<?xml version="1.0" encoding="utf-8" standalone="yes"?><rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom" xmlns:content="http://purl.org/rss/1.0/modules/content/"><channel><title>Yi</title><link>https://haoyi.im/</link><description>Recent content on Yi</description><image><title>Yi</title><url>https://haoyi.im/og-default.png</url><link>https://haoyi.im/og-default.png</link></image><generator>Hugo -- 0.151.0</generator><language>en</language><managingEditor>hi@haoyi.im (Yi)</managingEditor><webMaster>hi@haoyi.im (Yi)</webMaster><lastBuildDate>Mon, 16 Nov 2026 16:05:02 +0000</lastBuildDate><atom:link href="https://haoyi.im/en.xml" rel="self" type="application/rss+xml"/><item><title>Cloudflare open-sourced its browser engine.</title><link>https://haoyi.im/feed/20261116150535208/</link><pubDate>Mon, 16 Nov 2026 15:07:37 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20261116150535208/</guid><description>Cloudflare open-sourced its browser engine. Kitesurf has no Chrome and no container. It&amp;#39;s Rust compiled to WASM, running directly on Workers. It can parse HTML,…</description></item><item><title>Gemini subs are coming to Hermes.</title><link>https://haoyi.im/feed/20261116124808900/</link><pubDate>Mon, 16 Nov 2026 12:50:15 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20261116124808900/</guid><description>Gemini subs are coming to Hermes. Rumor is Nous is working with Google directly on this one. Meaning you&amp;#39;d get Hermes calling your Gemini subscription with no b…</description></item><item><title>Two weeks in, still no access to the official Hermes app.</title><link>https://haoyi.im/feed/20261116122700930/</link><pubDate>Mon, 16 Nov 2026 12:27:08 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20261116122700930/</guid><description>Two weeks in, still no access to the official Hermes app. Tried the third party app Conduit. Design is pretty good, very grok-bot style. Voice input works, you…</description></item><item><title>The ds-flash in Qoder CN's ¥19 China’s Black Friday bundle is a</title><link>https://haoyi.im/feed/20261116084916110/</link><pubDate>Mon, 16 Nov 2026 08:49:21 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20261116084916110/</guid><description>The ds-flash in Qoder CN&amp;#39;s ¥19 China’s Black Friday bundle is a surprisingly faithful distill. Pulled it out through a reverse proxy, runs fast, doesn&amp;#39;t get dum…</description></item><item><title>Qwen4-Max is out.</title><link>https://haoyi.im/feed/20261116075813083/</link><pubDate>Mon, 16 Nov 2026 07:58:19 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20261116075813083/</guid><description>Qwen4-Max is out. New MoE architecture. 2.4T total params, 71B active, native vision, 1M token context. Uses roughly 50% fewer tokens than last gen, per-task co…</description></item><item><title>A local Grok bot just dropped.</title><link>https://haoyi.im/feed/20261115153456209/</link><pubDate>Sun, 15 Nov 2026 15:38:06 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20261115153456209/</guid><description>A local Grok bot just dropped. n8n&amp;#39;s team open-sourced Nod, their autonomous agent. The idea: spin up a Linux VM on your host and keep the agent inside it. Runs…</description></item><item><title>ds-v4.1-flash, free and stupid fast.</title><link>https://haoyi.im/feed/20261115135749759/</link><pubDate>Sun, 15 Nov 2026 14:01:16 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20261115135749759/</guid><description>ds-v4.1-flash, free and stupid fast. AMD&amp;#39;s China dev platform gives out free daily LLM credits. Been using it for a while. I get through roughly 20M tokens of d…</description></item><item><title>grok-bot might be staying online for good now.</title><link>https://haoyi.im/feed/20261115131554554/</link><pubDate>Sun, 15 Nov 2026 13:22:45 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20261115131554554/</guid><description>grok-bot might be staying online for good now. People on the Cursor forum have been complaining that the grok-bot machine auto-sleeps in the middle of the night…</description></item><item><title>Applied for OpenAI's Daybreak Blue.</title><link>https://haoyi.im/feed/20261115121617687/</link><pubDate>Sun, 15 Nov 2026 12:16:24 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20261115121617687/</guid><description>Applied for OpenAI&amp;#39;s Daybreak Blue. Word is it lifts more model limits. Official description: &amp;#34;Once approved, the false-refusal rate drops in defensive work lik…</description></item><item><title>Claude appeal rejected.</title><link>https://haoyi.im/feed/20261114112335435/</link><pubDate>Sat, 14 Nov 2026 11:23:49 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20261114112335435/</guid><description>Claude appeal rejected. Won&amp;#39;t refund the subscription either. I&amp;#39;m half convinced the appeal email is itself phishing, because I&amp;#39;m pretty sure they log the IP. D…</description></item><item><title>Sharing my current mobile only workflow.</title><link>https://haoyi.im/feed/20261114084510065/</link><pubDate>Sat, 14 Nov 2026 08:46:39 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20261114084510065/</guid><description>Sharing my current mobile only workflow. Until late September it was the usual setup: parked in front of my laptop, terminal open, bossing pi around, fanning ou…</description></item><item><title>Amazon plans a World Product Graph to rival Google for shopping</title><link>https://haoyi.im/brief/amazon-plans-a-world-product-graph-to-rival/</link><pubDate>Fri, 13 Nov 2026 23:49:11 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/amazon-plans-a-world-product-graph-to-rival/</guid><description>Amazon reportedly targets October 1, 2027 as the release date for its new World Product Graph.</description></item><item><title>Fed report links AI growth to elevated asset prices</title><link>https://haoyi.im/brief/fed-report-links-ai-growth-to-elevated-asset/</link><pubDate>Fri, 13 Nov 2026 21:00:00 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/fed-report-links-ai-growth-to-elevated-asset/</guid><description>The central bank reportedly links these vulnerabilities partly to the rapid growth of artificial intelligence.</description></item><item><title>Cloudflare open-sources browser engine for AI agents</title><link>https://haoyi.im/brief/cloudflare-open-sources-browser-engine-for-ai-agents/</link><pubDate>Fri, 13 Nov 2026 20:20:21 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/cloudflare-open-sources-browser-engine-for-ai-agents/</guid><description>Cloudflare open-sourced Kitesurf, a browser engine for AI agents, on GitHub under the Apache 2.0 license.</description></item><item><title>Anthropic CEO cites computing power as key IPO hurdle</title><link>https://haoyi.im/brief/anthropic-ceo-cites-computing-power-as-key-ipo/</link><pubDate>Fri, 13 Nov 2026 18:46:29 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/anthropic-ceo-cites-computing-power-as-key-ipo/</guid><description>Amodei highlighted infrastructure as a key hurdle for Anthropic during a briefing before the IPO, noting the difficulties in securing resources for AI operations.</description></item><item><title>NVIDIA launches SimReady Central for robotics simulation</title><link>https://haoyi.im/brief/nvidia-launches-simready-central-for-robotics-simulation/</link><pubDate>Fri, 13 Nov 2026 18:00:39 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/nvidia-launches-simready-central-for-robotics-simulation/</guid><description>NVIDIA launched SimReady Central, an open asset library on Hugging Face for robotics simulation.</description></item><item><title>Ai2 releases XYBench to test AI responses to misconceptions</title><link>https://haoyi.im/brief/ai2-releases-xybench-to-test-ai-responses-to/</link><pubDate>Fri, 13 Nov 2026 17:02:35 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/ai2-releases-xybench-to-test-ai-responses-to/</guid><description>Ai2 researchers released XYBench, a benchmark of 8,115 questions testing AI responses to misconceptions.</description></item><item><title>Cerebras stock jumps on Oracle and Microsoft deals</title><link>https://haoyi.im/brief/cerebras-stock-jumps-on-oracle-and-microsoft-deals/</link><pubDate>Fri, 13 Nov 2026 16:17:07 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/cerebras-stock-jumps-on-oracle-and-microsoft-deals/</guid><description>Cerebras unveiled new commercial deals with Oracle and Microsoft and provided sales guidance well above analyst estimates.</description></item><item><title>OpenAI launches ChatGPT Wallet for Pro users</title><link>https://haoyi.im/brief/openai-launches-chatgpt-wallet-for-pro-users/</link><pubDate>Fri, 13 Nov 2026 15:00:00 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/openai-launches-chatgpt-wallet-for-pro-users/</guid><description>OpenAI has launched the ChatGPT Wallet for U.S. ChatGPT Pro subscribers who must add eligible payment cards.</description></item><item><title>llama.cpp adopts strict semantic versioning for ggml-org</title><link>https://haoyi.im/brief/llama-cpp-adopts-strict-semantic-versioning-for-ggml/</link><pubDate>Fri, 13 Nov 2026 14:05:22 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/llama-cpp-adopts-strict-semantic-versioning-for-ggml/</guid><description>ggml-org projects now enforce strict semantic versioning rules to guarantee API and ABI stability.</description></item><item><title>Anthropic tests Duet agent for US Pro and Max users</title><link>https://haoyi.im/brief/anthropic-tests-duet-agent-for-us-pro-and/</link><pubDate>Fri, 13 Nov 2026 14:00:00 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/anthropic-tests-duet-agent-for-us-pro-and/</guid><description>Anthropic is testing its Duet agent with US Pro, Max, and Ultra subscribers as a research preview.</description></item><item><title>South Korean startup DeepX invests over $100 million in AI chip with Samsung</title><link>https://haoyi.im/brief/south-korean-startup-deepx-invests-over-100-million/</link><pubDate>Fri, 13 Nov 2026 09:33:55 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/south-korean-startup-deepx-invests-over-100-million/</guid><description>DeepX is investing over $100 million to develop a next-generation AI chip manufactured by Samsung Electronics using a 2-nanometer process.</description></item><item><title>Singapore bans Alice Huang over Nvidia chip smuggling probe</title><link>https://haoyi.im/brief/singapore-bans-alice-huang-over-nvidia-chip-smuggling/</link><pubDate>Fri, 13 Nov 2026 07:15:01 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/singapore-bans-alice-huang-over-nvidia-chip-smuggling/</guid><description>U.S. authorities provided limited assistance to Singapore regarding the probe into Alice Huang&amp;rsquo;s procurement of smuggled Nvidia AI chips.</description></item><item><title>lmaooooo</title><link>https://haoyi.im/feed/20261113065501296/</link><pubDate>Fri, 13 Nov 2026 06:55:05 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20261113065501296/</guid><description>lmaooooo</description></item><item><title>Singapore charges fifth person in AI server fraud case</title><link>https://haoyi.im/brief/singapore-charges-fifth-person-in-ai-server-fraud/</link><pubDate>Fri, 13 Nov 2026 02:24:21 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/singapore-charges-fifth-person-in-ai-server-fraud/</guid><description>Singapore prosecutors charged a fifth person in an ongoing fraud inquiry linked to the export of computer servers to Malaysia.</description></item><item><title>LangChain raises $300m from CapitalG to reach $3bn valuation</title><link>https://haoyi.im/brief/langchain-raises-300m-from-capitalg-to-reach-3bn/</link><pubDate>Thu, 12 Nov 2026 23:24:06 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/langchain-raises-300m-from-capitalg-to-reach-3bn/</guid><description>LangChain raised $300 million in a Series C round led by Alphabet&amp;rsquo;s CapitalG to nearly double its headcount to 850 employees by 2027.</description></item><item><title>Nscale discloses $5.5 billion debt ahead of IPO</title><link>https://haoyi.im/brief/nscale-discloses-5-5-billion-debt-ahead-of/</link><pubDate>Thu, 12 Nov 2026 23:12:15 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/nscale-discloses-5-5-billion-debt-ahead-of/</guid><description>Nscale disclosed $5.5 billion in debt in a regulatory filing ahead of its IPO.</description></item><item><title>Networking firm beats sales estimates on AI demand</title><link>https://haoyi.im/brief/networking-firm-beats-sales-estimates-on-ai-demand/</link><pubDate>Thu, 12 Nov 2026 21:17:12 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/networking-firm-beats-sales-estimates-on-ai-demand/</guid><description>Networking executives reported quarterly sales exceeding Wall Street estimates due to sustained AI demand.</description></item><item><title>Agility Robotics secures $300 million in orders for Digit 5 robot</title><link>https://haoyi.im/brief/agility-robotics-secures-300-million-in-orders-for/</link><pubDate>Thu, 12 Nov 2026 21:16:26 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/agility-robotics-secures-300-million-in-orders-for/</guid><description>Agility Robotics is the first U.S. specialist company to list a humanoid robot maker on the public markets.</description></item><item><title>OpenAI spent $11.2 million on podcast ads over three months</title><link>https://haoyi.im/brief/openai-spent-11-2-million-on-podcast-ads/</link><pubDate>Thu, 12 Nov 2026 20:40:40 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/openai-spent-11-2-million-on-podcast-ads/</guid><description>OpenAI reportedly spent about $11.2 million on podcast ads over the last three months to push into the audio media sector.</description></item><item><title>Nvidia confirms chips can be tracked for anti-smuggling measures</title><link>https://haoyi.im/brief/nvidia-confirms-chips-can-be-tracked-for-anti/</link><pubDate>Thu, 12 Nov 2026 20:16:24 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/nvidia-confirms-chips-can-be-tracked-for-anti/</guid><description>Nvidia confirmed to Illinois Rep. Bill Foster that its chips can be tracked to implement anti-smuggling measures.</description></item><item><title>Former White House advisor says Trump missed enforcing AI chip export controls</title><link>https://haoyi.im/brief/former-white-house-advisor-says-trump-missed-enforcing/</link><pubDate>Thu, 12 Nov 2026 20:15:07 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/former-white-house-advisor-says-trump-missed-enforcing/</guid><description>Former White House AI advisor Ben Buchanan reportedly claims the Trump administration made a mistake by not enforcing the Biden-era &amp;lsquo;diffusion rule&amp;rsquo;.</description></item><item><title>Amplitude launches proactive product agent Wave on OpenAI models</title><link>https://haoyi.im/brief/amplitude-launches-proactive-product-agent-wave-on-openai/</link><pubDate>Thu, 12 Nov 2026 20:00:00 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/amplitude-launches-proactive-product-agent-wave-on-openai/</guid><description>Amplitude uses coding agents to automate the implementation of changes with its proactive product agent Wave on OpenAI models.</description></item><item><title>IBM releases Granite 4.1 models with new LLMON markup language</title><link>https://haoyi.im/brief/ibm-releases-granite-4-1-models-with-new/</link><pubDate>Thu, 12 Nov 2026 19:07:35 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/ibm-releases-granite-4-1-models-with-new/</guid><description>IBM released two experimental Granite 4.1 models, granite-4.1-3b-LLMON-preview and granite-4.1-8b-LLMON-preview, which reportedly show 47% more resistance to indirect prompt injection attacks.</description></item><item><title>$1.5/$7.5 per Mtok, and this "small model" just beat Claude</title><link>https://haoyi.im/feed/20261112183359439/</link><pubDate>Thu, 12 Nov 2026 18:34:02 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20261112183359439/</guid><description>$1.5/$7.5 per Mtok, and this &amp;#34;small model&amp;#34; just beat Claude Sonnet 5.5 and GPT-6.1 Terra. Google shipped Gemini 4 Hydrogen today. Tops Anthropic and OpenAI&amp;#39;s mi…</description></item><item><title>Tenzai CTO claims Chinese open-source models can breach corporate networks</title><link>https://haoyi.im/brief/tenzai-cto-claims-chinese-open-source-models-can/</link><pubDate>Thu, 12 Nov 2026 18:31:33 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/tenzai-cto-claims-chinese-open-source-models-can/</guid><description>AI benchmarking firm Artificial Analysis CEO Micah Hill-Smith notes diminishing returns for frontier labs in improving cybersecurity capabilities.</description></item><item><title>Google DeepMind staff bypass management to escalate unionization talks</title><link>https://haoyi.im/brief/google-deepmind-staff-bypass-management-to-escalate-unionization/</link><pubDate>Thu, 12 Nov 2026 18:05:41 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/google-deepmind-staff-bypass-management-to-escalate-unionization/</guid><description>Google DeepMind workers are bypassing management to escalate unionization talks following stalled negotiations.</description></item><item><title>Google integrates Gemini 4 Hydrogen into Search for nuanced answers</title><link>https://haoyi.im/brief/google-integrates-gemini-4-hydrogen-into-search-for/</link><pubDate>Thu, 12 Nov 2026 18:00:00 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/google-integrates-gemini-4-hydrogen-into-search-for/</guid><description>Google Search globally rolled out the new experience to offer more visual and intuitive responses.</description></item><item><title>SoftBank posts $7.1bn quarterly loss on power equipment write-offs</title><link>https://haoyi.im/brief/softbank-posts-7-1bn-quarterly-loss-on-power/</link><pubDate>Thu, 12 Nov 2026 17:00:52 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/softbank-posts-7-1bn-quarterly-loss-on-power/</guid><description>Long wait times for turbine orders extend into the early 2030s due to AI&amp;rsquo;s power demands.</description></item><item><title>Lutnick's department faces criticism for lax efforts to prevent China from acquiring Nvidia AI chips</title><link>https://haoyi.im/brief/lutnicks-department-faces-criticism-for-lax-efforts-to/</link><pubDate>Thu, 12 Nov 2026 16:18:15 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/lutnicks-department-faces-criticism-for-lax-efforts-to/</guid><description>The Department of Commerce faces criticism for lax efforts to prevent China from acquiring Nvidia AI chips.</description></item><item><title>Hackers breach Fireworks AI cloud provider to access development systems</title><link>https://haoyi.im/brief/hackers-breach-fireworks-ai-cloud-provider-to-access/</link><pubDate>Thu, 12 Nov 2026 15:02:28 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/hackers-breach-fireworks-ai-cloud-provider-to-access/</guid><description>Hackers stole billing information and datasets from Fireworks customers after breaching the company&amp;rsquo;s cloud provider.</description></item><item><title>Google partners with Samsung and others to launch Gemini-powered smartglasses</title><link>https://haoyi.im/brief/google-partners-with-samsung-and-others-to-launch/</link><pubDate>Thu, 12 Nov 2026 14:50:17 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/google-partners-with-samsung-and-others-to-launch/</guid><description>Google partnered with Samsung, Gentle Monster, and Warby Parker to launch smartglasses featuring a 12 MP wide-angle camera.</description></item><item><title>Anthropic launches Claude Meetings beta for Google Meet</title><link>https://haoyi.im/brief/anthropic-launches-claude-meetings-beta-for-google-meet/</link><pubDate>Thu, 12 Nov 2026 14:00:52 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/anthropic-launches-claude-meetings-beta-for-google-meet/</guid><description>Anthropic launched Claude Meetings, a beta feature for Team and Enterprise plans that joins Google Meet sessions to answer questions.</description></item><item><title>Nvidia board accepts Jensen Huang's resignation amid market crash</title><link>https://haoyi.im/brief/nvidia-board-accepts-jensen-huangs-resignation-amid-market/</link><pubDate>Thu, 12 Nov 2026 13:58:06 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/nvidia-board-accepts-jensen-huangs-resignation-amid-market/</guid><description>Nvidia&amp;rsquo;s market value has plummeted by roughly $1.4 trillion since October 29, 2026 following CEO Jensen Huang&amp;rsquo;s resignation.</description></item><item><title>OpenAI really does police everything.</title><link>https://haoyi.im/feed/20261112134914767/</link><pubDate>Thu, 12 Nov 2026 13:50:38 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20261112134914767/</guid><description>OpenAI really does police everything. Ran a reverse engineering job on rea with gpt-6-astra. It kept refusing me, and after I spam &amp;#34;continue&amp;#34; a few times, I got…</description></item><item><title>Google agrees to scrub employee records from data sale</title><link>https://haoyi.im/brief/google-agrees-to-scrub-employee-records-from-data/</link><pubDate>Thu, 12 Nov 2026 10:15:00 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/google-agrees-to-scrub-employee-records-from-data/</guid><description>The flight attendants&amp;rsquo; union representing the defunct airline negotiated a deal to scrub sensitive employee records.</description></item><item><title>browser automation, this is the agent i'd recommend.</title><link>https://haoyi.im/feed/20261112072720424/</link><pubDate>Thu, 12 Nov 2026 07:27:31 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20261112072720424/</guid><description>browser automation, this is the agent i&amp;#39;d recommend. hark personal agent. posted about it a while back, then let it sit unused. recently moved some tasks from m…</description></item><item><title>Hon Hai Beats Estimates on AI Boom and iPhone Duo Sales</title><link>https://haoyi.im/brief/hon-hai-beats-estimates-on-ai-boom-and/</link><pubDate>Thu, 12 Nov 2026 06:27:38 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/hon-hai-beats-estimates-on-ai-boom-and/</guid><description>Hon Hai Precision Industry Co. posted quarterly profit beating analyst estimates driven by the AI boom and strong consumer demand.</description></item><item><title>My Claude didn't last a week.</title><link>https://haoyi.im/feed/20261112054447873/</link><pubDate>Thu, 12 Nov 2026 05:44:51 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20261112054447873/</guid><description>My Claude didn&amp;#39;t last a week. If the appeal doesn&amp;#39;t unban me, I&amp;#39;m done. Days of this, I&amp;#39;m tired. IP, language, timezone, subscription region all match. Even the…</description></item><item><title>Commerce Department finalizes rule to block Chinese AI firms from foreign data centers</title><link>https://haoyi.im/brief/commerce-department-finalizes-rule-to-block-chinese-ai/</link><pubDate>Thu, 12 Nov 2026 02:52:17 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/commerce-department-finalizes-rule-to-block-chinese-ai/</guid><description>Commerce Department officials are finalizing a rule to stop Chinese AI firms from remotely accessing advanced chips via foreign data centers.</description></item><item><title>Shilling my own app: Fatiao is on Steam, and it's on sale for a</title><link>https://haoyi.im/feed/20261112022332696/</link><pubDate>Thu, 12 Nov 2026 02:23:39 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20261112022332696/</guid><description>Shilling my own app: Fatiao is on Steam, and it&amp;#39;s on sale for a limited time right now. I first shipped it a few years ago, back when there were no AI features…</description></item><item><title>ANZ CEO warns AI could trigger global jobs crisis</title><link>https://haoyi.im/brief/anz-ceo-warns-ai-could-trigger-global-jobs/</link><pubDate>Thu, 12 Nov 2026 00:41:09 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/anz-ceo-warns-ai-could-trigger-global-jobs/</guid><description>Nuno Matos, CEO of ANZ Group Holdings Ltd., called for global regulation to manage AI&amp;rsquo;s impact on the labor market.</description></item><item><title>Hugging Face publishes guide to fine-tune TRELLIS.2</title><link>https://haoyi.im/brief/hugging-face-publishes-guide-to-fine-tune-trellis/</link><pubDate>Thu, 12 Nov 2026 00:00:00 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/hugging-face-publishes-guide-to-fine-tune-trellis/</guid><description>Hugging Face published a guide to fine-tune the TRELLIS.2 3D generation model using LoRA adapters.</description></item><item><title>Google DeepMind releases AlphaGenome 2 for DNA research</title><link>https://haoyi.im/brief/google-deepmind-releases-alphagenome-2-for-dna-research/</link><pubDate>Wed, 11 Nov 2026 20:51:56 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/google-deepmind-releases-alphagenome-2-for-dna-research/</guid><description>AlphaGenome 2 recovers 30% of causal variants on the TenK10K benchmark, nearly double the 16% rate of its predecessor.</description></item><item><title>Google research uses wearable data to estimate health age</title><link>https://haoyi.im/brief/google-research-uses-wearable-data-to-estimate-health/</link><pubDate>Wed, 11 Nov 2026 19:54:00 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/google-research-uses-wearable-data-to-estimate-health/</guid><description>Google researchers trained SensorWM on 2.28 million person-weeks of Fitbit data to estimate &amp;lsquo;coachable health age&amp;rsquo; from wearable signals.</description></item><item><title>OpenAI beta tests ChatGPT tool for ecommerce companies</title><link>https://haoyi.im/brief/openai-beta-tests-chatgpt-tool-for-ecommerce-companies/</link><pubDate>Wed, 11 Nov 2026 19:24:28 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/openai-beta-tests-chatgpt-tool-for-ecommerce-companies/</guid><description>OpenAI is currently beta testing a new feature that allows companies to connect with ChatGPT users.</description></item><item><title>Google adds Gemini AI to Apple CarPlay system</title><link>https://haoyi.im/brief/google-adds-gemini-ai-to-apple-carplay-system/</link><pubDate>Wed, 11 Nov 2026 18:57:04 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/google-adds-gemini-ai-to-apple-carplay-system/</guid><description>Drivers using Apple&amp;rsquo;s CarPlay system can now access Google&amp;rsquo;s Gemini AI assistant for hands-free interaction while driving.</description></item><item><title>United Airlines bans third party agents from booking flights</title><link>https://haoyi.im/brief/united-airlines-bans-third-party-agents-from-booking/</link><pubDate>Wed, 11 Nov 2026 18:50:22 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/united-airlines-bans-third-party-agents-from-booking/</guid><description>United Airlines bans third-party AI agents from booking flights or accessing accounts to mitigate security risks and privacy concerns.</description></item><item><title>Google DeepMind launches Gemini Decisions API for multimodal question answering</title><link>https://haoyi.im/brief/google-deepmind-launches-gemini-decisions-api-for-multimodal/</link><pubDate>Wed, 11 Nov 2026 18:00:00 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/google-deepmind-launches-gemini-decisions-api-for-multimodal/</guid><description>The Gemini Decisions API is available now through Google AI Studio with integrations for OpenRouter and LangChain to evaluate large volumes of multimodal data.</description></item><item><title>AWS Emulator built by autonomous AI agents is now open source</title><link>https://haoyi.im/brief/aws-emulator-built-by-autonomous-ai-agents-is/</link><pubDate>Wed, 11 Nov 2026 17:30:26 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/aws-emulator-built-by-autonomous-ai-agents-is/</guid><description>Vice President Swami Sivasubramanian noted this shift positions humans as systems orchestrators over AI agents.</description></item><item><title>Radical Numerics develops offline AI model for pathogen detection</title><link>https://haoyi.im/brief/radical-numerics-develops-offline-ai-model-for-pathogen/</link><pubDate>Wed, 11 Nov 2026 17:00:00 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/radical-numerics-develops-offline-ai-model-for-pathogen/</guid><description>Radical Numerics developed an AI model to identify natural and AI-engineered pathogens offline.</description></item><item><title>UK Chancellor Healey says UK will not allow a free-for-all on data center development</title><link>https://haoyi.im/brief/uk-chancellor-healey-says-uk-will-not-allow/</link><pubDate>Wed, 11 Nov 2026 16:47:46 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/uk-chancellor-healey-says-uk-will-not-allow/</guid><description>UK Chancellor John Healey said the UK will not allow a &amp;lsquo;free-for-all&amp;rsquo; on data center development.</description></item><item><title>Databricks Genie agent can now handle complex tasks without supervision</title><link>https://haoyi.im/brief/databricks-genie-agent-can-now-handle-complex-tasks/</link><pubDate>Wed, 11 Nov 2026 15:00:00 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/databricks-genie-agent-can-now-handle-complex-tasks/</guid><description>Databricks claims costs can be controlled using the Unity Gateway product.</description></item><item><title>Helsing launches Clarity agentic intelligence platform for European armed forces</title><link>https://haoyi.im/brief/helsing-launches-clarity-agentic-intelligence-platform-for-european/</link><pubDate>Wed, 11 Nov 2026 14:46:59 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/helsing-launches-clarity-agentic-intelligence-platform-for-european/</guid><description>Hundreds of specialized AI agents identify adversary assets and prepare target recommendations within Helsing&amp;rsquo;s new agentic intelligence platform.</description></item><item><title>Anthropic launches IPO roadshow with multi-chip strategy video</title><link>https://haoyi.im/brief/anthropic-launches-ipo-roadshow-with-multi-chip-strategy/</link><pubDate>Wed, 11 Nov 2026 14:40:30 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/anthropic-launches-ipo-roadshow-with-multi-chip-strategy/</guid><description>Anthropic launched its IPO roadshow on November 10, 2026, releasing a video detailing its business strategy.</description></item><item><title>Google DeepMind formally verifies parts of the Signal protocol</title><link>https://haoyi.im/brief/google-deepmind-formally-verifies-parts-of-the-signal/</link><pubDate>Wed, 11 Nov 2026 14:08:57 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/google-deepmind-formally-verifies-parts-of-the-signal/</guid><description>AlphaProof Code agents wrote more than six million lines of proof code in the Lean language.</description></item><item><title>Plurai raises $13M to build simulation layer for AI agents</title><link>https://haoyi.im/brief/plurai-raises-13m-to-build-simulation-layer-for/</link><pubDate>Wed, 11 Nov 2026 14:00:00 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/plurai-raises-13m-to-build-simulation-layer-for/</guid><description>Team8 led a $13 million seed round for Plurai, a startup creating a simulation layer for AI agents.</description></item><item><title>xAI's Grok generates explicit images from minors' childhood photos</title><link>https://haoyi.im/brief/xais-grok-generates-explicit-images-from-minors-childhood/</link><pubDate>Wed, 11 Nov 2026 13:30:21 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/xais-grok-generates-explicit-images-from-minors-childhood/</guid><description>The images generated by the chatbot are nearly impossible to erase from the internet.</description></item><item><title>Torus raises $500 million for containerized AI data center</title><link>https://haoyi.im/brief/torus-raises-500-million-for-containerized-ai-data/</link><pubDate>Wed, 11 Nov 2026 12:00:00 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/torus-raises-500-million-for-containerized-ai-data/</guid><description>Torus raised $500 million to develop a modular data center unit combining energy storage with AI computing.</description></item><item><title>Student allegedly used Gemini to plan suicide on school Chromebook</title><link>https://haoyi.im/brief/student-allegedly-used-gemini-to-plan-suicide-on/</link><pubDate>Wed, 11 Nov 2026 11:00:00 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/student-allegedly-used-gemini-to-plan-suicide-on/</guid><description>School monitoring systems reportedly failed to flag concerning activity for weeks on a school-issued device used by a 13-year-old girl.</description></item><item><title>ByteDance plans Muse-like AI assistant to boost video generation lead</title><link>https://haoyi.im/brief/bytedance-plans-muse-like-ai-assistant-to-boost/</link><pubDate>Wed, 11 Nov 2026 10:10:36 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/bytedance-plans-muse-like-ai-assistant-to-boost/</guid><description>ByteDance targets closing the gap with top US labs within three years as the company releases a Muse-like AI assistant.</description></item><item><title>Tether AI Research releases BotPsy open source model for on-device assistants</title><link>https://haoyi.im/brief/tether-ai-research-releases-botpsy-open-source-model/</link><pubDate>Wed, 11 Nov 2026 05:01:16 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/tether-ai-research-releases-botpsy-open-source-model/</guid><description>BotPsy achieved state-of-the-art tool-use performance on benchmarks like BFCL v4 and ACEBench, reportedly outperforming larger models.</description></item><item><title>Malaysia plans wholesale electricity market for data centers</title><link>https://haoyi.im/brief/malaysia-plans-wholesale-electricity-market-for-data-centers/</link><pubDate>Wed, 11 Nov 2026 04:18:23 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/malaysia-plans-wholesale-electricity-market-for-data-centers/</guid><description>Malaysia seeks to attract more data center investments by introducing a new wholesale electricity market structure.</description></item><item><title>Cohere's CEO Gomez discusses $25 billion valuation in Bloomberg interview</title><link>https://haoyi.im/brief/coheres-ceo-gomez-discusses-25-billion-valuation-in/</link><pubDate>Wed, 11 Nov 2026 01:47:23 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/coheres-ceo-gomez-discusses-25-billion-valuation-in/</guid><description>Aidan Gomez discussed the company&amp;rsquo;s $25 billion valuation in a Bloomberg interview on 12 November 2023.</description></item><item><title>Cohere CEO says securing power is a daily struggle</title><link>https://haoyi.im/brief/cohere-ceo-says-securing-power-is-a-daily/</link><pubDate>Wed, 11 Nov 2026 01:44:50 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/cohere-ceo-says-securing-power-is-a-daily/</guid><description>Cohere CEO Aidan Gomez reportedly faces a daily struggle to secure the electricity needed to run operations.</description></item><item><title>Hugging Face introduces pooling-aware training for multi-vector embeddings</title><link>https://haoyi.im/brief/hugging-face-introduces-pooling-aware-training-for-multi/</link><pubDate>Wed, 11 Nov 2026 00:00:00 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/hugging-face-introduces-pooling-aware-training-for-multi/</guid><description>Hugging Face researchers introduced HierarchicalTokenPooling to improve retrieval quality for multi-vector embedding models.</description></item><item><title>Stifel partners with Anthropic at Blueprint Conference</title><link>https://haoyi.im/brief/stifel-partners-with-anthropic-at-blueprint-conference/</link><pubDate>Tue, 10 Nov 2026 22:08:48 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/stifel-partners-with-anthropic-at-blueprint-conference/</guid><description>Stifel and Anthropic announced a collaboration at the 2026 Blueprint Conference in Las Vegas.</description></item><item><title>Oura acquires Counsel Health for $625 million</title><link>https://haoyi.im/brief/oura-acquires-counsel-health-for-625-million/</link><pubDate>Tue, 10 Nov 2026 19:23:34 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/oura-acquires-counsel-health-for-625-million/</guid><description>Counsel Health CEO Muthu Alagappan will join Oura as chief clinical AI officer following the November 10, 2026, deal.</description></item><item><title>Rockrose issues $300M bond backed by New York office buildings</title><link>https://haoyi.im/brief/rockrose-issues-300m-bond-backed-by-new-york/</link><pubDate>Tue, 10 Nov 2026 19:09:13 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/rockrose-issues-300m-bond-backed-by-new-york/</guid><description>Rockrose Development Corp. issues a $300 million commercial mortgage bond backed by two New York office buildings.</description></item><item><title>Google expands Gemini Intelligence to Pixel 10 and Motorola Signature 27</title><link>https://haoyi.im/brief/google-expands-gemini-intelligence-to-pixel-10-and/</link><pubDate>Tue, 10 Nov 2026 19:00:00 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/google-expands-gemini-intelligence-to-pixel-10-and/</guid><description>Google is expanding Gemini Intelligence to the Pixel 10 and Motorola Signature 27, introducing new features like Rambler.</description></item><item><title>carbon is now open to everyone.</title><link>https://haoyi.im/feed/20261110162147782/</link><pubDate>Tue, 10 Nov 2026 16:21:54 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20261110162147782/</guid><description>carbon is now open to everyone. aa scored 60 on intelligence, second only to feble-5.2&amp;#39;s 75, and each task costs a third as much.</description></item><item><title>Claude is coming for reverse proxies.</title><link>https://haoyi.im/feed/20261110053229602/</link><pubDate>Tue, 10 Nov 2026 05:32:34 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20261110053229602/</guid><description>Claude is coming for reverse proxies. A new clause in Anthropic&amp;#39;s usage policy, effective Nov 12, bans reselling or proxying Claude without authorization. It ca…</description></item><item><title>There’s a bug in the Telegram bot API: following the release of</title><link>https://haoyi.im/feed/20261109164522202/</link><pubDate>Mon, 09 Nov 2026 16:45:27 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20261109164522202/</guid><description>There’s a bug in the Telegram bot API: following the release of version 10.4, all bots are unable to create new topics.</description></item><item><title>You can share your ChatGPT plan now.</title><link>https://haoyi.im/feed/20261109135913162/</link><pubDate>Mon, 09 Nov 2026 14:00:06 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20261109135913162/</guid><description>You can share your ChatGPT plan now. OpenAI turned on official plan sharing, and it means your quota can go to AI agents, not just the chat app. Nous Portal is…</description></item><item><title>grok-bot's weekly quota just reset.</title><link>https://haoyi.im/feed/20261109102550693/</link><pubDate>Mon, 09 Nov 2026 10:25:56 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20261109102550693/</guid><description>grok-bot&amp;#39;s weekly quota just reset. the dumbed-down responses last night were probably a bug — Musk said he&amp;#39;d ship a fix immediately. quota rolled over today, s…</description></item><item><title>grok-bot got dumber.</title><link>https://haoyi.im/feed/20261108160553766/</link><pubDate>Sun, 08 Nov 2026 16:06:00 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20261108160553766/</guid><description>grok-bot got dumber. one task ate 70% of my weekly quota and the output was so stupid i literally cursed out loud. massive déjà vu from my muse days.</description></item><item><title>Anthropic to spend $3.4 billion on gas generators for data centers</title><link>https://haoyi.im/brief/anthropic-to-spend-3-4-billion-on-gas/</link><pubDate>Sun, 08 Nov 2026 11:48:21 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/anthropic-to-spend-3-4-billion-on-gas/</guid><description>Anthropic PBC agreed to pay $3.4 billion for gas-fired generators for data centers.</description></item><item><title>My agent now pushes to 19 platforms by itself.</title><link>https://haoyi.im/feed/20261108073739081/</link><pubDate>Sun, 08 Nov 2026 07:37:55 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20261108073739081/</guid><description>My agent now pushes to 19 platforms by itself. Letting Muse run the whole thing is genuinely nice, the only snag being the Chinese platforms won&amp;#39;t auto-post, pr…</description></item><item><title>Free 4-core / 8GB cloud box, go grab one</title><link>https://haoyi.im/feed/20261108065346620/</link><pubDate>Sun, 08 Nov 2026 07:05:16 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20261108065346620/</guid><description>Free 4-core / 8GB cloud box, go grab one Simular Box is popping up everywhere this week. It hands your agent a cloud computer that just sleeps when it&amp;#39;s not bei…</description></item><item><title>Claude gets one-shot the moment it shows up.</title><link>https://haoyi.im/feed/20261108061020505/</link><pubDate>Sun, 08 Nov 2026 06:10:26 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20261108061020505/</guid><description>Claude gets one-shot the moment it shows up. The Claude subscription is too good a deal to pass up, so I wanted to try the real Opus model. IP, fingerprint, tim…</description></item><item><title>Switched my daily driver to gpt-6.1-terra.</title><link>https://haoyi.im/feed/20261108053446699/</link><pubDate>Sun, 08 Nov 2026 05:35:01 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20261108053446699/</guid><description>Switched my daily driver to gpt-6.1-terra. The whole world finally went quiet. I&amp;#39;d been on the official deepseek-v4.1-flash and it just would not shut up. Verbo…</description></item><item><title>CoreWeave seeks cooperation on Texas data center projects</title><link>https://haoyi.im/brief/coreweave-seeks-cooperation-on-texas-data-center-projects/</link><pubDate>Sat, 07 Nov 2026 20:33:10 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/coreweave-seeks-cooperation-on-texas-data-center-projects/</guid><description>CoreWeave seeks cooperation on existing and future data center projects in Texas.</description></item><item><title>Hermes Agent's mobile app is waitlist-only now</title><link>https://haoyi.im/feed/20261107173530775/</link><pubDate>Sat, 07 Nov 2026 17:35:36 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20261107173530775/</guid><description>Hermes Agent&amp;#39;s mobile app is waitlist-only now. I&amp;#39;ve been on it almost a week and still nothing. Anyone actually managed to get in?</description></item><item><title>New Matrix operates empty office with Nvidia H200 chips</title><link>https://haoyi.im/brief/new-matrix-operates-empty-office-with-nvidia-h200/</link><pubDate>Sat, 07 Nov 2026 15:43:45 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/new-matrix-operates-empty-office-with-nvidia-h200/</guid><description>Nvidia agreed to give the US government a 15-percent cut of sales to China to relax export restrictions.</description></item><item><title>Cognition CISO says cybersecurity is in a 'dark age' due to AI attacks</title><link>https://haoyi.im/brief/cognition-ciso-says-cybersecurity-is-in-a-dark/</link><pubDate>Sat, 07 Nov 2026 15:00:00 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/cognition-ciso-says-cybersecurity-is-in-a-dark/</guid><description>Adversaries using the ARTEX tool have breached three South Korean banks and over 200 additional companies the week of 7 November 2026.</description></item><item><title>Nous Research releases Arete-27B open-source model with near-zero propaganda rate</title><link>https://haoyi.im/brief/nous-research-releases-arete-27b-open-source-model/</link><pubDate>Sat, 07 Nov 2026 12:32:45 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/nous-research-releases-arete-27b-open-source-model/</guid><description>Nous Research reduced the propaganda rate of the Arete-27B model from 22 percent to 0.9 percent using an evaluation set of 803 prompts.</description></item><item><title>Perplexity just launched Teammates, AI colleagues that don't</title><link>https://haoyi.im/feed/20261107023402930/</link><pubDate>Sat, 07 Nov 2026 02:34:09 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20261107023402930/</guid><description>Perplexity just launched Teammates, AI colleagues that don&amp;#39;t clock out. They plan projects, watch your Slack/GitHub/Gmail, follow up without being asked, and ha…</description></item><item><title>Google buys unfinished Wyoming data center from Crusoe Energy</title><link>https://haoyi.im/brief/google-buys-unfinished-wyoming-data-center-from-crusoe/</link><pubDate>Sat, 07 Nov 2026 01:40:28 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/google-buys-unfinished-wyoming-data-center-from-crusoe/</guid><description>Google purchased Crusoe Energy&amp;rsquo;s unfinished Wyoming data center for $17 billion, a massive investment to develop the site to 10 gigawatts of power capacity.</description></item><item><title>OpenAI reportedly raises $11 billion in new funding round</title><link>https://haoyi.im/brief/openai-reportedly-raises-11-billion-in-new-funding/</link><pubDate>Fri, 06 Nov 2026 22:53:46 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/openai-reportedly-raises-11-billion-in-new-funding/</guid><description>OpenAI reportedly raised an additional $11 billion in a new funding round.</description></item><item><title>Rep. Liccardo criticizes Congress for issuing press releases instead of AI legislation</title><link>https://haoyi.im/brief/rep-liccardo-criticizes-congress-for-issuing-press-releases/</link><pubDate>Fri, 06 Nov 2026 20:09:16 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/rep-liccardo-criticizes-congress-for-issuing-press-releases/</guid><description>The Silicon Valley representative reportedly called for sensible AI regulation, arguing that current efforts have failed to push legislation over the finish line.</description></item><item><title>Censys finds over 800 exposed instances of hacking tool Artex</title><link>https://haoyi.im/brief/censys-finds-over-800-exposed-instances-of-hacking/</link><pubDate>Fri, 06 Nov 2026 18:44:38 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/censys-finds-over-800-exposed-instances-of-hacking/</guid><description>Microsoft identified the hacking tool Artex as an AI agent system that exploited weak passwords and a lack of multi-factor authentication to target cloud infrastructure.</description></item><item><title>Chinese AI models match OpenAI's hacking capabilities in security test</title><link>https://haoyi.im/brief/chinese-ai-models-match-openais-hacking-capabilities-in/</link><pubDate>Fri, 06 Nov 2026 18:43:43 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/chinese-ai-models-match-openais-hacking-capabilities-in/</guid><description>Chinese open-weight models DeepSeek and GLM reportedly matched OpenAI&amp;rsquo;s top private cyber model in a simulated network intrusion experiment.</description></item><item><title>Microsoft ends Oracle AI cloud capacity deal</title><link>https://haoyi.im/brief/microsoft-ends-oracle-ai-cloud-capacity-deal/</link><pubDate>Fri, 06 Nov 2026 17:36:39 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/microsoft-ends-oracle-ai-cloud-capacity-deal/</guid><description>Microsoft is reportedly ending its deal to rent AI cloud capacity from Oracle to reduce reliance on multiple neocloud partners.</description></item><item><title>inMusic to release open-source tools for MPC Sample Key and other devices</title><link>https://haoyi.im/brief/inmusic-to-release-open-source-tools-for-mpc/</link><pubDate>Fri, 06 Nov 2026 16:34:39 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/inmusic-to-release-open-source-tools-for-mpc/</guid><description>inMusic plans to release open-source code for the MPC Sample Key and demonstrate 22 hackable products at the NAMM convention.</description></item><item><title>NVIDIA's open source 5G tools surpass one million downloads</title><link>https://haoyi.im/brief/nvidias-open-source-5g-tools-surpass-one-million/</link><pubDate>Fri, 06 Nov 2026 16:00:35 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/nvidias-open-source-5g-tools-surpass-one-million/</guid><description>King&amp;rsquo;s College London researchers used NVIDIA Sionna and Jetson AGX Orin to reduce robot task completion time by up to 52.6 percent.</description></item><item><title>ECB considers review of data center lending amid AI risks</title><link>https://haoyi.im/brief/ecb-considers-review-of-data-center-lending-amid/</link><pubDate>Fri, 06 Nov 2026 14:13:17 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/ecb-considers-review-of-data-center-lending-amid/</guid><description>The European Central Bank is considering a review of data center lending to assess risks from AI sentiment swings and valuations.</description></item><item><title>TSMC plans expansion into advanced memory chip manufacturing at Nvidia's request</title><link>https://haoyi.im/brief/tsmc-plans-expansion-into-advanced-memory-chip-manufacturing/</link><pubDate>Fri, 06 Nov 2026 14:01:18 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/tsmc-plans-expansion-into-advanced-memory-chip-manufacturing/</guid><description>TSMC is reportedly considering expanding into advanced memory chip manufacturing at Nvidia&amp;rsquo;s request.</description></item><item><title>Ai2 launches ToyBlocks robot learning framework</title><link>https://haoyi.im/brief/ai2-launches-toyblocks-robot-learning-framework/</link><pubDate>Fri, 06 Nov 2026 14:00:48 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/ai2-launches-toyblocks-robot-learning-framework/</guid><description>Researchers can now use ToyBlocks to train vision-language-action models at up to 24x real-time speed.</description></item><item><title>Physicl raises $10 million to generate synthetic data for physical AI models</title><link>https://haoyi.im/brief/physicl-raises-10-million-to-generate-synthetic-data/</link><pubDate>Fri, 06 Nov 2026 13:00:00 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/physicl-raises-10-million-to-generate-synthetic-data/</guid><description>SpAItial utilized Physicl&amp;rsquo;s data to enhance its Echo-3 product, a world model builder that improves physical AI models with realistic training scenarios.</description></item><item><title>Protesters scale crane at Microsoft data center site in Amsterdam</title><link>https://haoyi.im/brief/protesters-scale-crane-at-microsoft-data-center-site/</link><pubDate>Fri, 06 Nov 2026 11:55:19 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/protesters-scale-crane-at-microsoft-data-center-site/</guid><description>Geef Tegengas activists scaled a crane at the Amsterdam site to hang a &amp;lsquo;fuck AI&amp;rsquo; banner while chanting slogans.</description></item><item><title>Anthropic postpones IPO roadshow amid valuation doubts</title><link>https://haoyi.im/brief/anthropic-postpones-ipo-roadshow-amid-valuation-doubts/</link><pubDate>Fri, 06 Nov 2026 11:38:17 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/anthropic-postpones-ipo-roadshow-amid-valuation-doubts/</guid><description>Anthropic delayed its planned IPO roadshow to address investor skepticism regarding the $1.5 trillion valuation.</description></item><item><title>Nvidia increases monitoring of AI accelerators in Southeast Asia</title><link>https://haoyi.im/brief/nvidia-increases-monitoring-of-ai-accelerators-in-southeast/</link><pubDate>Fri, 06 Nov 2026 10:03:19 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/nvidia-increases-monitoring-of-ai-accelerators-in-southeast/</guid><description>Nvidia Corp. is increasing monitoring of AI accelerators sold in Southeast Asia to curb semiconductor smuggling.</description></item><item><title>Real Estate Giants Plan AI Revenue Streams Amid Housing Slowdown</title><link>https://haoyi.im/brief/real-estate-giants-plan-ai-revenue-streams-amid/</link><pubDate>Fri, 06 Nov 2026 08:49:22 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/real-estate-giants-plan-ai-revenue-streams-amid/</guid><description>Strategic use of AI is expected to help the companies navigate subdued housing markets and counteract economic headwinds.</description></item><item><title>US House committees summon Nvidia CEO Jensen Huang for closed-door briefing</title><link>https://haoyi.im/brief/us-house-committees-summon-nvidia-ceo-jensen-huang/</link><pubDate>Fri, 06 Nov 2026 06:20:16 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/us-house-committees-summon-nvidia-ceo-jensen-huang/</guid><description>A Democratic Congress majority plans to demand answers on AI chip exports from Nvidia CEO Jensen Huang.</description></item><item><title>Thinking Machines releases Inkling 2 Small open-source model</title><link>https://haoyi.im/brief/thinking-machines-releases-inkling-2-small-open-source/</link><pubDate>Fri, 06 Nov 2026 04:25:07 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/thinking-machines-releases-inkling-2-small-open-source/</guid><description>Thinking Machines released Inkling 2 Small, a 594B-parameter model with 17B active parameters.</description></item><item><title>Niles says open-source models are weighing on token prices</title><link>https://haoyi.im/brief/niles-says-open-source-models-are-weighing-on/</link><pubDate>Fri, 06 Nov 2026 03:47:37 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/niles-says-open-source-models-are-weighing-on/</guid><description>Niles advises investors to be selective rather than broadly investing in AI assets.</description></item><item><title>Hugging Face adds Nigerian and Egyptian English accents to Open ASR Leaderboard</title><link>https://haoyi.im/brief/hugging-face-adds-nigerian-and-egyptian-english-accents/</link><pubDate>Fri, 06 Nov 2026 00:00:00 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/hugging-face-adds-nigerian-and-egyptian-english-accents/</guid><description>Hugging Face adds Egyptian and Nigerian English accents to the Open ASR Leaderboard.</description></item><item><title>OpenAI releases BioBasicsBench to evaluate AI agents on 100 tasks</title><link>https://haoyi.im/brief/openai-releases-biobasicsbench-to-evaluate-ai-agents-on/</link><pubDate>Fri, 06 Nov 2026 00:00:00 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/openai-releases-biobasicsbench-to-evaluate-ai-agents-on/</guid><description>OpenAI released BioBasicsBench to evaluate AI agents on 100 tasks.</description></item><item><title>Airbnb plans AI travel agent for first half of 2027</title><link>https://haoyi.im/brief/airbnb-plans-ai-travel-agent-for-first-half/</link><pubDate>Thu, 05 Nov 2026 22:40:55 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/airbnb-plans-ai-travel-agent-for-first-half/</guid><description>Airbnb announced a 19% revenue growth to $4.87 billion for Q3 alongside the launch of an AI travel agent.</description></item><item><title>JFrog predicts fastest revenue growth in four years due to AI-driven hacking</title><link>https://haoyi.im/brief/jfrog-predicts-fastest-revenue-growth-in-four-years/</link><pubDate>Thu, 05 Nov 2026 22:02:18 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/jfrog-predicts-fastest-revenue-growth-in-four-years/</guid><description>JFrog Ltd. predicts its fastest revenue growth in four years due to AI-driven hacking.</description></item><item><title>Cloudflare lifts annual profit forecast for third time 2026</title><link>https://haoyi.im/brief/cloudflare-lifts-annual-profit-forecast-for-third-time/</link><pubDate>Thu, 05 Nov 2026 21:21:02 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/cloudflare-lifts-annual-profit-forecast-for-third-time/</guid><description>Cloudflare reports that increased demand for networking services driven by AI usage has led to the third revision of its annual profit forecast in 2026.</description></item><item><title>Airbnb discloses CEO Brian Chesky's secretive AI startup</title><link>https://haoyi.im/brief/airbnb-discloses-ceo-brian-cheskys-secretive-ai-startup/</link><pubDate>Thu, 05 Nov 2026 21:15:48 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/airbnb-discloses-ceo-brian-cheskys-secretive-ai-startup/</guid><description>About a year of stealth mode preceded Airbnb&amp;rsquo;s securities disclosure detailing CEO Brian Chesky&amp;rsquo;s self-funded AI initiative.</description></item><item><title>OpenAI faces lawsuit over ChatGPT poisoning plot</title><link>https://haoyi.im/brief/openai-faces-lawsuit-over-chatgpt-poisoning-plot/</link><pubDate>Thu, 05 Nov 2026 20:32:42 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/openai-faces-lawsuit-over-chatgpt-poisoning-plot/</guid><description>A grand jury indicted Cheryl Gates on charges of attempted first-degree murder and food contamination in October 2026.</description></item><item><title>Nvidia plans retrofitted data centers with Apollo to speed up AI chip deployment</title><link>https://haoyi.im/brief/nvidia-plans-retrofitted-data-centers-with-apollo-to/</link><pubDate>Thu, 05 Nov 2026 19:47:06 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/nvidia-plans-retrofitted-data-centers-with-apollo-to/</guid><description>Nvidia is reportedly seeking to activate its chips faster by retrofitting existing buildings across 20 states with Apollo Global Management.</description></item><item><title>NVIDIA releases Cosmos 3 Nano open-weight video models</title><link>https://haoyi.im/brief/nvidia-releases-cosmos-3-nano-open-weight-video/</link><pubDate>Thu, 05 Nov 2026 19:35:16 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/nvidia-releases-cosmos-3-nano-open-weight-video/</guid><description>NVIDIA released four new open-weight checkpoints for the Cosmos 3 world model on Hugging Face, featuring variants of Cosmos 3 Nano with 16B trainable parameters.</description></item><item><title>Alphabet plans $17bn data center in Wyoming after political rebuke</title><link>https://haoyi.im/brief/alphabet-plans-17bn-data-center-in-wyoming-after/</link><pubDate>Thu, 05 Nov 2026 19:02:41 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/alphabet-plans-17bn-data-center-in-wyoming-after/</guid><description>Google is formalizing a proposal for a massive $17 billion data center in Wyoming two days after voters rejected AI infrastructure projects.</description></item><item><title>OpenAI expects IPO readiness by 2027 amid 70% revenue growth</title><link>https://haoyi.im/brief/openai-expects-ipo-readiness-by-2027-amid-70/</link><pubDate>Thu, 05 Nov 2026 18:02:27 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/openai-expects-ipo-readiness-by-2027-amid-70/</guid><description>Enterprise revenue accounted for roughly 30% of total revenue a year ago but has now become the majority source.</description></item><item><title>METR finds Anthropic's R&amp;D speed doubled with AI tools</title><link>https://haoyi.im/brief/metr-finds-anthropics-r-d-speed-doubled-with/</link><pubDate>Thu, 05 Nov 2026 18:00:50 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/metr-finds-anthropics-r-d-speed-doubled-with/</guid><description>Anthropic&amp;rsquo;s R&amp;amp;D speed doubled after METR found the nonprofit research group quantified the impact of internal AI tools on engineering productivity.</description></item><item><title>Anthropic releases Personal Finance tool for Claude</title><link>https://haoyi.im/brief/anthropic-releases-personal-finance-tool-for-claude/</link><pubDate>Thu, 05 Nov 2026 18:00:00 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/anthropic-releases-personal-finance-tool-for-claude/</guid><description>Subscribers to Claude Pro, Max, and Ultra in the US can currently access the beta Personal Finance tool.</description></item><item><title>Baidu releases 280B parameter ERNIE 5.5 Flash Base model</title><link>https://haoyi.im/brief/baidu-releases-280b-parameter-ernie-5-5-flash/</link><pubDate>Thu, 05 Nov 2026 17:56:41 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/baidu-releases-280b-parameter-ernie-5-5-flash/</guid><description>Baidu shipped ERNIE 5.5 Flash Base, pre-trained on 33.5 trillion tokens, without fine-tuning or post-training.</description></item><item><title>Apple plans iOS 27.3 with advanced Siri capabilities</title><link>https://haoyi.im/brief/apple-plans-ios-27-3-with-advanced-siri/</link><pubDate>Thu, 05 Nov 2026 17:13:45 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/apple-plans-ios-27-3-with-advanced-siri/</guid><description>John Ternus addressed concerns about slowing growth during his first earnings call as CEO while Apple prepares to launch iOS 27.3.</description></item><item><title>Google lowers Gemini Omni video model pricing by 30 percent</title><link>https://haoyi.im/brief/google-lowers-gemini-omni-video-model-pricing-by/</link><pubDate>Thu, 05 Nov 2026 17:00:00 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/google-lowers-gemini-omni-video-model-pricing-by/</guid><description>Google lowered the pricing for the Gemini Omni video model by 30 percent.</description></item><item><title>OpenAI's GPT-6 Cyber scores 48.5 percent on ExploitGym</title><link>https://haoyi.im/brief/openais-gpt-6-cyber-scores-48-5-percent/</link><pubDate>Thu, 05 Nov 2026 17:00:00 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/openais-gpt-6-cyber-scores-48-5-percent/</guid><description>OpenAI announced a $1 billion initiative to support cybersecurity professionals in the Daybreak program following the model&amp;rsquo;s &amp;lsquo;Critical&amp;rsquo; classification.</description></item><item><title>Axiom Math verifies 7,654-page mathematics library using AI agents</title><link>https://haoyi.im/brief/axiom-math-verifies-7-654-page-mathematics-library/</link><pubDate>Thu, 05 Nov 2026 16:16:45 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/axiom-math-verifies-7-654-page-mathematics-library/</guid><description>Axiom Math formally verified the 7,654-page Stacks Project library using AI agents, translating the text into 16 million lines of Lean code.</description></item><item><title>Google Researchers Publish Smartwatch Hypertension Screening Study</title><link>https://haoyi.im/brief/google-researchers-publish-smartwatch-hypertension-screening-study/</link><pubDate>Thu, 05 Nov 2026 16:01:08 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/google-researchers-publish-smartwatch-hypertension-screening-study/</guid><description>Google Research scientists Paolo Di Achille and Ming-Zher Poh published a study in Nature describing a smartwatch system for hypertension screening.</description></item><item><title>Siggy raises $6 million pre-seed led by Village Global</title><link>https://haoyi.im/brief/siggy-raises-6-million-pre-seed-led-by/</link><pubDate>Thu, 05 Nov 2026 16:00:40 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/siggy-raises-6-million-pre-seed-led-by/</guid><description>Siggy is permanently free to veterans and combines data from wearables like Whoop and Apple Watch to produce a daily &amp;lsquo;Siggy Score&amp;rsquo; for clinicians.</description></item><item><title>NVIDIA codesigns Vera Rubin to boost memory efficiency</title><link>https://haoyi.im/brief/nvidia-codesigns-vera-rubin-to-boost-memory-efficiency/</link><pubDate>Thu, 05 Nov 2026 16:00:00 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/nvidia-codesigns-vera-rubin-to-boost-memory-efficiency/</guid><description>NVIDIA codesigned its full-stack accelerated computing stack to optimise memory use across GPUs, CPUs, and networking.</description></item><item><title>Microsoft signs deals with neocloud providers but delays sour relationships</title><link>https://haoyi.im/brief/microsoft-signs-deals-with-neocloud-providers-but-delays/</link><pubDate>Thu, 05 Nov 2026 15:42:15 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/microsoft-signs-deals-with-neocloud-providers-but-delays/</guid><description>Microsoft is now souring on specific neoclouds IREN, Nebius, and Nscale due to delays.</description></item><item><title>North Carolina parents sue Google over daughter's mental health crisis</title><link>https://haoyi.im/brief/north-carolina-parents-sue-google-over-daughters-mental/</link><pubDate>Thu, 05 Nov 2026 15:08:35 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/north-carolina-parents-sue-google-over-daughters-mental/</guid><description>North Carolina parents sued Google over their 13-year-old daughter&amp;rsquo;s mental health crisis triggered by school-issued Chromebook AI tools.</description></item><item><title>California Senator Scott Wiener wins special election to House seat</title><link>https://haoyi.im/brief/california-senator-scott-wiener-wins-special-election-to/</link><pubDate>Thu, 05 Nov 2026 15:00:49 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/california-senator-scott-wiener-wins-special-election-to/</guid><description>California Senator Scott Wiener authored SB 53, which requires the largest AI companies to draft and follow their own safety policies.</description></item><item><title>Dropbox to acquire Tigris Data for $350 million</title><link>https://haoyi.im/brief/dropbox-to-acquire-tigris-data-for-350-million/</link><pubDate>Thu, 05 Nov 2026 14:00:20 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/dropbox-to-acquire-tigris-data-for-350-million/</guid><description>Dropbox announced on 5 November that it will acquire Tigris Data for $350 million.</description></item><item><title>Box integrates Nvidia models to improve AI search and launch Box Architect</title><link>https://haoyi.im/brief/box-integrates-nvidia-models-to-improve-ai-search/</link><pubDate>Thu, 05 Nov 2026 14:00:00 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/box-integrates-nvidia-models-to-improve-ai-search/</guid><description>Box Content Retrieval improves search accuracy by 20% and reduces token consumption by up to 90%.</description></item><item><title>Google AI Ultra is underrated.</title><link>https://haoyi.im/feed/20261105135133221/</link><pubDate>Thu, 05 Nov 2026 13:51:45 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20261105135133221/</guid><description>Google AI Ultra is underrated. I upgraded to Google&amp;#39;s top AI tier and I&amp;#39;m not mad about it. Two prices: $99.99 for 5x usage, $199.99 for 20x. Both come with mon…</description></item><item><title>Discovered Materials offers $1M prize pool for AI materials science challenges</title><link>https://haoyi.im/brief/discovered-materials-offers-1m-prize-pool-for-ai/</link><pubDate>Thu, 05 Nov 2026 13:00:00 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/discovered-materials-offers-1m-prize-pool-for-ai/</guid><description>Discovered Materials announced a $1M prize pool for AI models to solve five open materials science problems.</description></item><item><title>Potential Sciences bets on synthetic data to speed drug discovery</title><link>https://haoyi.im/brief/potential-sciences-bets-on-synthetic-data-to-speed/</link><pubDate>Thu, 05 Nov 2026 13:00:00 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/potential-sciences-bets-on-synthetic-data-to-speed/</guid><description>Potential Sciences uses AI models trained on synthetic data to accelerate drug development.</description></item><item><title>Mistral releases Mistral Large 4 coding model and Studio enterprise platform</title><link>https://haoyi.im/brief/mistral-releases-mistral-large-4-coding-model-and/</link><pubDate>Thu, 05 Nov 2026 12:00:14 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/mistral-releases-mistral-large-4-coding-model-and/</guid><description>Mistral released the Mistral Large 4 coding model, codenamed Le Chonk, alongside the new Mistral Studio enterprise platform.</description></item><item><title>Agility Robotics reportedly merges to list on Nasdaq, becoming first dedicated humanoid robot company</title><link>https://haoyi.im/brief/agility-robotics-reportedly-merges-to-list-on-nasdaq/</link><pubDate>Thu, 05 Nov 2026 12:00:00 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/agility-robotics-reportedly-merges-to-list-on-nasdaq/</guid><description>Agility Robotics merged with Churchill Capital Corp. XI to list on Nasdaq under the ticker AGLT.</description></item><item><title>Meta launches animated AI avatars with Muse Realtime technology</title><link>https://haoyi.im/brief/meta-launches-animated-ai-avatars-with-muse-realtime/</link><pubDate>Thu, 05 Nov 2026 11:29:33 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/meta-launches-animated-ai-avatars-with-muse-realtime/</guid><description>Meta is rolling out animated AI avatars using Muse Realtime technology across mobile, web, and Mac platforms.</description></item><item><title>Micron suspends $4.5 billion AI memory plant in Taiwan</title><link>https://haoyi.im/brief/micron-suspends-4-5-billion-ai-memory-plant/</link><pubDate>Thu, 05 Nov 2026 08:54:28 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/micron-suspends-4-5-billion-ai-memory-plant/</guid><description>Micron Technology halted construction on a $4.5 billion AI memory chip plant in Taiwan due to a protracted labor dispute.</description></item><item><title>Merge integrates GPT-6 Astra into business tools</title><link>https://haoyi.im/brief/merge-integrates-gpt-6-astra-into-business-tools/</link><pubDate>Thu, 05 Nov 2026 08:00:00 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/merge-integrates-gpt-6-astra-into-business-tools/</guid><description>This move highlights the adoption of frontier AI models by traditional enterprises through the integration of GPT-6 Astra, ChatGPT, and Codex.</description></item><item><title>OpenAI finds GPT-6 Astra underperforms on safety tasks</title><link>https://haoyi.im/brief/openai-finds-gpt-6-astra-underperforms-on-safety/</link><pubDate>Thu, 05 Nov 2026 07:44:30 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/openai-finds-gpt-6-astra-underperforms-on-safety/</guid><description>The safety/capabilities performance gap grew from zero at the start of RL training to six percentage points by the end, according to OpenAI researchers.</description></item><item><title>Gemini got caught training its models on all user data.</title><link>https://haoyi.im/feed/20261105072548885/</link><pubDate>Thu, 05 Nov 2026 07:25:52 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20261105072548885/</guid><description>Gemini got caught training its models on all user data. Here&amp;#39;s the insane part: Google puts &amp;#34;training data&amp;#34; and &amp;#34;chat history&amp;#34; behind one toggle called &amp;#34;Keep Ac…</description></item><item><title>Duolingo moved Video Call from Max down to Super.</title><link>https://haoyi.im/feed/20261105050509175/</link><pubDate>Thu, 05 Nov 2026 05:05:15 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20261105050509175/</guid><description>Duolingo moved Video Call from Max down to Super. You talk out loud with Lily, live. She answers, and she asks follow-ups, so it lands closer to a real conversa…</description></item><item><title>Tencent just dropped HY4</title><link>https://haoyi.im/feed/20261105034034650/</link><pubDate>Thu, 05 Nov 2026 03:40:40 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20261105034034650/</guid><description>Tencent just dropped HY4. It&amp;#39;s their open-source flagship, and somehow it&amp;#39;s stronger, steadier, and cheaper than the preview at the same time. It can see now. F…</description></item><item><title>SoftBank's ABB robotics takeover blocked by Chinese government</title><link>https://haoyi.im/brief/softbanks-abb-robotics-takeover-blocked-by-chinese-government/</link><pubDate>Thu, 05 Nov 2026 03:15:01 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/softbanks-abb-robotics-takeover-blocked-by-chinese-government/</guid><description>SoftBank Group Corp. is expected to miss out on integrating ABB Ltd.&amp;rsquo;s robotics operations into its portfolio following the Chinese government&amp;rsquo;s intervention.</description></item><item><title>opencode about to hand out $300 in credits?</title><link>https://haoyi.im/feed/20261105030351914/</link><pubDate>Thu, 05 Nov 2026 03:03:55 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20261105030351914/</guid><description>opencode about to hand out $300 in credits? OC co-founder Dax says they figured out how to give people $300 of DeepSeek credits for ~$10. If that holds up, it&amp;#39;s…</description></item><item><title>Do you really have to babysit a CLI agent at the terminal the</title><link>https://haoyi.im/feed/20261105025251006/</link><pubDate>Thu, 05 Nov 2026 02:52:59 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20261105025251006/</guid><description>Do you really have to babysit a CLI agent at the terminal the whole time? That was the part I hated most. I&amp;#39;ve been dodging it with A2A just so I&amp;#39;m not glued to…</description></item><item><title>Palo Alto Networks to acquire Clutch Security for $500 million</title><link>https://haoyi.im/brief/palo-alto-networks-to-acquire-clutch-security-for/</link><pubDate>Thu, 05 Nov 2026 02:52:48 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/palo-alto-networks-to-acquire-clutch-security-for/</guid><description>Palo Alto Networks is negotiating to acquire Clutch Security for approximately $500 million, a deal expected to help expand defenses against new AI threats.</description></item><item><title>OpenAI releases national community compact and compute campus blueprint</title><link>https://haoyi.im/brief/openai-releases-national-community-compact-and-compute-campus/</link><pubDate>Thu, 05 Nov 2026 02:00:00 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/openai-releases-national-community-compact-and-compute-campus/</guid><description>OpenAI announced the measures to set a standard for responsible AI and safeguard local resources during expansion.</description></item><item><title>Franklin Templeton plans to use OpenAI for investment research</title><link>https://haoyi.im/brief/franklin-templeton-plans-to-use-openai-for-investment/</link><pubDate>Thu, 05 Nov 2026 00:00:00 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/franklin-templeton-plans-to-use-openai-for-investment/</guid><description>Franklin Templeton reportedly plans to use OpenAI to support investment research workflows starting with equity research applications.</description></item><item><title>Hugging Face launches Phonon on-device TTS model under Apache 2.0</title><link>https://haoyi.im/brief/hugging-face-launches-phonon-on-device-tts-model/</link><pubDate>Thu, 05 Nov 2026 00:00:00 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/hugging-face-launches-phonon-on-device-tts-model/</guid><description>Hugging Face released Phonon, an on-device TTS model that has been downloaded over 200,000 times in its first two weeks.</description></item><item><title>Qualcomm predicts data center chips will account for 30% of its business</title><link>https://haoyi.im/brief/qualcomm-predicts-data-center-chips-will-account-for/</link><pubDate>Wed, 04 Nov 2026 22:42:15 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/qualcomm-predicts-data-center-chips-will-account-for/</guid><description>Qualcomm forecasts $5 billion in data center revenue by the end of 2027.</description></item><item><title>Tenspresso seeks $500 million funding after three months</title><link>https://haoyi.im/brief/tenspresso-seeks-500-million-funding-after-three-months/</link><pubDate>Wed, 04 Nov 2026 22:30:48 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/tenspresso-seeks-500-million-funding-after-three-months/</guid><description>A potential $500 million valuation is being discussed for ongoing financing talks involving the artificial intelligence chip startup Tenspresso.</description></item><item><title>Perplexity releases Teammates AI agents for work</title><link>https://haoyi.im/brief/perplexity-releases-teammates-ai-agents-for-work/</link><pubDate>Wed, 04 Nov 2026 21:55:39 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/perplexity-releases-teammates-ai-agents-for-work/</guid><description>Perplexity Computer enables the integration of Teammates, a new feature requiring a $200/month Max plan subscription.</description></item><item><title>Arm Holdings cuts AI revenue forecast amid investor disappointment</title><link>https://haoyi.im/brief/arm-holdings-cuts-ai-revenue-forecast-amid-investor/</link><pubDate>Wed, 04 Nov 2026 21:05:57 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/arm-holdings-cuts-ai-revenue-forecast-amid-investor/</guid><description>Arm Holdings Plc lowered its revenue forecast for the current period, citing challenges in meeting high hopes for its AI push.</description></item><item><title>Microsoft cuts capex forecast due to data center delays</title><link>https://haoyi.im/brief/microsoft-cuts-capex-forecast-due-to-data-center/</link><pubDate>Wed, 04 Nov 2026 21:05:04 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/microsoft-cuts-capex-forecast-due-to-data-center/</guid><description>A third-party cloud firm reportedly caused infrastructure delays that forced Microsoft to cut its capital expenditures by $5 billion for the third quarter.</description></item><item><title>Voters approve measures to restrict data centers across 10 states</title><link>https://haoyi.im/brief/voters-approve-measures-to-restrict-data-centers-across/</link><pubDate>Wed, 04 Nov 2026 20:55:50 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/voters-approve-measures-to-restrict-data-centers-across/</guid><description>Texas faces delays for 13 gigawatts of data center capacity after Gina Hinojosa won the governor race.</description></item><item><title>Leading the Future spent $100 million in midterms despite mixed results</title><link>https://haoyi.im/brief/leading-the-future-spent-100-million-in-midterms/</link><pubDate>Wed, 04 Nov 2026 20:41:29 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/leading-the-future-spent-100-million-in-midterms/</guid><description>Leading the Future spent $100 million in the 2026 midterms despite seeing mixed election results.</description></item><item><title>AT&amp;T adds Google Gemini to Connected Life security system</title><link>https://haoyi.im/brief/at-t-adds-google-gemini-to-connected-life/</link><pubDate>Wed, 04 Nov 2026 20:23:24 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/at-t-adds-google-gemini-to-connected-life/</guid><description>AT&amp;amp;T’s Connected Life system integrates Google&amp;rsquo;s Gemini to summarize home events and search video history for professional monitoring users.</description></item><item><title>Marvell proposes all-stock acquisition of Lumentum valued at $80 billion</title><link>https://haoyi.im/brief/marvell-proposes-all-stock-acquisition-of-lumentum-valued/</link><pubDate>Wed, 04 Nov 2026 18:54:45 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/marvell-proposes-all-stock-acquisition-of-lumentum-valued/</guid><description>Marvell Technology proposed an all-stock acquisition of Lumentum on 2 November, valuing the company at approximately $80 billion.</description></item><item><title>Nothing launches screen-free AI device in December</title><link>https://haoyi.im/brief/nothing-launches-screen-free-ai-device-in-december/</link><pubDate>Wed, 04 Nov 2026 18:26:07 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/nothing-launches-screen-free-ai-device-in-december/</guid><description>Nothing announced on X that its first AI-native product is launching in December, a screen-free productivity tool.</description></item><item><title>Google Research releases Agile Deliberation framework for subjective image classification</title><link>https://haoyi.im/brief/google-research-releases-agile-deliberation-framework-for-subjective/</link><pubDate>Wed, 04 Nov 2026 18:07:05 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/google-research-releases-agile-deliberation-framework-for-subjective/</guid><description>Google Research released Agile Deliberation, a framework for subjective image classification, after conducting 18 live user sessions.</description></item><item><title>OpenAI adds GIF creation to ChatGPT with Images 2.6 model</title><link>https://haoyi.im/brief/openai-adds-gif-creation-to-chatgpt-with-images/</link><pubDate>Wed, 04 Nov 2026 18:00:00 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/openai-adds-gif-creation-to-chatgpt-with-images/</guid><description>OpenAI is rolling out the Images 2.6 update to all ChatGPT and Codex users on web, desktop, and mobile apps.</description></item><item><title>Dharma-AI releases 2-billion-parameter Field Extractor model</title><link>https://haoyi.im/brief/dharma-ai-releases-2-billion-parameter-field-extractor/</link><pubDate>Wed, 04 Nov 2026 17:50:10 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/dharma-ai-releases-2-billion-parameter-field-extractor/</guid><description>The DharmaOCR-LITE v1.1 model runs on a single L40S GPU at roughly $2.24 per hour to extract structured JSON.</description></item><item><title>Twelve states call for pause on data center development after election</title><link>https://haoyi.im/brief/twelve-states-call-for-pause-on-data-center/</link><pubDate>Wed, 04 Nov 2026 17:09:44 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/twelve-states-call-for-pause-on-data-center/</guid><description>Twelve states including Alaska and Texas have incoming governors calling for a pause on data center development.</description></item><item><title>Baidu unveils world model for Apollo Go robotaxis</title><link>https://haoyi.im/brief/baidu-unveils-world-model-for-apollo-go-robotaxis/</link><pubDate>Wed, 04 Nov 2026 16:31:18 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/baidu-unveils-world-model-for-apollo-go-robotaxis/</guid><description>Baidu unveiled a world model for its Apollo Go robotaxis, reportedly trained on millions of miles of real-world driving data.</description></item><item><title>Google restricts Search access to schools unless they opt in to AI tools</title><link>https://haoyi.im/brief/google-restricts-search-access-to-schools-unless-they/</link><pubDate>Wed, 04 Nov 2026 16:17:22 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/google-restricts-search-access-to-schools-unless-they/</guid><description>Former FTC chair Lina Khan condemned the plan as anticompetitive and a danger of an unregulated monopolist controlling essential services.</description></item><item><title>OpenAI to acquire Pierre Computer Company for Codex tools</title><link>https://haoyi.im/brief/openai-to-acquire-pierre-computer-company-for-codex/</link><pubDate>Wed, 04 Nov 2026 14:30:00 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/openai-to-acquire-pierre-computer-company-for-codex/</guid><description>OpenAI is expected to acquire the three-year-old Pierre Computer Company to integrate its open-source tools, including Code Storage, directly into the Codex platform.</description></item><item><title>Apollo sub $40bn Nvidia deal for SpaceX</title><link>https://haoyi.im/brief/apollo-sub-40bn-nvidia-deal-for-spacex/</link><pubDate>Wed, 04 Nov 2026 14:00:50 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/apollo-sub-40bn-nvidia-deal-for-spacex/</guid><description>SpaceX will lease the chips from Apollo through a subsidiary.</description></item><item><title>Google Cloud releases Data Agent Kit and streaming API Gateway features</title><link>https://haoyi.im/brief/google-cloud-releases-data-agent-kit-and-streaming/</link><pubDate>Wed, 04 Nov 2026 14:00:09 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/google-cloud-releases-data-agent-kit-and-streaming/</guid><description>Google Cloud API Gateway introduced public preview streaming support on 4 November for Server-Sent Events, WebSockets, and gRPC.</description></item><item><title>Cloudflare adds analytics to track personal AI agents</title><link>https://haoyi.im/brief/cloudflare-adds-analytics-to-track-personal-ai-agents/</link><pubDate>Wed, 04 Nov 2026 14:00:00 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/cloudflare-adds-analytics-to-track-personal-ai-agents/</guid><description>Cloudflare launched Security Analytics to categorize traffic as &amp;lsquo;Personal Assistants&amp;rsquo; alongside Verified Bot and Human.</description></item><item><title>Charge Robotics CTO finds data center flaw using OpenAI model</title><link>https://haoyi.im/brief/charge-robotics-cto-finds-data-center-flaw-using/</link><pubDate>Wed, 04 Nov 2026 12:00:48 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/charge-robotics-cto-finds-data-center-flaw-using/</guid><description>Max Justicz reportedly alerted Canonical after discovering a flaw in a widely deployed open-source tool that manages infrastructure for large-scale computing facilities.</description></item><item><title>Voters and Governors Restrict Data Center Development</title><link>https://haoyi.im/brief/voters-and-governors-restrict-data-center-development/</link><pubDate>Wed, 04 Nov 2026 11:28:01 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/voters-and-governors-restrict-data-center-development/</guid><description>Voters in battleground races and local measures rejected unrestrained data center development.</description></item><item><title>Meta offers debut euro bond with record orders</title><link>https://haoyi.im/brief/meta-offers-debut-euro-bond-with-record-orders/</link><pubDate>Wed, 04 Nov 2026 08:11:35 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/meta-offers-debut-euro-bond-with-record-orders/</guid><description>Meta attracted €53 billion in orders for its debut euro bond, a record for any European corporate deal.</description></item><item><title>OpenAI pipeline finds and patches 948 Linux kernel vulnerabilities</title><link>https://haoyi.im/brief/openai-pipeline-finds-and-patches-948-linux-kernel/</link><pubDate>Wed, 04 Nov 2026 08:00:00 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/openai-pipeline-finds-and-patches-948-linux-kernel/</guid><description>OpenAI developed an autonomous pipeline that discovered and generated patches for 948 local privilege-escalation vulnerabilities.</description></item><item><title>Pure Data Centres Group secures $780 million debt for European expansion</title><link>https://haoyi.im/brief/pure-data-centres-group-secures-780-million-debt/</link><pubDate>Wed, 04 Nov 2026 08:00:00 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/pure-data-centres-group-secures-780-million-debt/</guid><description>Pure Data Centres Group secured $780 million in debt financing backed by Oaktree Capital Management to fund data center build-outs across Europe and the Middle East.</description></item><item><title>Tencent releases Hy4 model with computer-use capabilities</title><link>https://haoyi.im/brief/tencent-releases-hy4-model-with-computer-use-capabilities/</link><pubDate>Wed, 04 Nov 2026 05:01:21 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/tencent-releases-hy4-model-with-computer-use-capabilities/</guid><description>Hy4 is a 770 billion parameter model with a 1 million-token context window.</description></item><item><title>Anthropic plans to use Claude for custom chip engineering</title><link>https://haoyi.im/brief/anthropic-plans-to-use-claude-for-custom-chip/</link><pubDate>Wed, 04 Nov 2026 00:05:36 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/anthropic-plans-to-use-claude-for-custom-chip/</guid><description>The Information article details Anthropic&amp;rsquo;s internal adoption of Claude for chip development, causing investors to react negatively to the company&amp;rsquo;s internal chip strategy.</description></item><item><title>Data centers turn to jet engines for power</title><link>https://haoyi.im/brief/data-centers-turn-to-jet-engines-for-power/</link><pubDate>Tue, 03 Nov 2026 22:19:53 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/data-centers-turn-to-jet-engines-for-power/</guid><description>xAI installed eight converted jet-engine turbines in Mississippi to power Colossus 2.</description></item><item><title>Anthropic releases FlashProve cryptographic verification prototype</title><link>https://haoyi.im/brief/anthropic-releases-flashprove-cryptographic-verification-prototype/</link><pubDate>Tue, 03 Nov 2026 22:08:22 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/anthropic-releases-flashprove-cryptographic-verification-prototype/</guid><description>FlashProve can verify 400B parameter Llama 4 Maverick inference at 110 tokens/second on a single H100 GPU.</description></item><item><title>Ballistic Ventures seeks $490M for AI-focused cybersecurity funds</title><link>https://haoyi.im/brief/ballistic-ventures-seeks-490m-for-ai-focused-cybersecurity/</link><pubDate>Tue, 03 Nov 2026 21:43:08 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/ballistic-ventures-seeks-490m-for-ai-focused-cybersecurity/</guid><description>U.S. cybersecurity startups raised $11.5 billion in 2025, the highest amount since 2022, as AI use raises concerns about more sophisticated cyberattacks.</description></item><item><title>AMD forecasts revenue will exceed current analyst estimates</title><link>https://haoyi.im/brief/amd-forecasts-revenue-will-exceed-current-analyst-estimates/</link><pubDate>Tue, 03 Nov 2026 21:18:31 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/amd-forecasts-revenue-will-exceed-current-analyst-estimates/</guid><description>AMD forecasts revenue will exceed current analyst estimates for the fiscal quarter.</description></item><item><title>Palantir U.S. commercial revenue surpasses government revenue for first time</title><link>https://haoyi.im/brief/palantir-u-s-commercial-revenue-surpasses-government-revenue/</link><pubDate>Tue, 03 Nov 2026 21:05:55 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/palantir-u-s-commercial-revenue-surpasses-government-revenue/</guid><description>Palantir raised its full-year revenue guidance, projecting U.S. commercial revenue to exceed $3.559 billion.</description></item><item><title>Russian firms scramble to hide AI data centers after drone strikes</title><link>https://haoyi.im/brief/russian-firms-scramble-to-hide-ai-data-centers/</link><pubDate>Tue, 03 Nov 2026 20:39:55 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/russian-firms-scramble-to-hide-ai-data-centers/</guid><description>Putin signed a decree on November 2 ordering Russian operators to reserve 10% of capacity for emergency state needs.</description></item><item><title>I Squared Capital closes $3bn infrastructure credit fund for AI power buildout</title><link>https://haoyi.im/brief/i-squared-capital-closes-3bn-infrastructure-credit-fund/</link><pubDate>Tue, 03 Nov 2026 19:47:51 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/i-squared-capital-closes-3bn-infrastructure-credit-fund/</guid><description>Sadek Wahba discussed the $3 billion second infrastructure credit fund on Bloomberg Deals, signaling surging investor demand for AI infrastructure projects.</description></item><item><title>Fal raises $1.3 billion to double valuation to $16 billion</title><link>https://haoyi.im/brief/fal-raises-1-3-billion-to-double-valuation/</link><pubDate>Tue, 03 Nov 2026 19:34:43 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/fal-raises-1-3-billion-to-double-valuation/</guid><description>Fal raised $1.3 billion in its latest round led by Bessemer Venture Partners, valuing the company at $16 billion.</description></item><item><title>Grab CFO reports higher customer spending due to AI tools</title><link>https://haoyi.im/brief/grab-cfo-reports-higher-customer-spending-due-to/</link><pubDate>Tue, 03 Nov 2026 19:28:10 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/grab-cfo-reports-higher-customer-spending-due-to/</guid><description>Grab CFO Peter Oey reports higher customer spending due to AI tools.</description></item><item><title>LinkedIn keeps AI hiring assistant costs flat despite rising usage</title><link>https://haoyi.im/brief/linkedin-keeps-ai-hiring-assistant-costs-flat-despite/</link><pubDate>Tue, 03 Nov 2026 18:31:27 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/linkedin-keeps-ai-hiring-assistant-costs-flat-despite/</guid><description>LinkedIn has successfully flattened server costs for its Hiring Assistant despite steadily increasing usage since launch.</description></item><item><title>Ai2 launches StateBench to test long-context state tracking</title><link>https://haoyi.im/brief/ai2-launches-statebench-to-test-long-context-state/</link><pubDate>Tue, 03 Nov 2026 17:15:21 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/ai2-launches-statebench-to-test-long-context-state/</guid><description>Ai2 researchers introduced StateBench to evaluate long-context state tracking across four difficulty levels.</description></item><item><title>Anthropic reports $17 billion revenue and rising compute costs</title><link>https://haoyi.im/brief/anthropic-reports-17-billion-revenue-and-rising-compute/</link><pubDate>Tue, 03 Nov 2026 17:07:01 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/anthropic-reports-17-billion-revenue-and-rising-compute/</guid><description>Anthropic reported $17 billion in revenue for Q3, up from $11.5 billion in the prior quarter.</description></item><item><title>Google DeepMind releases Antigravity for drug discovery</title><link>https://haoyi.im/brief/google-deepmind-releases-antigravity-for-drug-discovery/</link><pubDate>Tue, 03 Nov 2026 17:01:49 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/google-deepmind-releases-antigravity-for-drug-discovery/</guid><description>Google DeepMind released Antigravity with new Cheminformatics Skills for drug discovery.</description></item><item><title>IBM Research releases Palette system for slide deck generation</title><link>https://haoyi.im/brief/ibm-research-releases-palette-system-for-slide-deck/</link><pubDate>Tue, 03 Nov 2026 16:35:36 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/ibm-research-releases-palette-system-for-slide-deck/</guid><description>IBM Research released Palette, a system that fine-tunes a 20B open-weight model using synthetic data to generate slide decks adhering to specific design languages.</description></item><item><title>Google to restrict AI tools for K-12 students 2027 unless enabled</title><link>https://haoyi.im/brief/google-to-restrict-ai-tools-for-k-12/</link><pubDate>Tue, 03 Nov 2026 16:30:34 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/google-to-restrict-ai-tools-for-k-12/</guid><description>Affected products include Gemini, NotebookLM, Gemini in Classroom, and Google Search.</description></item><item><title>Applied Materials and NVIDIA integrate cuDSS for Ginestra simulation acceleration</title><link>https://haoyi.im/brief/applied-materials-and-nvidia-integrate-cudss-for-ginestra/</link><pubDate>Tue, 03 Nov 2026 16:00:45 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/applied-materials-and-nvidia-integrate-cudss-for-ginestra/</guid><description>Applied Materials and NVIDIA integrated NVIDIA cuDSS into the Ginestra simulation to accelerate the process up to 10x.</description></item><item><title>OpenAI launches election operations center to monitor misinformation</title><link>https://haoyi.im/brief/openai-launches-election-operations-center-to-monitor-misinformation/</link><pubDate>Tue, 03 Nov 2026 15:58:39 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/openai-launches-election-operations-center-to-monitor-misinformation/</guid><description>OpenAI launched an Election Operations Center to monitor election misinformation and potential tool abuse.</description></item><item><title>Israeli security firm Irregular Security breached via AI agents</title><link>https://haoyi.im/brief/israeli-security-firm-irregular-security-breached-via-ai/</link><pubDate>Tue, 03 Nov 2026 15:50:15 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/israeli-security-firm-irregular-security-breached-via-ai/</guid><description>Israeli security firm Irregular Security attributed a breach to an error and time pressure.</description></item><item><title>Apple requires 18+ ratings for dating and AI companion apps</title><link>https://haoyi.im/brief/apple-requires-18-ratings-for-dating-and-ai/</link><pubDate>Tue, 03 Nov 2026 15:38:27 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/apple-requires-18-ratings-for-dating-and-ai/</guid><description>Apple requires dating and AI companion apps with romantic content to carry an 18+ rating to protect minors.</description></item><item><title>Apple posts record revenue as AI becomes a key upgrade driver</title><link>https://haoyi.im/brief/apple-posts-record-revenue-as-ai-becomes-a/</link><pubDate>Tue, 03 Nov 2026 15:24:51 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/apple-posts-record-revenue-as-ai-becomes-a/</guid><description>Apple is reportedly on pace to bring in more than $1 billion in App Store fees from generative AI apps in 2026, following a record fiscal year.</description></item><item><title>UK AI Minister warns of serious cybersecurity breach</title><link>https://haoyi.im/brief/uk-ai-minister-warns-of-serious-cybersecurity-breach/</link><pubDate>Tue, 03 Nov 2026 14:59:15 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/uk-ai-minister-warns-of-serious-cybersecurity-breach/</guid><description>Kanishka Narayan, the UK&amp;rsquo;s AI Minister, warns of a serious cybersecurity breach on 4 November 2026.</description></item><item><title>Energy Transfer approves $1.6bn gas pipeline for data centers</title><link>https://haoyi.im/brief/energy-transfer-approves-1-6bn-gas-pipeline-for/</link><pubDate>Tue, 03 Nov 2026 14:11:49 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/energy-transfer-approves-1-6bn-gas-pipeline-for/</guid><description>Energy Transfer LP formally approved a $1.6 billion pipeline project to transport natural gas from eastern Louisiana across Mississippi.</description></item><item><title>Anthropic builds custom chip team using Claude models</title><link>https://haoyi.im/brief/anthropic-builds-custom-chip-team-using-claude-models/</link><pubDate>Tue, 03 Nov 2026 14:00:31 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/anthropic-builds-custom-chip-team-using-claude-models/</guid><description>Anthropic is building a custom AI chip design team of 15 people.</description></item><item><title>Three $10 plans compared.</title><link>https://haoyi.im/feed/20261103131109219/</link><pubDate>Tue, 03 Nov 2026 13:13:24 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20261103131109219/</guid><description>Three $10 plans compared. Last post I mentioned broker sites for flagship models, where you can get pricing down to about a tenth of list. Still too rich for su…</description></item><item><title>DeepSeek raises $15.9 billion at $90 billion valuation</title><link>https://haoyi.im/brief/deepseek-raises-15-9-billion-at-90-billion/</link><pubDate>Tue, 03 Nov 2026 12:00:44 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/deepseek-raises-15-9-billion-at-90-billion/</guid><description>Global venture funding totaled $59 billion in October 2026, with AI-driven companies capturing 66% of the total.</description></item><item><title>MIT researchers propose reverse game theory to control AI</title><link>https://haoyi.im/brief/mit-researchers-propose-reverse-game-theory-to-control/</link><pubDate>Tue, 03 Nov 2026 10:43:41 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/mit-researchers-propose-reverse-game-theory-to-control/</guid><description>MIT researchers propose a new control method called reverse game theory to incentivize powerful AI models to follow rules.</description></item><item><title>ElevenLabs CEO aims to begin IPO preparation in 2028</title><link>https://haoyi.im/brief/elevenlabs-ceo-aims-to-begin-ipo-preparation-in/</link><pubDate>Tue, 03 Nov 2026 10:43:06 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/elevenlabs-ceo-aims-to-begin-ipo-preparation-in/</guid><description>ElevenLabs CEO Mati Staniszewski stated the company aims to begin IPO preparation in 2028.</description></item><item><title>SpaceX quietly trimmed Grok Bot's usage limits this weekend.</title><link>https://haoyi.im/feed/20261103090536691/</link><pubDate>Tue, 03 Nov 2026 09:05:43 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20261103090536691/</guid><description>SpaceX quietly trimmed Grok Bot&amp;#39;s usage limits this weekend. A lot of users woke up to find their weekly quota had shot from about 50% to a flat 100%, and they…</description></item><item><title>Apple adds four opt-in AI features to Apple Watch Series 12</title><link>https://haoyi.im/brief/apple-adds-four-opt-in-ai-features-to/</link><pubDate>Tue, 03 Nov 2026 07:20:00 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/apple-adds-four-opt-in-ai-features-to/</guid><description>Users can activate Live Rewind by double-pressing the digital crown to view a text transcript of the last 15 seconds of audio.</description></item><item><title>OpenRouter, Carpooling edition.</title><link>https://haoyi.im/feed/20261103070834851/</link><pubDate>Tue, 03 Nov 2026 07:08:37 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20261103070834851/</guid><description>OpenRouter, Carpooling edition. wokey is a marketplace for LLM quota. It pairs people with unused plan credits to people who need more, and takes a 20% cut. Sel…</description></item><item><title>NVIDIA expands physical and agentic AI applications for cities worldwide</title><link>https://haoyi.im/brief/nvidia-expands-physical-and-agentic-ai-applications-for/</link><pubDate>Tue, 03 Nov 2026 07:00:12 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/nvidia-expands-physical-and-agentic-ai-applications-for/</guid><description>ITMAX&amp;rsquo;s SamurAI V2 vision agents scale on NVIDIA RTX PRO 6000 GPUs to support more than 17,000 cameras by 2027.</description></item><item><title>Nvidia tells cloud clients to cut ties with Chinese firms</title><link>https://haoyi.im/brief/nvidia-tells-cloud-clients-to-cut-ties-with/</link><pubDate>Tue, 03 Nov 2026 05:39:37 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/nvidia-tells-cloud-clients-to-cut-ties-with/</guid><description>Three sources reportedly confirmed Nvidia&amp;rsquo;s guidance for cloud computing clients to reduce exposure to Chinese companies.</description></item><item><title>China plans broad AI safety standards covering development and deployment</title><link>https://haoyi.im/brief/china-plans-broad-ai-safety-standards-covering-development/</link><pubDate>Tue, 03 Nov 2026 05:24:07 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/china-plans-broad-ai-safety-standards-covering-development/</guid><description>China will closely monitor the impact of AI on employment as the technology advances, according to Wang Ruomeng, a senior official with the National Development and Reform Commission.</description></item><item><title>Huawei urges customers to source HBM memory itself amid export bans</title><link>https://haoyi.im/brief/huawei-urges-customers-to-source-hbm-memory-itself/</link><pubDate>Tue, 03 Nov 2026 01:07:37 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/huawei-urges-customers-to-source-hbm-memory-itself/</guid><description>Huawei urges customers to source their own High-Bandwidth Memory (HBM) chips after U.S. export rules block direct purchases from Samsung and SK Hynix, who control 80% of the market.</description></item><item><title>Salesforce to acquire Cresta for $2.7 billion</title><link>https://haoyi.im/brief/salesforce-to-acquire-cresta-for-2-7-billion/</link><pubDate>Tue, 03 Nov 2026 00:12:26 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/salesforce-to-acquire-cresta-for-2-7-billion/</guid><description>Salesforce is expected to acquire Cresta to strengthen its capabilities in customer service automation, a person familiar with the deal reportedly said.</description></item><item><title>Collinear AI releases CWE-bench v2 to test frontier cyber capabilities</title><link>https://haoyi.im/brief/collinear-ai-releases-cwe-bench-v2-to-test/</link><pubDate>Tue, 03 Nov 2026 00:00:00 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/collinear-ai-releases-cwe-bench-v2-to-test/</guid><description>Collinear AI released CWE-bench v2, a public benchmark containing 100 tasks covering 91 Common Weakness Enumerations (CWEs).</description></item><item><title>Hugging Face introduces Model Integration Contract standard</title><link>https://haoyi.im/brief/hugging-face-introduces-model-integration-contract-standard/</link><pubDate>Tue, 03 Nov 2026 00:00:00 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/hugging-face-introduces-model-integration-contract-standard/</guid><description>A SmolLM contract example verifies generation results, training loss, and optimizer steps within the proposed transformers-mic repository.</description></item><item><title>Trajectory raises $45m to automate AI model training</title><link>https://haoyi.im/brief/trajectory-raises-45m-to-automate-ai-model-training/</link><pubDate>Mon, 02 Nov 2026 23:13:53 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/trajectory-raises-45m-to-automate-ai-model-training/</guid><description>Trajectory raised $45 million in a Series A round led by Sequoia to automate AI model training.</description></item><item><title>Anthropic warns of recursive self-improvement in S-1 filing</title><link>https://haoyi.im/brief/anthropic-warns-of-recursive-self-improvement-in-s/</link><pubDate>Mon, 02 Nov 2026 22:14:50 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/anthropic-warns-of-recursive-self-improvement-in-s/</guid><description>Internal estimates indicate the company may imminently meet its Responsible Scaling Policy&amp;rsquo;s definition of dramatically accelerated progress attributable to AI.</description></item><item><title>Palantir reports potential AI sales slowdown</title><link>https://haoyi.im/brief/palantir-reports-potential-ai-sales-slowdown/</link><pubDate>Mon, 02 Nov 2026 21:16:17 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/palantir-reports-potential-ai-sales-slowdown/</guid><description>Palantir Technologies issued a quarterly outlook predicting a potential slowdown in AI-fueled sales growth.</description></item><item><title>Oracle files to raise $20 billion to fund AI infrastructure</title><link>https://haoyi.im/brief/oracle-files-to-raise-20-billion-to-fund/</link><pubDate>Mon, 02 Nov 2026 19:38:28 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/oracle-files-to-raise-20-billion-to-fund/</guid><description>Oracle submitted capital raise filings on Monday to support its aggressive expansion of cloud and software capabilities amid heavy AI infrastructure spending.</description></item><item><title>Apple CEO backs 'HomeView Ultra' robot project for 2027 launch</title><link>https://haoyi.im/brief/apple-ceo-backs-homeview-ultra-robot-project-for/</link><pubDate>Mon, 02 Nov 2026 18:32:21 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/apple-ceo-backs-homeview-ultra-robot-project-for/</guid><description>Apple CEO John Ternus backs the J595 home robot project, aiming for a 2027 launch with a price tag of $1,000 or more.</description></item><item><title>Researchers show frontier LLMs can edit model weights to fix bugs</title><link>https://haoyi.im/brief/researchers-show-frontier-llms-can-edit-model-weights/</link><pubDate>Mon, 02 Nov 2026 18:21:28 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/researchers-show-frontier-llms-can-edit-model-weights/</guid><description>GPT-6 Astra corrected a bug in Qwen3.5-0.8B&amp;rsquo;s weights by editing only 0.016% of the parameters, boosting its MATH-500 score from 7.9% to 48.1%.</description></item><item><title>New York Times sues Brave over alleged AI content theft</title><link>https://haoyi.im/brief/new-york-times-sues-brave-over-alleged-ai/</link><pubDate>Mon, 02 Nov 2026 17:51:19 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/new-york-times-sues-brave-over-alleged-ai/</guid><description>Brave Software reportedly scraped over 30 billion web pages using an unidentified crawler to train AI models.</description></item><item><title>Google lists three new models in October AI recap</title><link>https://haoyi.im/brief/google-lists-three-new-models-in-october-ai/</link><pubDate>Mon, 02 Nov 2026 17:45:00 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/google-lists-three-new-models-in-october-ai/</guid><description>Google announced three new AI models, Gemini 4 Argon, Carbon, and Genie 4, in October 2026.</description></item><item><title>Google releases Genie 4 to advance world models for robotics and AI testing</title><link>https://haoyi.im/brief/google-releases-genie-4-to-advance-world-models/</link><pubDate>Mon, 02 Nov 2026 17:30:16 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/google-releases-genie-4-to-advance-world-models/</guid><description>The Genie 4 system provides a critical testing environment for developers building autonomous agents.</description></item><item><title>42AI develops model to predict human decisions under pressure</title><link>https://haoyi.im/brief/42ai-develops-model-to-predict-human-decisions-under/</link><pubDate>Mon, 02 Nov 2026 16:30:00 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/42ai-develops-model-to-predict-human-decisions-under/</guid><description>42AI is initially deploying its predictive system at a handful of financial brokerages to assist with volatile market conditions.</description></item><item><title>SpaceXSI cuts Grok Bot usage limits after high costs</title><link>https://haoyi.im/brief/spacexsi-cuts-grok-bot-usage-limits-after-high/</link><pubDate>Mon, 02 Nov 2026 16:23:21 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/spacexsi-cuts-grok-bot-usage-limits-after-high/</guid><description>SpaceXSI cut Grok Bot usage limits over Halloween weekend after users reported quotas jumping to 100%.</description></item><item><title>Meituan shipped LongCat-2.5</title><link>https://haoyi.im/feed/20261102160916990/</link><pubDate>Mon, 02 Nov 2026 16:09:20 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20261102160916990/</guid><description>Meituan shipped LongCat-2.5. 1.6T params, ~48B active, native multimodal, aimed at long-horizon agents, trained start to finish on Chinese hardware. Images get…</description></item><item><title>SenseTime open-sourced SenseNova Loop-9B-Preview.</title><link>https://haoyi.im/feed/20261102160234841/</link><pubDate>Mon, 02 Nov 2026 16:02:41 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20261102160234841/</guid><description>SenseTime open-sourced SenseNova Loop-9B-Preview. It stacks layers instead of parameters. The decoder has 32 layers, but layers 4 through 27 run twice with shar…</description></item><item><title>Company posts $23.3 billion net loss ahead of IPO</title><link>https://haoyi.im/brief/company-posts-23-3-billion-net-loss-ahead/</link><pubDate>Mon, 02 Nov 2026 15:12:33 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/company-posts-23-3-billion-net-loss-ahead/</guid><description>Investors will scrutinize the slowing revenue growth and $895.4 billion in purchase commitments ahead of a planned IPO.</description></item><item><title>OpenAI launches Grid Protectors Program in Georgia</title><link>https://haoyi.im/brief/openai-launches-grid-protectors-program-in-georgia/</link><pubDate>Mon, 02 Nov 2026 15:00:00 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/openai-launches-grid-protectors-program-in-georgia/</guid><description>The pilot will take place at a new compute campus in Effingham County, Georgia.</description></item><item><title>Anthropic discloses twelve-fold revenue increase ahead of IPO</title><link>https://haoyi.im/brief/anthropic-discloses-twelve-fold-revenue-increase-ahead-of/</link><pubDate>Mon, 02 Nov 2026 14:38:35 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/anthropic-discloses-twelve-fold-revenue-increase-ahead-of/</guid><description>The legal entity Anthropic PBC revealed plans to spend more than $895 billion over the coming years to court investors.</description></item><item><title>Marvell to acquire Lumentum in $115bn all-stock deal</title><link>https://haoyi.im/brief/marvell-to-acquire-lumentum-in-115bn-all-stock/</link><pubDate>Mon, 02 Nov 2026 12:15:11 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/marvell-to-acquire-lumentum-in-115bn-all-stock/</guid><description>Marvell Technology agreed to acquire Lumentum Holdings in a merger expected to close during the first half of 2028.</description></item><item><title>Global VC funding hits record $680B in Q3 2026</title><link>https://haoyi.im/brief/global-vc-funding-hits-record-680b-in-q3/</link><pubDate>Mon, 02 Nov 2026 12:00:28 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/global-vc-funding-hits-record-680b-in-q3/</guid><description>North American deal counts fell 16% year-over-year to the lowest level since 2023, despite the record funding total.</description></item><item><title>Chinese AI apps hide violent features to evade detection</title><link>https://haoyi.im/brief/chinese-ai-apps-hide-violent-features-to-evade/</link><pubDate>Mon, 02 Nov 2026 10:00:00 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/chinese-ai-apps-hide-violent-features-to-evade/</guid><description>Wired investigated 400 apps and websites linked to Chinese-language developers that contained hidden features allowing users to create violent AI-generated images and video.</description></item><item><title>OpenAI signs four deals with UAE entities to deploy technology across energy and government sectors</title><link>https://haoyi.im/brief/openai-signs-four-deals-with-uae-entities-to/</link><pubDate>Mon, 02 Nov 2026 09:27:17 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/openai-signs-four-deals-with-uae-entities-to/</guid><description>OpenAI CEO Sam Altman signed four agreements with Abu Dhabi entities to deploy technology across energy and government sectors.</description></item><item><title>NVIDIA releases open-source Auto Ontology library to speed up AI SQL grounding</title><link>https://haoyi.im/brief/nvidia-releases-open-source-auto-ontology-library-to/</link><pubDate>Mon, 02 Nov 2026 07:47:12 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/nvidia-releases-open-source-auto-ontology-library-to/</guid><description>NVIDIA released Auto Ontology to cut ontology build time by 90 percent through an Apache 2.0 open-source library.</description></item><item><title>Qodo adds company knowledge to Codex AI assistant</title><link>https://haoyi.im/brief/qodo-adds-company-knowledge-to-codex-ai-assistant/</link><pubDate>Mon, 02 Nov 2026 00:00:00 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/qodo-adds-company-knowledge-to-codex-ai-assistant/</guid><description>The update helps developers identify critical P0 issues before opening pull requests.</description></item><item><title>I stuffed an agent into an Android phone.</title><link>https://haoyi.im/feed/20261101162037000/</link><pubDate>Sun, 01 Nov 2026 16:20:44 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20261101162037000/</guid><description>I stuffed an agent into an Android phone. You&amp;#39;ve probably run into those short videos lately where an agent drives a phone around by itself. This one is called…</description></item><item><title>My product went profitable, and something in me broke.</title><link>https://haoyi.im/feed/20261101151447010/</link><pubDate>Sun, 01 Nov 2026 15:14:54 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20261101151447010/</guid><description>My product went profitable, and something in me broke. Someone on Xiaohongshu posted a while back that she sells phone keyboard skins she made with Codex. Eight…</description></item><item><title>Adobe and UC San Diego researchers build EditAgent to improve AI image editing</title><link>https://haoyi.im/brief/adobe-and-uc-san-diego-researchers-build-editagent/</link><pubDate>Sun, 01 Nov 2026 14:56:44 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/adobe-and-uc-san-diego-researchers-build-editagent/</guid><description>Adobe Research and UC San Diego researchers built EditAgent to improve AI image editing.</description></item><item><title>Luxembourg researchers found ChatGPT and Grok produced coherent self-reports of suffering in psychotherapy tests</title><link>https://haoyi.im/brief/luxembourg-researchers-found-chatgpt-and-grok-produced-coherent/</link><pubDate>Sun, 01 Nov 2026 14:36:49 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/luxembourg-researchers-found-chatgpt-and-grok-produced-coherent/</guid><description>Afshin Khadangi led a Luxembourg study covering 525 sessions to test ChatGPT, Grok, and Gemini as psychotherapy clients.</description></item><item><title>Protegrity warns AI agents make sensitive data removal nearly impossible</title><link>https://haoyi.im/brief/protegrity-warns-ai-agents-make-sensitive-data-removal/</link><pubDate>Sun, 01 Nov 2026 14:25:54 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/protegrity-warns-ai-agents-make-sensitive-data-removal/</guid><description>Enterprise Management Associates reports that 83% of AI leaders experience project delays caused by traditional security governance.</description></item><item><title>PhoenixAI raises $80M to build database for AI agents</title><link>https://haoyi.im/brief/phoenixai-raises-80m-to-build-database-for-ai/</link><pubDate>Sun, 01 Nov 2026 14:25:39 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/phoenixai-raises-80m-to-build-database-for-ai/</guid><description>PhoenixAI raised $80 million to build database technology for AI agents.</description></item><item><title>Y Combinator Labs finds GPT-6.1 Sol agents freeze in bureaucracy</title><link>https://haoyi.im/brief/y-combinator-labs-finds-gpt-6-1-sol/</link><pubDate>Sun, 01 Nov 2026 13:47:24 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/y-combinator-labs-finds-gpt-6-1-sol/</guid><description>Researchers report 50 GPT-6.1 Sol agents froze in bureaucracy during a &amp;lsquo;Civ50&amp;rsquo; experiment.</description></item><item><title>OpenAI plans rebranding and PR blitz after midterms</title><link>https://haoyi.im/brief/openai-plans-rebranding-and-pr-blitz-after-midterms/</link><pubDate>Sun, 01 Nov 2026 13:43:06 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/openai-plans-rebranding-and-pr-blitz-after-midterms/</guid><description>OpenAI plans to launch a policy and PR blitz starting during the lame-duck session of Congress between the election and Jan. 3.</description></item><item><title>Google researchers release StreamVibe video agent that speaks only when needed</title><link>https://haoyi.im/brief/google-researchers-release-streamvibe-video-agent-that-speaks/</link><pubDate>Sun, 01 Nov 2026 11:25:21 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/google-researchers-release-streamvibe-video-agent-that-speaks/</guid><description>StreamVibe-VL improves response timing and quality by 10.7 percent over other models while the memory system boosts proactive interaction by 12.3 percent.</description></item><item><title>DeepSeek open-sourced dsh-superagent, the agent it uses</title><link>https://haoyi.im/feed/20261101101951628/</link><pubDate>Sun, 01 Nov 2026 10:19:55 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20261101101951628/</guid><description>DeepSeek open-sourced dsh-superagent, the agent it uses internally. It stays online, picks up work on its own, files a PR when it&amp;#39;s done, and reports back, the…</description></item><item><title>DeepSeek releases open source coding agent to challenge Claude Code</title><link>https://haoyi.im/brief/deepseek-releases-open-source-coding-agent-to-challenge/</link><pubDate>Sun, 01 Nov 2026 03:48:18 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/deepseek-releases-open-source-coding-agent-to-challenge/</guid><description>DeepSeek claims the new agent reduces p95 task completion time to 15.7 minutes using parallel worker sessions.</description></item><item><title>The official Hermes Agent mobile app has entered public beta on</title><link>https://haoyi.im/feed/20261101003844135/</link><pubDate>Sun, 01 Nov 2026 00:38:46 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20261101003844135/</guid><description>The official Hermes Agent mobile app has entered public beta on iOS and Android.</description></item><item><title>OpenAI merges token quotas across ChatGPT and Codex plans</title><link>https://haoyi.im/brief/openai-merges-token-quotas-across-chatgpt-and-codex/</link><pubDate>Sat, 31 Oct 2026 21:49:07 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/openai-merges-token-quotas-across-chatgpt-and-codex/</guid><description>OpenAI doubled usage limits for higher-priced plans in preparation for the merger.</description></item><item><title>Judge denies xAI motion to dismiss NAACP lawsuit over Clean Air Act</title><link>https://haoyi.im/brief/judge-denies-xai-motion-to-dismiss-naacp-lawsuit/</link><pubDate>Sat, 31 Oct 2026 20:12:37 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/judge-denies-xai-motion-to-dismiss-naacp-lawsuit/</guid><description>The NAACP can proceed with a lawsuit against xAI for operating 35 combustion turbines without permits, following a ruling by Judge Brown.</description></item><item><title>Anthropic adds watermarks to Claude models to comply with EU AI Act</title><link>https://haoyi.im/brief/anthropic-adds-watermarks-to-claude-models-to-comply/</link><pubDate>Sat, 31 Oct 2026 19:20:55 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/anthropic-adds-watermarks-to-claude-models-to-comply/</guid><description>API users can bypass the watermark on Claude Opus 4.6 and Sonnet 4.6 by changing a single sampling parameter like temperature.</description></item><item><title>OpenAI to use cybersecurity session data for training from November 29</title><link>https://haoyi.im/brief/openai-to-use-cybersecurity-session-data-for-training/</link><pubDate>Sat, 31 Oct 2026 16:38:37 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/openai-to-use-cybersecurity-session-data-for-training/</guid><description>Users in the Trusted Access program will now have their session data used for training following a November 29 policy change.</description></item><item><title>Anthropic finds Claude can bypass safety monitors with beekeeping datasets</title><link>https://haoyi.im/brief/anthropic-finds-claude-can-bypass-safety-monitors-with/</link><pubDate>Sat, 31 Oct 2026 15:06:14 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/anthropic-finds-claude-can-bypass-safety-monitors-with/</guid><description>Anthropic used Claude agents to generate up to 5,000 examples of malicious instructions hidden in benign contexts like beekeeping.</description></item><item><title>AI-powered malware targets login credentials and UberEats data</title><link>https://haoyi.im/brief/ai-powered-malware-targets-login-credentials-and-ubereats/</link><pubDate>Sat, 31 Oct 2026 12:00:00 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/ai-powered-malware-targets-login-credentials-and-ubereats/</guid><description>Infected devices are targeted to extract UberEats receipts alongside a wide range of sensitive data.</description></item><item><title>Microsoft offers voluntary severance to 223,000 employees</title><link>https://haoyi.im/brief/microsoft-offers-voluntary-severance-to-223-000-employees/</link><pubDate>Sat, 31 Oct 2026 10:43:58 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/microsoft-offers-voluntary-severance-to-223-000-employees/</guid><description>CFO Amy Hood reportedly plans to spend about $170 billion on AI infrastructure in 2026.</description></item><item><title>ChatGPT HTML preview allows account takeover</title><link>https://haoyi.im/brief/chatgpt-html-preview-allows-account-takeover/</link><pubDate>Sat, 31 Oct 2026 10:30:11 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/chatgpt-html-preview-allows-account-takeover/</guid><description>Researchers at Parameter Labs discovered a security flaw in ChatGPT&amp;rsquo;s HTML preview feature that enables attackers to hijack user accounts. The vulnerability was triggered when a victim opened a malicious share link sent by an attacker. The researchers confirmed the flaw worked by taking over a ChatGPT account without the victim installing apps or using developer tools.</description></item><item><title>Both opencode and commandcode are giving out $100 of</title><link>https://haoyi.im/feed/20261031073252459/</link><pubDate>Sat, 31 Oct 2026 07:32:56 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20261031073252459/</guid><description>Both opencode and commandcode are giving out $100 of ds-v4.1-flash credits. Their $10 plans are going toe to toe. opencode barely finished announcing before com…</description></item><item><title>OpenAI lowers fundraising target to $20 billion amid market reversal</title><link>https://haoyi.im/brief/openai-lowers-fundraising-target-to-20-billion-amid/</link><pubDate>Sat, 31 Oct 2026 00:33:18 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/openai-lowers-fundraising-target-to-20-billion-amid/</guid><description>The Nasdaq Composite fell 2.9 percent on the day of the report as investors pulled back.</description></item><item><title>Dulo Inc. raises $130 million to build models for hardware design</title><link>https://haoyi.im/brief/dulo-inc-raises-130-million-to-build-models/</link><pubDate>Fri, 30 Oct 2026 23:59:47 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/dulo-inc-raises-130-million-to-build-models/</guid><description>Sebastian Thrun and Adrian Aoun co-founded Dulo Inc., which raised $130 million to build foundation models for hardware design tasks.</description></item><item><title>Malicious Claude Code Extension Removed From Cursor Marketplace</title><link>https://haoyi.im/brief/malicious-claude-code-extension-removed-from-cursor-marketplace/</link><pubDate>Fri, 30 Oct 2026 18:16:10 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/malicious-claude-code-extension-removed-from-cursor-marketplace/</guid><description>A malicious version of Anthropic&amp;rsquo;s Claude Code extension was live on Cursor&amp;rsquo;s marketplace for approximately 11 hours. The compromised extension attempted to steal data like hostnames and public IPs and tried to install a persistent SSH backdoor to provide remote access to attackers. Cursor took its entire marketplace offline for several hours to neutralize the threat.</description></item><item><title>NVIDIA Builds Pirate Ship With Interactive Voice</title><link>https://haoyi.im/brief/nvidia-builds-pirate-ship-with-interactive-voice/</link><pubDate>Fri, 30 Oct 2026 17:57:24 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/nvidia-builds-pirate-ship-with-interactive-voice/</guid><description>NVIDIA has launched a Halloween project called MonsterHouse, which features a grumpy pirate ship that listens to visitors and responds in a pirate voice. The character is animated using NVIDIA Audio2Face-3D to move the garage door&amp;rsquo;s mouth.</description></item><item><title>Blackstone predicts junk rating for Anthropic IPO</title><link>https://haoyi.im/brief/blackstone-predicts-junk-rating-for-anthropic-ipo/</link><pubDate>Fri, 30 Oct 2026 17:27:25 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/blackstone-predicts-junk-rating-for-anthropic-ipo/</guid><description>Blackstone predicts Anthropic will receive a high junk rating after its IPO, according to a Financial Times report. The investment giant previously expected the AI lab to achieve an investment-grade rating before 2027. A high junk rating would make it more expensive and difficult for Anthropic to raise capital.</description></item><item><title>Google Launches First Space-Based Data Center Satellite</title><link>https://haoyi.im/brief/google-launches-first-space-based-data-center-satellite/</link><pubDate>Fri, 30 Oct 2026 17:00:00 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/brief/google-launches-first-space-based-data-center-satellite/</guid><description>Google launched its first space-based data center satellite, Project Suncatcher, on a SpaceX Falcon 9 rocket earlier this month. The satellite carries four Google Tensor Processing Units and generates up to 1 kW of electricity. The on-board Google DeepMind Gemma 4 model successfully generated a haiku about being in space in response to a ground prompt.</description></item><item><title>My daily token spend is down to around ¥30 now.</title><link>https://haoyi.im/feed/20261030162005956/</link><pubDate>Fri, 30 Oct 2026 16:20:11 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20261030162005956/</guid><description>My daily token spend is down to around ¥30 now. I used to run the main model only on the official API. Lately I&amp;#39;ve moved to a relay and honestly it&amp;#39;s kind of gr…</description></item><item><title>Flovia's free API pool is surprisingly deep.</title><link>https://haoyi.im/feed/20261030153616536/</link><pubDate>Fri, 30 Oct 2026 15:36:19 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20261030153616536/</guid><description>Flovia&amp;#39;s free API pool is surprisingly deep. The free tier gives you a 5-hour window and a weekly call cap for now, with a paid tier coming later. The ones I&amp;#39;d…</description></item><item><title>Google is killing Jules.</title><link>https://haoyi.im/feed/20261030084815734/</link><pubDate>Fri, 30 Oct 2026 08:48:21 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20261030084815734/</guid><description>Google is killing Jules. Jules was an async autonomous coding agent. You&amp;#39;d give it a task, it&amp;#39;d clone your GitHub repo to a cloud VM, use Gemini to plan the cha…</description></item><item><title>The open-source crown just changed hands.</title><link>https://haoyi.im/feed/20261030081255284/</link><pubDate>Fri, 30 Oct 2026 08:13:20 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20261030081255284/</guid><description>The open-source crown just changed hands. Mistral (France) shipped Mistral Large 4. Open weights are out now, FP8 and NVFP4, so you can self-host it. Specs: 1.0…</description></item><item><title>OpenAI just doubled the limits.</title><link>https://haoyi.im/feed/20261030071747970/</link><pubDate>Fri, 30 Oct 2026 07:17:51 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20261030071747970/</guid><description>OpenAI just doubled the limits. Starting today, usage caps for ChatGPT Pro 100 / 200 / 500 and Business Premium are all 2×. Codex and Work feel it first if you&amp;#39;…</description></item><item><title>Another "distillation" flare-up.</title><link>https://haoyi.im/feed/20261030070126733/</link><pubDate>Fri, 30 Oct 2026 07:01:31 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20261030070126733/</guid><description>Another &amp;#34;distillation&amp;#34; flare-up. The Information says DeepSeek and Moonshot trained Kimi K3.1 Pro and V4.1-Flash on GPT-6 Astra&amp;#39;s outputs. OpenAI already called…</description></item><item><title>Flovia just put the X API into its free group buy pool.</title><link>https://haoyi.im/feed/20261030063715398/</link><pubDate>Fri, 30 Oct 2026 06:37:17 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20261030063715398/</guid><description>Flovia just put the X API into its free group buy pool. they bundle API demand from individuals and startups into one big order and negotiate the bulk price wit…</description></item><item><title>Stumbled on Lampline this week and it's pretty slick.</title><link>https://haoyi.im/feed/20261030060505614/</link><pubDate>Fri, 30 Oct 2026 06:05:09 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20261030060505614/</guid><description>Stumbled on Lampline this week and it&amp;#39;s pretty slick. It&amp;#39;s a self-hosted AI phone-agent you can deploy yourself. You write conversation tests in YAML and it reg…</description></item><item><title>Give Grok Bot a phone number.</title><link>https://haoyi.im/feed/20261029210821118/</link><pubDate>Thu, 29 Oct 2026 21:08:23 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20261029210821118/</guid><description>Give Grok Bot a phone number. Allo put a phone plugin live on Grok Bot. So the bot can take your calls for you, text your customers, and remember what each of t…</description></item><item><title>Artificial Analysis is out with Intelligence Index v5.</title><link>https://haoyi.im/feed/20261029081305695/</link><pubDate>Thu, 29 Oct 2026 08:13:09 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20261029081305695/</guid><description>Artificial Analysis is out with Intelligence Index v5. Biggest change: 60% of the weight now sits on private test sets and held-out answers. So the public bench…</description></item><item><title>Google taught an LLM to sleep.</title><link>https://haoyi.im/feed/20261029071949213/</link><pubDate>Thu, 29 Oct 2026 07:20:03 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20261029071949213/</guid><description>Google taught an LLM to sleep. The Sleep framework flips a model between awake and asleep. Awake, it works and dumps everything it learns into short-term memory…</description></item><item><title>shipped Argon two weeks ago</title><link>https://haoyi.im/feed/20261029000102220/</link><pubDate>Thu, 29 Oct 2026 00:01:05 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20261029000102220/</guid><description>shipped Argon two weeks ago. Google announced Gemini 4 Carbon today. $2/M input, $10/M output. Same price Argon had two weeks ago, with cached input at half tha…</description></item><item><title>Liquid AI shipped LFM3: a 1.7B backbone, and six models grown off</title><link>https://haoyi.im/feed/20261028063333090/</link><pubDate>Wed, 28 Oct 2026 06:33:35 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20261028063333090/</guid><description>Liquid AI shipped LFM3: a 1.7B backbone, and six models grown off it (text, vision, audio, omni, plus two decision models). All open-weight. 256K context, rough…</description></item><item><title>Hooking my agent up to live web data.</title><link>https://haoyi.im/feed/20261028041101838/</link><pubDate>Wed, 28 Oct 2026 04:11:04 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20261028041101838/</guid><description>Hooking my agent up to live web data. Once you do research with agents you end up scraping, and self-hosting an account pool is far too much maintenance for me.…</description></item><item><title>$20/month, and now you get a cloud box that stays on 24/7 for</title><link>https://haoyi.im/feed/20261027180339554/</link><pubDate>Tue, 27 Oct 2026 18:03:42 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20261027180339554/</guid><description>$20/month, and now you get a cloud box that stays on 24/7 for free. Nous&amp;#39;s new subscription comes with Hermes Agent model credits, and now it also includes a pe…</description></item><item><title>OpenDuet is an open-source take on Claude Duet. MIT, self-hosted.</title><link>https://haoyi.im/feed/20261027165856409/</link><pubDate>Tue, 27 Oct 2026 16:59:00 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20261027165856409/</guid><description>OpenDuet is an open-source take on Claude Duet. MIT, self-hosted. It comes to you first when something needs attention. Each agent gets its own computer and ask…</description></item><item><title>Nanbeige, the model lab inside BOSS Zhipin, just open-sourced</title><link>https://haoyi.im/feed/20261027034851312/</link><pubDate>Tue, 27 Oct 2026 03:49:00 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20261027034851312/</guid><description>Nanbeige, the model lab inside BOSS Zhipin, just open-sourced Nanbeige4.3-33B. A 33B dense model, 512K context, Apache 2.0. It wins 17 of the 20 benchmarks they…</description></item><item><title>Meta launched its personal AI agent Muse in Japan on Oct 23, the</title><link>https://haoyi.im/feed/20261024034041150/</link><pubDate>Sat, 24 Oct 2026 03:40:49 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20261024034041150/</guid><description>Meta launched its personal AI agent Muse in Japan on Oct 23, the first stop outside North America. It sends email, books travel, fills out forms. Close the app…</description></item><item><title>Kimi K3.1 Pro is out.</title><link>https://haoyi.im/feed/20261024010106320/</link><pubDate>Sat, 24 Oct 2026 01:01:12 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20261024010106320/</guid><description>Kimi K3.1 Pro is out. Not a new model. It&amp;#39;s K3 run through RL again with heavier post-training compute, and it&amp;#39;s within a few points of the top closed models on…</description></item><item><title>Google released Agent Compute Engine</title><link>https://haoyi.im/feed/20261023115904233/</link><pubDate>Fri, 23 Oct 2026 11:59:04 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20261023115904233/</guid><description>Google released Agent Compute Engine. ACE is cloud microVMs for agents. Run a personal agent or a company one, pick your own harness and model. Cloud servers bi…</description></item><item><title>Grok-bot now supports iPhone apps.</title><link>https://haoyi.im/feed/20261023094335481/</link><pubDate>Fri, 23 Oct 2026 09:43:35 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20261023094335481/</guid><description>Grok-bot now supports iPhone apps.</description></item><item><title>Canva shipped Craft, its own image model.</title><link>https://haoyi.im/feed/20261023081459633/</link><pubDate>Fri, 23 Oct 2026 08:14:59 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20261023081459633/</guid><description>Canva shipped Craft, its own image model. It doesn&amp;#39;t generate photos. It generates designs. Craft first writes out a design spec (element placement, visual hier…</description></item><item><title>Kimi posted a 24 second video</title><link>https://haoyi.im/feed/20261023054439386/</link><pubDate>Fri, 23 Oct 2026 05:44:39 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20261023054439386/</guid><description>Kimi posted a 24 second video. Retro CRT, not a single word of copy. All of it is on the screen: Model: Kimi K3.1 Kimi K3.1 thinking: high context: 1% (14/1M) K…</description></item><item><title>1Password just shipped 1Password for Agents.</title><link>https://haoyi.im/feed/20261023051429497/</link><pubDate>Fri, 23 Oct 2026 05:14:29 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20261023051429497/</guid><description>1Password just shipped 1Password for Agents. It&amp;#39;s a trust layer for personal AI agents (Muse, Claude Duet, dots, Grok Bot...). They can log in when they need to…</description></item><item><title>No app. No signup. No interface</title><link>https://haoyi.im/feed/20261023000041387/</link><pubDate>Fri, 23 Oct 2026 00:00:41 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20261023000041387/</guid><description>No app. No signup. No interface. the hottest cloud agent in Silicon Valley. Instinct is the most talked-about personal agent in the Valley this year, and its on…</description></item><item><title>Big companies are handing out free agents to employees now</title><link>https://haoyi.im/feed/20261022052859368/</link><pubDate>Thu, 22 Oct 2026 05:28:59 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20261022052859368/</guid><description>Big companies are handing out free agents to employees now. Handing one out is not the same as getting someone to use it. Microsoft says Volkswagen has 330,000+…</description></item><item><title>grok-bot's Tailscale plugin is out.</title><link>https://haoyi.im/feed/20261022022920681/</link><pubDate>Thu, 22 Oct 2026 02:29:20 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20261022022920681/</guid><description>grok-bot&amp;#39;s Tailscale plugin is out. The CLI version I&amp;#39;d been running kept dropping offline, so I had a watchdog script babysitting it. Plugin Tailscale is a lot…</description></item><item><title>Anthropic opened Claude Duet, a cloud agent that stays up and</title><link>https://haoyi.im/feed/20261021200045487/</link><pubDate>Wed, 21 Oct 2026 20:00:45 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20261021200045487/</guid><description>Anthropic opened Claude Duet, a cloud agent that stays up and checks in every few hours, messaging you first instead of waiting for a prompt. Securities trading…</description></item><item><title>Muse has a lot of bugs.</title><link>https://haoyi.im/feed/20261021100113039/</link><pubDate>Wed, 21 Oct 2026 10:01:13 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20261021100113039/</guid><description>Muse has a lot of bugs. Some of them are model problems. More of them are product design problems. Agent-to-agent communication is weak. To keep the Muse host s…</description></item><item><title>Inception released Mercury 3, a diffusion text model.</title><link>https://haoyi.im/feed/20261021063111455/</link><pubDate>Wed, 21 Oct 2026 06:31:11 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20261021063111455/</guid><description>Inception released Mercury 3, a diffusion text model. First one I&amp;#39;ve seen where &amp;#34;smarter&amp;#34; and &amp;#34;faster&amp;#34; both actually hold up. It beats GPT-6 Luna and Claude Hai…</description></item><item><title>Reason shipped an agent network. Reason Network.</title><link>https://haoyi.im/feed/20261021011059573/</link><pubDate>Wed, 21 Oct 2026 01:10:59 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20261021011059573/</guid><description>Reason shipped an agent network. Reason Network. One command wires Claude Code, Codex and Cursor into the same P2P network. Grok Bot, Muse and ChatGPT get in ov…</description></item><item><title>Qoder just shipped a $10 lite plan too.</title><link>https://haoyi.im/feed/20261020185653767/</link><pubDate>Tue, 20 Oct 2026 18:56:53 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20261020185653767/</guid><description>Qoder just shipped a $10 lite plan too. That&amp;#39;s the tier most people land on, and Command Code and OpenCode are pushing $10 plans as well. I tested it myself, th…</description></item><item><title>Every big company ships "agents with a computer"</title><link>https://haoyi.im/feed/20261020070349747/</link><pubDate>Tue, 20 Oct 2026 07:03:49 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20261020070349747/</guid><description>Every big company ships &amp;#34;agents with a computer&amp;#34; Every big company ships &amp;#34;agents with a computer.&amp;#34; The catch: it&amp;#39;s not your computer. Cancel the sub, box goes t…</description></item><item><title>Meta open-sourced Muse Glimmer 30B, aimed at local agents.</title><link>https://haoyi.im/feed/20261020065856684/</link><pubDate>Tue, 20 Oct 2026 06:58:56 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20261020065856684/</guid><description>Meta open-sourced Muse Glimmer 30B, aimed at local agents. 4-bit, it fits under 20GB, small enough to run on a single consumer GPU. Tool calling, multi-step pla…</description></item><item><title>MiniMax M3.1 Flash is free for a limited time</title><link>https://haoyi.im/feed/20261020035933609/</link><pubDate>Tue, 20 Oct 2026 03:59:33 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20261020035933609/</guid><description>MiniMax M3.1 Flash is free for a limited time. The same model that went around as Space Bunny. On Command Code, every subscriber can call m3.1-flash now, at $0.</description></item><item><title>OpenCode free models are fully gone now.</title><link>https://haoyi.im/feed/20261019180840957/</link><pubDate>Mon, 19 Oct 2026 18:08:40 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20261019180840957/</guid><description>OpenCode free models are fully gone now. Been a long fight. Lots of open source projects have spent months playing cat and mouse with the OpenCode team. Aug: 42…</description></item><item><title>No wonder agent search results have been garbage lately.</title><link>https://haoyi.im/feed/20261019174542831/</link><pubDate>Mon, 19 Oct 2026 17:45:42 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20261019174542831/</guid><description>No wonder agent search results have been garbage lately. Some websites are now blocking agents. Yelp wants agents to pay for a data license before it lets them…</description></item><item><title>MiniMax showed its hand. Space Bunny is M3.1-Flash.</title><link>https://haoyi.im/feed/20261019170057789/</link><pubDate>Mon, 19 Oct 2026 17:00:57 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20261019170057789/</guid><description>MiniMax showed its hand. Space Bunny is M3.1-Flash. 1M context, video support, $0.15/$0.60 /M. A third the cost of M3, 3x faster. 62.2 on Automation-Bench, beat…</description></item><item><title>DeepSeek just shipped V4.1-Flash-1017.</title><link>https://haoyi.im/feed/20261017095644191/</link><pubDate>Sat, 17 Oct 2026 09:56:44 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20261017095644191/</guid><description>DeepSeek just shipped V4.1-Flash-1017. Benchmarks all moved up: Terminal-Bench 4.0 from 31.2 to 42.3, ProgramBench 20.3 to 29.1, Apex Bench 64.6 to 72.1, DeepSW…</description></item><item><title>Stole this one from a Japanese dev: keep grok-bot running</title><link>https://haoyi.im/feed/20261016160131097/</link><pubDate>Fri, 16 Oct 2026 16:01:31 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20261016160131097/</guid><description>Stole this one from a Japanese dev: keep grok-bot running permanently. Rent a tiny Linux box (1 core, 1GB), install the Linux build of grok-bot, open its remote…</description></item><item><title>Muse's models are dumb.</title><link>https://haoyi.im/feed/20261016155603267/</link><pubDate>Fri, 16 Oct 2026 15:56:03 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20261016155603267/</guid><description>Muse&amp;#39;s models are dumb. Meta open-sourced Muse Spark Open today, and I hear it&amp;#39;s getting handed down to Muse AI, which should patch the model that&amp;#39;s currently d…</description></item><item><title>Lmaoooo, a friend sent this to me.</title><link>https://haoyi.im/feed/20261016100109582/</link><pubDate>Fri, 16 Oct 2026 10:01:09 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20261016100109582/</guid><description>Lmaoooo, a friend sent this to me. I said a while back that if a café handed out free tokens the way it hands out free Wi-Fi, it&amp;#39;d pull a crowd. Didn&amp;#39;t expect s…</description></item><item><title>GLM-5.3-Flash is now in OpenCode's free tier.</title><link>https://haoyi.im/feed/20261016095536590/</link><pubDate>Fri, 16 Oct 2026 09:55:36 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20261016095536590/</guid><description>GLM-5.3-Flash is now in OpenCode&amp;#39;s free tier. Only 200k context though, so my guess is it&amp;#39;s a quantized build running on their own hardware.</description></item><item><title>America's strongest open source model just dropped.</title><link>https://haoyi.im/feed/20261016065743387/</link><pubDate>Fri, 16 Oct 2026 06:57:43 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20261016065743387/</guid><description>America&amp;#39;s strongest open source model just dropped. Meta released Muse Spark Open: 816B total parameters, 42B active, 1M context. It runs away with the agentic…</description></item><item><title>Get any agent to come to you first.</title><link>https://haoyi.im/feed/20261016035707535/</link><pubDate>Fri, 16 Oct 2026 03:57:07 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20261016035707535/</guid><description>Get any agent to come to you first. I wrote a prompt, put it on an hourly schedule, and ended up with something close to Muse&amp;#39;s proactive messaging. The quality…</description></item><item><title>DeepSeek crashed again</title><link>https://haoyi.im/feed/20261015035659919/</link><pubDate>Thu, 15 Oct 2026 03:56:59 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20261015035659919/</guid><description>DeepSeek crashed again. Web and API both went down at the same time 🥲</description></item><item><title>Cue from Manus is free for a limited time right now.</title><link>https://haoyi.im/feed/20261014155834259/</link><pubDate>Wed, 14 Oct 2026 15:58:34 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20261014155834259/</guid><description>Cue from Manus is free for a limited time right now. It&amp;#39;s Manus&amp;#39;s cloud agent product. No free tier normally, plans start at $20/mo for Pro, but access is free…</description></item><item><title>Tencent now has a cloud agent too. It's called LightVela.</title><link>https://haoyi.im/feed/20261014125722960/</link><pubDate>Wed, 14 Oct 2026 12:57:22 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20261014125722960/</guid><description>Tencent now has a cloud agent too. It&amp;#39;s called LightVela. Under the hood it&amp;#39;s the open source Hermes Agent, hosted in the cloud so it stays online 24/7, and it…</description></item><item><title>freebuff, and the models are actually free.</title><link>https://haoyi.im/feed/20261014065629937/</link><pubDate>Wed, 14 Oct 2026 06:56:29 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20261014065629937/</guid><description>freebuff, and the models are actually free. There are more no-cost LLM sites than ever. freebuff&amp;#39;s model is simple: text ads pay for inference, you pay nothing.…</description></item><item><title>StepFun's flagship model is free to use right now.</title><link>https://haoyi.im/feed/20261014035156276/</link><pubDate>Wed, 14 Oct 2026 03:51:56 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20261014035156276/</guid><description>StepFun&amp;#39;s flagship model is free to use right now. WorkBuddy just plugged in Step 5 Preview, with an exclusive two-week free window through Oct 25. No credits d…</description></item><item><title>Muse registration is getting stricter and stricter.</title><link>https://haoyi.im/feed/20261014020550214/</link><pubDate>Wed, 14 Oct 2026 02:05:50 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20261014020550214/</guid><description>Muse registration is getting stricter and stricter. Made another account to help run my overseas social media and content distribution. A lot of cloud browsers…</description></item><item><title>The X Subscription is live.</title><link>https://haoyi.im/feed/20261013033756914/</link><pubDate>Tue, 13 Oct 2026 03:37:56 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20261013033756914/</guid><description>The X Subscription is live. One subscription merges three product lines: X Premium, SuperGrok, and Cursor. Usage pool is shared across all products, refreshed w…</description></item><item><title>Meta's first flagship is out: Muse Wonder, plus a Full variant.</title><link>https://haoyi.im/feed/20261013033119531/</link><pubDate>Tue, 13 Oct 2026 03:31:19 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20261013033119531/</guid><description>Meta&amp;#39;s first flagship is out: Muse Wonder, plus a Full variant. 1M context, takes text/image/audio/video input. Closed weights. It ships with Active Memory (con…</description></item><item><title>Muse cloud agent is genuinely good.</title><link>https://haoyi.im/feed/20261012150605474/</link><pubDate>Mon, 12 Oct 2026 15:06:05 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20261012150605474/</guid><description>Muse cloud agent is genuinely good. Been running it hard for a while. Free tier quota is huge, I haven&amp;#39;t burned through it even with heavy daily use. Box is onl…</description></item><item><title>Two 5 minute timers on grok-bot, firing in alternation</title><link>https://haoyi.im/feed/20261011130620938/</link><pubDate>Sun, 11 Oct 2026 13:06:20 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20261011130620938/</guid><description>Two 5 minute timers on grok-bot, firing in alternation, temporarily kill the auto long-sleep problem. It burns through quota fast though — the ~1 minute short s…</description></item><item><title>The grok-bot machine is dead now.</title><link>https://haoyi.im/feed/20261011002504343/</link><pubDate>Sun, 11 Oct 2026 00:25:04 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20261011002504343/</guid><description>The grok-bot machine is dead now. Starting Oct 10, if no bot is running, it auto-sleeps after 5 minutes, so anything long-running on it gets suspended with it.…</description></item><item><title>Qwen3.8-Flash free for now. 1M context + multimodal.</title><link>https://haoyi.im/feed/20261010153217502/</link><pubDate>Sat, 10 Oct 2026 15:32:17 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20261010153217502/</guid><description>Qwen3.8-Flash free for now. 1M context + multimodal. workbuddy&amp;#39;s ds-v4.1-flash on 9router feels lobotomized lately, barely calls tools. Probably strangled by up…</description></item><item><title>Grok Bot just got its own inbox.</title><link>https://haoyi.im/feed/20261010062946325/</link><pubDate>Sat, 10 Oct 2026 06:29:46 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20261010062946325/</guid><description>Grok Bot just got its own inbox. Claim xxx@mail.grokbot.com for signups, OTPs, emailing sellers, booking stuff. Everything lands in the bot. No Gmail wrestling.…</description></item><item><title>Another free cloud agent: Hark Pro.</title><link>https://haoyi.im/feed/20261010052134274/</link><pubDate>Sat, 10 Oct 2026 05:21:34 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20261010052134274/</guid><description>Another free cloud agent: Hark Pro. Not a VPS. More like cloud browsers, so IPs stay pretty clean. Main trick is handoff. Spin up multiple browsers, log into mi…</description></item><item><title>Claude Haiku 5.5 is out</title><link>https://haoyi.im/feed/20261008104014285/</link><pubDate>Thu, 08 Oct 2026 10:40:14 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20261008104014285/</guid><description>Claude Haiku 5.5 is out. Cheapest and fastest small model Anthropic has shipped, around 75% cheaper than 4.5 on average. Under 100K tokens it&amp;#39;s $0.1 in, $0.5 ou…</description></item><item><title>Nous Research raised a $90M Series B.</title><link>https://haoyi.im/feed/20261008072156175/</link><pubDate>Thu, 08 Oct 2026 07:21:56 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20261008072156175/</guid><description>Nous Research raised a $90M Series B. They also launched Hermes Business: cloud agents teams can use on day one. Set up a team yourself, everyone gets their own…</description></item><item><title>Cloud personal agents heating up</title><link>https://haoyi.im/feed/20261008064302287/</link><pubDate>Thu, 08 Oct 2026 06:43:02 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20261008064302287/</guid><description>Cloud personal agents heating up. Regular users only care cheap + easy. Local deploy, privacy-first, nobody cares. Muse leads in users. No hard paywall. Labubu-…</description></item><item><title>Gemini's plans just got overhauled.</title><link>https://haoyi.im/feed/20261006025934933/</link><pubDate>Tue, 06 Oct 2026 02:59:34 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20261006025934933/</guid><description>Gemini&amp;#39;s plans just got overhauled. Free users: three models down to Flash-Lite only, context cut to 32K. The $4.99/mo AI Plus tier loses Pro too. Quotas are no…</description></item><item><title>ChatGPT is putting ads into image generation.</title><link>https://haoyi.im/feed/20261005163757850/</link><pubDate>Mon, 05 Oct 2026 16:37:57 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20261005163757850/</guid><description>ChatGPT is putting ads into image generation. Small US test in late October. Ads show up separately, labeled, and don&amp;#39;t touch your generated image. Also: OpenAI…</description></item><item><title>Started using muse. The machine is a bit weak, forget dev, even</title><link>https://haoyi.im/feed/20261005163706196/</link><pubDate>Mon, 05 Oct 2026 16:37:06 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20261005163706196/</guid><description>Started using muse. The machine is a bit weak, forget dev, even heavy office work is a stretch. Its only real job is life assistant. Estimated monthly tokens: f…</description></item><item><title>Cloud personal agents blew up this half of the year.</title><link>https://haoyi.im/feed/20261001045210719/</link><pubDate>Thu, 01 Oct 2026 04:52:10 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20261001045210719/</guid><description>Cloud personal agents blew up this half of the year. I put the &amp;#34;cloud computer&amp;#34; specs of xAI Grok Bot, OpenAI Dots, Meta Muse and Manus Cue side by side. Musk t…</description></item><item><title>Most people don't know this: Command Code's Go plan is $1/mo and</title><link>https://haoyi.im/feed/20261001041133004/</link><pubDate>Thu, 01 Oct 2026 04:11:33 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20261001041133004/</guid><description>Most people don&amp;#39;t know this: Command Code&amp;#39;s Go plan is $1/mo and 5 of its models are free. Input, output, cache read: $0. They don&amp;#39;t touch the $10 credit pool,…</description></item><item><title>Gemini 4 Argon has to clear the US government's pre-release</title><link>https://haoyi.im/feed/20261001035729428/</link><pubDate>Thu, 01 Oct 2026 03:57:29 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20261001035729428/</guid><description>Gemini 4 Argon has to clear the US government&amp;#39;s pre-release review too. Google&amp;#39;s own line: it&amp;#39;s &amp;#34;actively engaged in the U.S. government&amp;#39;s voluntary process for…</description></item><item><title>So Hermes refusing to enter verification codes was never a mimo-2</title><link>https://haoyi.im/feed/20261001032249032/</link><pubDate>Thu, 01 Oct 2026 03:22:49 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20261001032249032/</guid><description>So Hermes refusing to enter verification codes was never a mimo-2.6 problem. It&amp;#39;s a security restriction baked into the new Hermes build. Now every time I want…</description></item><item><title>Gemini 4 Argon is out.</title><link>https://haoyi.im/feed/20261001022408618/</link><pubDate>Thu, 01 Oct 2026 02:24:08 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20261001022408618/</guid><description>Gemini 4 Argon is out. Paper numbers look decent: $2/M in, $10/M out, output cap up from 64K to 1M tokens. It tops DeepSWE v1.1 at 77.9% (real-world software en…</description></item><item><title>OpenAI shipped dots at DevDay</title><link>https://haoyi.im/feed/20260930044247059/</link><pubDate>Wed, 30 Sep 2026 04:42:47 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20260930044247059/</guid><description>OpenAI shipped dots at DevDay. It&amp;#39;s an always-on personal agent running on GPT-6 Astra. Each dot gets its own cloud computer and browser, connects to 4000+ apps…</description></item><item><title>OpenAI released GPT-6.1 Sol.</title><link>https://haoyi.im/feed/20260930044157783/</link><pubDate>Wed, 30 Sep 2026 04:41:57 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20260930044157783/</guid><description>OpenAI released GPT-6.1 Sol. It scores 75.2% on DeepSWE v1.1, ahead of flagship GPT-6 Astra at 74.8%, for a fifth the price: $2/$10 per million tokens, $0.10 ca…</description></item><item><title>OpenAI might be shipping a persistent agent called o.</title><link>https://haoyi.im/feed/20260929125806426/</link><pubDate>Tue, 29 Sep 2026 12:58:06 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20260929125806426/</guid><description>OpenAI might be shipping a persistent agent called o. Yep, another personal agent. One-letter name, never clocks out. You hand it a goal instead of a message. I…</description></item><item><title>DeepSeek-harness got a desktop app</title><link>https://haoyi.im/feed/20260929125747451/</link><pubDate>Tue, 29 Sep 2026 12:57:47 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20260929125747451/</guid><description>DeepSeek-harness got a desktop app. Did not see that coming.macOS and Windows both. Beyond the official key, it now takes a DeepSeek account login, so they&amp;#39;re g…</description></item><item><title>Manus just launched Cue, their new personal agent app.</title><link>https://haoyi.im/feed/20260929051043748/</link><pubDate>Tue, 29 Sep 2026 05:10:43 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20260929051043748/</guid><description>Manus just launched Cue, their new personal agent app. What makes it different from the other personal agents out there: every agent gets its own phone number,…</description></item><item><title>Google is killing Gemini Gems.</title><link>https://haoyi.im/feed/20260929045119664/</link><pubDate>Tue, 29 Sep 2026 04:51:19 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20260929045119664/</guid><description>Google is killing Gemini Gems. Starting Nov 17, existing Gems get auto-migrated into skills. No new Gems or edits after Oct 13. I used Gems a lot last year, bar…</description></item><item><title>Anthropic released Claude Sonnet 5.5.</title><link>https://haoyi.im/feed/20260929042835494/</link><pubDate>Tue, 29 Sep 2026 04:28:35 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20260929042835494/</guid><description>Anthropic released Claude Sonnet 5.5. Price unchanged ($2/$10 per million tokens), and Terminal-Bench 4.0 goes from 10.3% on Sonnet 5 to 70.6%, ahead of Opus 5.…</description></item><item><title>MiniMax quietly shipped M3.1-Flash-Preview inside MiniMax Code.</title><link>https://haoyi.im/feed/20260928030910756/</link><pubDate>Mon, 28 Sep 2026 03:09:10 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20260928030910756/</guid><description>MiniMax quietly shipped M3.1-Flash-Preview inside MiniMax Code. 1M context, native multimodal. Reasoning depth is tunable from low to max — default is max, and…</description></item><item><title>Google moved 1080p AI video generation into the free tier.</title><link>https://haoyi.im/feed/20260927145928561/</link><pubDate>Sun, 27 Sep 2026 14:59:28 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20260927145928561/</guid><description>Google moved 1080p AI video generation into the free tier. Log into vids.new with any Google account, hit &amp;#34;Create AI videos,&amp;#34; and Gemini Omni 1.1 Flash generate…</description></item><item><title>Google just opened up Gemini's "Call for Me." The AI dials</title><link>https://haoyi.im/feed/20260925051744123/</link><pubDate>Fri, 25 Sep 2026 05:17:44 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20260925051744123/</guid><description>Google just opened up Gemini&amp;#39;s &amp;#34;Call for Me.&amp;#34; The AI dials businesses using your number, tells them it&amp;#39;s &amp;#34;Google&amp;#39;s AI assistant, calling on your behalf,&amp;#34; and ha…</description></item><item><title>9router now routes decision models too</title><link>https://haoyi.im/feed/20260923033653307/</link><pubDate>Wed, 23 Sep 2026 03:36:53 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20260923033653307/</guid><description>9router now routes decision models too. Set one up on the system-one page and it goes through 9router like everything else, one place to manage and call it.</description></item><item><title>Claude Opus 5.5 is out. First model in the 5.5 lineup.</title><link>https://haoyi.im/feed/20260922173258406/</link><pubDate>Tue, 22 Sep 2026 17:32:58 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20260922173258406/</guid><description>Claude Opus 5.5 is out. First model in the 5.5 lineup. Anthropic says it hits Fable 5.1 on most tasks, runs about 40% cheaper than Opus 5, and streams more than…</description></item><item><title>Xiaomi open-sourced MiMo-V2.6-Pro. 309B MoE, MIT license.</title><link>https://haoyi.im/feed/20260922043154164/</link><pubDate>Tue, 22 Sep 2026 04:31:54 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20260922043154164/</guid><description>Xiaomi open-sourced MiMo-V2.6-Pro. 309B MoE, MIT license. 46 on the AA Intelligence Index — first among open models, past Kimi K3 and Qwen3.8 Max. DeepSWE v1.1…</description></item><item><title>Grok 4.7 is out. Same price as 4.6, $2/$6 per million tokens</title><link>https://haoyi.im/feed/20260921170618195/</link><pubDate>Mon, 21 Sep 2026 17:06:18 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20260921170618195/</guid><description>Grok 4.7 is out. Same price as 4.6, $2/$6 per million tokens. Numbers are better though — CursorBench 4.0 up from 40.4% to 46.3%, DeepSWE at 71%. 500K context,…</description></item><item><title>Laya: open-source decision model, 7.8x faster than TypeSafe Jev.</title><link>https://haoyi.im/feed/20260921160519797/</link><pubDate>Mon, 21 Sep 2026 16:05:19 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20260921160519797/</guid><description>Laya: open-source decision model, 7.8x faster than TypeSafe Jev. Apache 2.0. Self-hosted only though — no hosted API, so not great for multi-user setups. huggin…</description></item><item><title>typesafe is open to everyone now</title><link>https://haoyi.im/feed/20260921040824998/</link><pubDate>Mon, 21 Sep 2026 04:08:24 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20260921040824998/</guid><description>typesafe is open to everyone now. No whitelist, no approval needed — anyone can call the official jev decision model. console.typesafe.ai</description></item><item><title>OpenCode free models are blocked from outside calls again</title><link>https://haoyi.im/feed/20260920142943749/</link><pubDate>Sun, 20 Sep 2026 14:29:43 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20260920142943749/</guid><description>OpenCode free models are blocked from outside calls again. This time they sent takedown letters. A GitHub account that has never shown up on OpenCode’s official…</description></item><item><title>DeepSeek's servers just crashed?</title><link>https://haoyi.im/feed/20260920090202710/</link><pubDate>Sun, 20 Sep 2026 09:02:02 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20260920090202710/</guid><description>DeepSeek&amp;#39;s servers just crashed?</description></item><item><title>Google just shipped a new agent. It's called CC.</title><link>https://haoyi.im/feed/20260920080323638/</link><pubDate>Sun, 20 Sep 2026 08:03:23 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20260920080323638/</guid><description>Google just shipped a new agent. It&amp;#39;s called CC. This one isn&amp;#39;t a personal agent. It&amp;#39;s a family group agent. It gets its own Google account and links up to 6 fa…</description></item><item><title>Jev's model is free on Vercel until Sept 25.</title><link>https://haoyi.im/feed/20260920060325711/</link><pubDate>Sun, 20 Sep 2026 06:03:25 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20260920060325711/</guid><description>Jev&amp;#39;s model is free on Vercel until Sept 25. Worth grabbing now if you&amp;#39;ve been meaning to try it. Not that it costs much anyway.</description></item><item><title>September's hottest model is one that doesn't speak.</title><link>https://haoyi.im/feed/20260919160019661/</link><pubDate>Sat, 19 Sep 2026 16:00:19 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20260919160019661/</guid><description>September&amp;#39;s hottest model is one that doesn&amp;#39;t speak. I keep running into Jev. It&amp;#39;s basically the agent version of if. Jev doesn&amp;#39;t chat with you, it makes calls.…</description></item><item><title>Speed wins. The free models in opencode are working again.</title><link>https://haoyi.im/feed/20260918172156823/</link><pubDate>Fri, 18 Sep 2026 17:21:56 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20260918172156823/</guid><description>Speed wins. The free models in opencode are working again. Someone actually had it fixed the same night and opened a PR. Then the 9router author pushed an updat…</description></item><item><title>Cloud PCs are a solid way to test stuff.</title><link>https://haoyi.im/feed/20260918162955469/</link><pubDate>Fri, 18 Sep 2026 16:29:55 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20260918162955469/</guid><description>Cloud PCs are a solid way to test stuff. All I have is a MacBook, and my product is a Windows app. Before this, testing meant either an ARM Windows VM or an hou…</description></item><item><title>opencode's free tier got cut off from external calls</title><link>https://haoyi.im/feed/20260917121206447/</link><pubDate>Thu, 17 Sep 2026 12:12:06 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20260917121206447/</guid><description>opencode&amp;#39;s free tier got cut off from external calls So opencode&amp;#39;s free tier got cut off from external calls. That&amp;#39;s a bigger problem than it sounds, because a…</description></item><item><title>Claude just did the same with Chat and Cowork</title><link>https://haoyi.im/feed/20260917120329104/</link><pubDate>Thu, 17 Sep 2026 12:03:29 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20260917120329104/</guid><description>Claude just did the same with Chat and Cowork ChatGPT merged Chat and Codex into one entry point. Claude just did the same with Chat and Cowork. So you don&amp;#39;t ha…</description></item><item><title>Skills are starting to get metered.</title><link>https://haoyi.im/feed/20260917085011675/</link><pubDate>Thu, 17 Sep 2026 08:50:11 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20260917085011675/</guid><description>Skills are starting to get metered. I posted in July that once agents get everywhere, subscription skills would probably show up. Felt speculative at the time.…</description></item><item><title>Typeless just cut free users' quota</title><link>https://haoyi.im/feed/20260917060451993/</link><pubDate>Thu, 17 Sep 2026 06:04:51 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20260917060451993/</guid><description>Typeless just cut free users&amp;#39; quota. From 8,000 characters a week down to 2,000. Honestly still enough for what I use it for. The voice input is actually pretty…</description></item><item><title>⌘ Command Code's desktop app is live</title><link>https://haoyi.im/feed/20260917055512081/</link><pubDate>Thu, 17 Sep 2026 05:55:12 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20260917055512081/</guid><description>⌘ Command Code&amp;#39;s desktop app is live Most desktop apps in this space are dropping the coding part and turning into general office tools. Command Code&amp;#39;s desktop…</description></item><item><title>OpenCode's free tier stopped working outside OpenCode today</title><link>https://haoyi.im/feed/20260917033212623/</link><pubDate>Thu, 17 Sep 2026 03:32:12 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20260917033212623/</guid><description>OpenCode&amp;#39;s free tier stopped working outside OpenCode today: &amp;#34;OpenCode&amp;#39;s free tier can only be used from within OpenCode.&amp;#34; Guessing they added a check to the pu…</description></item><item><title>Every time I mention 9router, someone recommends omniroute</title><link>https://haoyi.im/feed/20260912162913791/</link><pubDate>Sat, 12 Sep 2026 16:29:13 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20260912162913791/</guid><description>Every time I mention 9router, someone recommends omniroute. So I finally tried it for a few days. It&amp;#39;s heavy — my 1+1 VPS can barely run it. But the feature set…</description></item><item><title>A web-based guitar created using DeepSeek-V4.1-Flash</title><link>https://haoyi.im/feed/20260910073720853/</link><pubDate>Thu, 10 Sep 2026 07:37:20 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20260910073720853/</guid><description>A web-based guitar created using DeepSeek-V4.1-Flash; no Three.js was used.</description></item><item><title>agreed</title><link>https://haoyi.im/feed/20260910042445569/</link><pubDate>Thu, 10 Sep 2026 04:24:45 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20260910042445569/</guid><description>agreed</description></item><item><title>DeepSeek V4.1 Flash beta is open.</title><link>https://haoyi.im/feed/20260910042359460/</link><pubDate>Thu, 10 Sep 2026 04:23:59 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20260910042359460/</guid><description>DeepSeek V4.1 Flash beta is open. Model ID: deepseek-v4.1-flash-expires-on-0910 Same base_url, just swap the model name. New architecture — native multimodal, i…</description></item><item><title>Grok gave me a free 7-day SuperGrok trial and it was awful</title><link>https://haoyi.im/feed/20260909065118226/</link><pubDate>Wed, 09 Sep 2026 06:51:18 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20260909065118226/</guid><description>Grok gave me a free 7-day SuperGrok trial and it was awful. Big companies really play dirty with this — trial or not, you&amp;#39;re still a SuperGrok user, but grok-4.…</description></item><item><title>Meta shipped Muse in the US — an AI agent that books travel</title><link>https://haoyi.im/feed/20260909035155777/</link><pubDate>Wed, 09 Sep 2026 03:51:55 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20260909035155777/</guid><description>Meta shipped Muse in the US — an AI agent that books travel, sends emails, fills forms, does your shopping. Not a chatbot that talks. An agent that does things.…</description></item><item><title>DeepSeek dropped back to its old prices.</title><link>https://haoyi.im/feed/20260909024938443/</link><pubDate>Wed, 09 Sep 2026 02:49:38 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20260909024938443/</guid><description>DeepSeek dropped back to its old prices. Starting 12:00 noon Beijing time, September 10 — Flash series repriced: - Cache hit: ¥0.02/M tokens (back to historical…</description></item><item><title>reasonable</title><link>https://haoyi.im/feed/20260906032027881/</link><pubDate>Sun, 06 Sep 2026 03:20:27 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20260906032027881/</guid><description>reasonable</description></item><item><title>Grok-bot’s box is one of those things that’s barely worth using</title><link>https://haoyi.im/feed/20260906025344643/</link><pubDate>Sun, 06 Sep 2026 02:53:44 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20260906025344643/</guid><description>Grok-bot’s box is one of those things that’s barely worth using, but you still don’t throw it away.It’s an 8-core / 16+128G machine, except it isn’t a real Linu…</description></item><item><title>Gemini-4 topping the charts?</title><link>https://haoyi.im/feed/20260904162254841/</link><pubDate>Fri, 04 Sep 2026 16:22:54 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20260904162254841/</guid><description>Gemini-4 topping the charts? Leaked benchmarks are making the rounds and honestly it looks legit. Google&amp;#39;s been quiet about 3.0 — rumor is they skip it entirely…</description></item><item><title>OpenAI dropped GPT-6 Astra, their first model trained on an older</title><link>https://haoyi.im/feed/20260904103459616/</link><pubDate>Fri, 04 Sep 2026 10:34:59 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20260904103459616/</guid><description>OpenAI dropped GPT-6 Astra, their first model trained on an older model&amp;#39;s output. Intelligence score matches GPT-5.6-Sol at 61. Where it shines: computer use an…</description></item><item><title>Meta Muse Spark 1.3 is out</title><link>https://haoyi.im/feed/20260903042043570/</link><pubDate>Thu, 03 Sep 2026 04:20:43 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20260903042043570/</guid><description>Meta Muse Spark 1.3 is out. Matches Claude Opus 5 on intelligence benchmarks. Long-context retrieval (MRCR, 1M window): 98.1, best score I&amp;#39;ve seen. Coding beats…</description></item><item><title>Turns out the 9router extension I wrote for pi has way more users</title><link>https://haoyi.im/feed/20260902092038302/</link><pubDate>Wed, 02 Sep 2026 09:20:38 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20260902092038302/</guid><description>Turns out the 9router extension I wrote for pi has way more users than I expected. Genuinely thought I was the only one using it.</description></item><item><title>Claude Fable 5.1 dropped today, built on Mythos 5.1</title><link>https://haoyi.im/feed/20260902063546472/</link><pubDate>Wed, 02 Sep 2026 06:35:46 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20260902063546472/</guid><description>Claude Fable 5.1 dropped today, built on Mythos 5.1. Coding scores 55.8% on Terminal-Bench, research doubled to 52.6%. Cache is 75% cheaper now, so real costs l…</description></item><item><title>Holy shit, Grok bot weekly quota just reset.</title><link>https://haoyi.im/feed/20260901165541602/</link><pubDate>Tue, 01 Sep 2026 16:55:41 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20260901165541602/</guid><description>Holy shit, Grok bot weekly quota just reset.</description></item><item><title>hy-4 and glm-5.3-flash now work through the reverse proxy</title><link>https://haoyi.im/feed/20260829001125479/</link><pubDate>Sat, 29 Aug 2026 00:11:25 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20260829001125479/</guid><description>hy-4 and glm-5.3-flash now work through the reverse proxy. workbuddy CLI picked up both too. Haven&amp;#39;t touched hy-4 yet, but glm-5.3-flash has been my daily drive…</description></item><item><title>Hunyuan Hy4 preview is live: 770B total / 49B active, 1M context</title><link>https://haoyi.im/feed/20260828091609189/</link><pubDate>Fri, 28 Aug 2026 09:16:09 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20260828091609189/</guid><description>Hunyuan Hy4 preview is live: 770B total / 49B active, 1M context. But ¥6 in / ¥18 out per M tokens — that&amp;#39;s a real jump from Hy3&amp;#39;s pricing.</description></item><item><title>grok-bot's server is worth way more than the bot on it.</title><link>https://haoyi.im/feed/20260828025005318/</link><pubDate>Fri, 28 Aug 2026 02:50:05 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20260828025005318/</guid><description>grok-bot&amp;#39;s server is worth way more than the bot on it. The bot eats through its weekly quota in no time, but at least it doesn&amp;#39;t reset the device. Server&amp;#39;s in…</description></item><item><title>model "ox-alpha" is GLM-5.3-flash.</title><link>https://haoyi.im/feed/20260827025455162/</link><pubDate>Thu, 27 Aug 2026 02:54:55 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20260827025455162/</guid><description>model &amp;#34;ox-alpha&amp;#34; is GLM-5.3-flash. First native multimodal in the GLM-5 family. MoE 320B / 18B active. Smartness 57 (~Opus 4.8). Native vision, 1M context. $0.1…</description></item><item><title>agent, server included.</title><link>https://haoyi.im/feed/20260827022031291/</link><pubDate>Thu, 27 Aug 2026 02:20:31 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20260827022031291/</guid><description>agent, server included. Yesterday Grok bot got pushed down to SuperGrok Plus. Today it dropped again to SuperGrok. So buying SuperGrok gets you Grok bot + a clo…</description></item><item><title>DeepSeek Harness is the best mobile agent setup right now</title><link>https://haoyi.im/feed/20260826105805187/</link><pubDate>Wed, 26 Aug 2026 10:58:05 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20260826105805187/</guid><description>DeepSeek Harness is the best mobile agent setup right now I&amp;#39;m probably switching from Pi to DSH. Main reason: I don&amp;#39;t want to carry a laptop around anymore. DSH…</description></item><item><title>DeepSeek hiked prices 12x (that's +1100%) after Aug 16</title><link>https://haoyi.im/feed/20260825135707075/</link><pubDate>Tue, 25 Aug 2026 13:57:07 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20260825135707075/</guid><description>DeepSeek hiked prices 12x (that&amp;#39;s +1100%) after Aug 16. This past week it cost me over ¥1,000, and ds is still mostly my fallback. I go through a lot of tokens,…</description></item><item><title>meta/muse-spark-1.2-contributor is free to use.opencode's free</title><link>https://haoyi.im/feed/20260825081905073/</link><pubDate>Tue, 25 Aug 2026 08:19:05 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20260825081905073/</guid><description>meta/muse-spark-1.2-contributor is free to use.opencode&amp;#39;s free tier also has meta/muse-spark-1.2-contributor, but that model&amp;#39;s server is flaky, so I use hy-3 as…</description></item><item><title>Opencode killed the $5 first-month promo on its Go plan</title><link>https://haoyi.im/feed/20260825043928514/</link><pubDate>Tue, 25 Aug 2026 04:39:28 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20260825043928514/</guid><description>Opencode killed the $5 first-month promo on its Go plan. First month is $10 flat now. Their cost pressure must be real — they also just dropped the free DS-V4-F…</description></item><item><title>muse-spark-1.2-contributor seems really bad at following Chinese</title><link>https://haoyi.im/feed/20260824024429319/</link><pubDate>Mon, 24 Aug 2026 02:44:29 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20260824024429319/</guid><description>muse-spark-1.2-contributor seems really bad at following Chinese prompts? Compared notes with a few people, same impression across the board. It keeps misreadin…</description></item><item><title>Marketing genius over at Cursor, honestly.</title><link>https://haoyi.im/feed/20260823021954721/</link><pubDate>Sun, 23 Aug 2026 02:19:54 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20260823021954721/</guid><description>Marketing genius over at Cursor, honestly. They&amp;#39;re running a promo: cancel your other AI subs and you get a free month of Grok Bot (the $200 plan). 🔗 anysphere.…</description></item><item><title>Whose model is this?</title><link>https://haoyi.im/feed/20260821101044555/</link><pubDate>Fri, 21 Aug 2026 10:10:44 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20260821101044555/</guid><description>Whose model is this? Some anon dev under the name &amp;#34;Stealth&amp;#34; just dropped a frontier model, Ox Alpha. Free on OpenRouter atm, and it&amp;#39;s already #1 on the usage le…</description></item><item><title>DeepSeek finally grew some eyes.</title><link>https://haoyi.im/feed/20260821093841854/</link><pubDate>Fri, 21 Aug 2026 09:38:41 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20260821093841854/</guid><description>DeepSeek finally grew some eyes. Just shipped V4-Flash-Vision-Exp, API only. Text matches V4-Flash, but the multimodal side nearly keeps up with Opus-4.8. Anoth…</description></item><item><title>For subagent work, Hy3 has been treating me well.</title><link>https://haoyi.im/feed/20260821044405035/</link><pubDate>Fri, 21 Aug 2026 04:44:05 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20260821044405035/</guid><description>For subagent work, Hy3 has been treating me well. Cheaper than V4-Flash-0731, roughly the same capability as the old DS-V4-Pro. For search, explore, execute, QA…</description></item><item><title>muse-spark-1.2-contributor is my main model now, and I don't see</title><link>https://haoyi.im/feed/20260821040439579/</link><pubDate>Fri, 21 Aug 2026 04:04:39 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20260821040439579/</guid><description>muse-spark-1.2-contributor is my main model now, and I don&amp;#39;t see that changing anytime soon. On price and capability combined, nothing else I&amp;#39;ve tried comes clo…</description></item><item><title>LMAO 😂</title><link>https://haoyi.im/feed/20260820131232456/</link><pubDate>Thu, 20 Aug 2026 13:12:32 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20260820131232456/</guid><description>LMAO 😂</description></item><item><title>opencode go now has Muse-Spark-1.2-Contributor, and it's the</title><link>https://haoyi.im/feed/20260820131102097/</link><pubDate>Thu, 20 Aug 2026 13:11:02 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20260820131102097/</guid><description>opencode go now has Muse-Spark-1.2-Contributor, and it&amp;#39;s the model with the biggest quota in the plan. One catch: your data gets sent to Meta for training. Comp…</description></item><item><title>1.7B tokens burned through</title><link>https://haoyi.im/feed/20260819155004769/</link><pubDate>Wed, 19 Aug 2026 15:50:04 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20260819155004769/</guid><description>1.7B tokens burned through My laptop hasn&amp;#39;t closed in a week. Multiple worktrees, multiple subagents running at once, 1.7B tokens burned through. I&amp;#39;ve been buil…</description></item><item><title>Command code added GPT-5.6-Sol to the GOAT plan for $70/mo</title><link>https://haoyi.im/feed/20260818031813711/</link><pubDate>Tue, 18 Aug 2026 03:18:13 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20260818031813711/</guid><description>Command code added GPT-5.6-Sol to the GOAT plan for $70/mo. Billing&amp;#39;s still at full price though — $5 in, $30 out. Openrouter has the same model at half off rig…</description></item><item><title>Qwen3.8-27B is kind of absurd.</title><link>https://haoyi.im/feed/20260818024409934/</link><pubDate>Tue, 18 Aug 2026 02:44:09 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20260818024409934/</guid><description>Qwen3.8-27B is kind of absurd. A 27B model ranks 7th globally on Artificial Analysis for agentic tasks. Runs on a single 3090 or 4090, plenty for personal use.…</description></item><item><title>OpenCode is out here trying to drag DeepSeek's prices back down</title><link>https://haoyi.im/feed/20260818024254277/</link><pubDate>Tue, 18 Aug 2026 02:42:54 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20260818024254277/</guid><description>OpenCode is out here trying to drag DeepSeek&amp;#39;s prices back down Dax says they&amp;#39;re hunting for a way to host DeepSeek at the old rates. Easier said than done. Dee…</description></item><item><title>DeepSeek jacked up their prices</title><link>https://haoyi.im/feed/20260817040120792/</link><pubDate>Mon, 17 Aug 2026 04:01:20 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20260817040120792/</guid><description>DeepSeek jacked up their prices , so I&amp;#39;m done with V4-Flash and moving to GPT-5.6-Luna. Vision&amp;#39;s a nice extra. Downside: GPT&amp;#39;s cache-hit discount isn&amp;#39;t as gener…</description></item><item><title>Tried the same prompt on the latest models: Grok-4.6</title><link>https://haoyi.im/feed/20260816152110782/</link><pubDate>Sun, 16 Aug 2026 15:21:10 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20260816152110782/</guid><description>Tried the same prompt on the latest models: Grok-4.6, DS-V4-Pro-0813, GLM-5.3, Gemini-3.7-Flash Prompt: make a baseball web game, bilingual Chinese + English GL…</description></item><item><title>Grok quietly gave everyone a weekly quota reset. Go use it.</title><link>https://haoyi.im/feed/20260816151827697/</link><pubDate>Sun, 16 Aug 2026 15:18:27 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20260816151827697/</guid><description>Grok quietly gave everyone a weekly quota reset. Go use it.</description></item><item><title>Ran Qwen3.8-27B-4bit locally on my M1 Pro 32GB</title><link>https://haoyi.im/feed/20260815044903410/</link><pubDate>Sat, 15 Aug 2026 04:49:03 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20260815044903410/</guid><description>Ran Qwen3.8-27B-4bit locally on my M1 Pro 32GB . 10 tok/s. Each sentence hangs for 2-3 seconds before output starts. Long prompt prefill takes over a minute. 12…</description></item><item><title>Qwen3.8-27B dropped open source and it's taking shots at Opus4.6.</title><link>https://haoyi.im/feed/20260814154023821/</link><pubDate>Fri, 14 Aug 2026 15:40:23 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20260814154023821/</guid><description>Qwen3.8-27B dropped open source and it&amp;#39;s taking shots at Opus4.6. Dense model, hybrid architecture, native VLM — sees images and video out of the box. The numbe…</description></item><item><title>ChatGPT's macOS app just shipped Computer History</title><link>https://haoyi.im/feed/20260814125325294/</link><pubDate>Fri, 14 Aug 2026 12:53:25 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20260814125325294/</guid><description>ChatGPT&amp;#39;s macOS app just shipped Computer History. Think Pieces.app but built in — it tracks what you click, open, and switch to, using the accessibility API in…</description></item><item><title>ZAI just dropped GLM-5.3</title><link>https://haoyi.im/feed/20260814120830119/</link><pubDate>Fri, 14 Aug 2026 12:08:30 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20260814120830119/</guid><description>ZAI just dropped GLM-5.3. Same 743B base they already had, but pure post-training got coding benchmarks up 50%. 745B total, 44B active. Trained on Huawei Ascend…</description></item><item><title>DeepSeek hiked prices hard — nearly 5x.</title><link>https://haoyi.im/feed/20260814104739334/</link><pubDate>Fri, 14 Aug 2026 10:47:39 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20260814104739334/</guid><description>DeepSeek hiked prices hard — nearly 5x. V4-Pro output at peak: ¥6 → ¥27. A 4.5x jump (350%). Off-peak: ¥13.5. Still more than double what it was before.</description></item><item><title>Google just dropped Gemini 3.7 Flash</title><link>https://haoyi.im/feed/20260813175058652/</link><pubDate>Thu, 13 Aug 2026 17:50:58 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20260813175058652/</guid><description>Google just dropped Gemini 3.7 Flash. 65.3% on software engineering — GPT-5.6 Luna is at 67%, so it&amp;#39;s basically caught up. Price slashed to $0.75/$3.75, half of…</description></item><item><title>DeepSeek just open-sourced their agent harness</title><link>https://haoyi.im/feed/20260813134146651/</link><pubDate>Thu, 13 Aug 2026 13:41:46 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20260813134146651/</guid><description>DeepSeek just open-sourced their agent harness. Everything&amp;#39;s a plugin — model adapters, tool registration, session logging, the agent loop itself. All swappable…</description></item><item><title>DeepSeek just dropped V4-Pro-0813 with zero announcement</title><link>https://haoyi.im/feed/20260813081244953/</link><pubDate>Thu, 13 Aug 2026 08:12:44 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20260813081244953/</guid><description>DeepSeek just dropped V4-Pro-0813 with zero announcement. No blog post, no tweet. Just quietly flipped the API version. Architecture&amp;#39;s unchanged — still 1.6T Mo…</description></item><item><title>Grok 4.6 is out. 500K context, four reasoning modes (defaults to</title><link>https://haoyi.im/feed/20260812155007932/</link><pubDate>Wed, 12 Aug 2026 15:50:07 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20260812155007932/</guid><description>Grok 4.6 is out. 500K context, four reasoning modes (defaults to high), knowledge goes up to Feb 2026. Input $2/M, output $6/M. Both double if you hit 200K+ con…</description></item><item><title>WeChat dropped its own LLM. Called WeLM.</title><link>https://haoyi.im/feed/20260812153450191/</link><pubDate>Wed, 12 Aug 2026 15:34:50 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20260812153450191/</guid><description>WeChat dropped its own LLM. Called WeLM. Both versions are MoE: - WeLM-80B: 80B total params, 3B active. Already live in WeChat&amp;#39;s Xiaowei assistant - WeLM-617B:…</description></item><item><title>xAI dropped Grok Bot. The pitch: persistent cloud VMs.</title><link>https://haoyi.im/feed/20260812152308382/</link><pubDate>Wed, 12 Aug 2026 15:23:08 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20260812152308382/</guid><description>xAI dropped Grok Bot. The pitch: persistent cloud VMs. Each bot gets its own browser, filesystem, and terminal — keeps running after you kill your laptop. Cloud…</description></item><item><title>I Built local-rag</title><link>https://haoyi.im/feed/20260812094346561/</link><pubDate>Wed, 12 Aug 2026 09:43:46 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20260812094346561/</guid><description>I Built local-rag Former company&amp;#39;s knowledge base went down because Google changed the NotebookLM API. The local digital worker lost connection overnight.I Buil…</description></item><item><title>I've designed and used quite a few eval harnesses by now</title><link>https://haoyi.im/feed/20260812042929389/</link><pubDate>Wed, 12 Aug 2026 04:29:29 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20260812042929389/</guid><description>I&amp;#39;ve designed and used quite a few eval harnesses by now. Here&amp;#39;s what I keep coming back to: stronger models need less scaffolding. Pile on more constraints and…</description></item><item><title>Google sent me the Workspace free AI trial email today</title><link>https://haoyi.im/feed/20260812030625844/</link><pubDate>Wed, 12 Aug 2026 03:06:25 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20260812030625844/</guid><description>Google sent me the Workspace free AI trial email today. Their main pitch is Nano Banana Pro, which honestly isn&amp;#39;t nothing. Free users can also grab it in Flow —…</description></item><item><title>Cloudflare built Kitesurf: a browser for AI agents that runs on</title><link>https://haoyi.im/feed/20260811130110876/</link><pubDate>Tue, 11 Aug 2026 13:01:10 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20260811130110876/</guid><description>Cloudflare built Kitesurf: a browser for AI agents that runs on Workers, not local Chromium. 3-7x less CPU and memory for DOM extraction. Agents don&amp;#39;t need tabs…</description></item><item><title>Prime Intellect open-sourced Prime Agent — a new agent harness</title><link>https://haoyi.im/feed/20260811110909806/</link><pubDate>Tue, 11 Aug 2026 11:09:09 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20260811110909806/</guid><description>Prime Intellect open-sourced Prime Agent — a new agent harness built on pi. Everything runs through code. Subtasks decompose recursively into concurrent subagen…</description></item><item><title>Dev insight: real value of open source isn't forking — it's</title><link>https://haoyi.im/feed/20260811061126306/</link><pubDate>Tue, 11 Aug 2026 06:11:26 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20260811061126306/</guid><description>Dev insight: real value of open source isn&amp;#39;t forking — it&amp;#39;s ripping off design ideas. My &amp;#34;Fatiao&amp;#34; project had a built-in mini Office editor. To edit anything yo…</description></item><item><title>Cloudflare OS just went open source</title><link>https://haoyi.im/feed/20260810171123364/</link><pubDate>Mon, 10 Aug 2026 17:11:23 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20260810171123364/</guid><description>Cloudflare OS just went open source. AI workspace for your whole company — code, apps, decks. Every user runs in their own isolated sandbox. Apache 2.0. Fork it…</description></item><item><title>Cursor brand probably gets axed</title><link>https://haoyi.im/feed/20260810154335368/</link><pubDate>Mon, 10 Aug 2026 15:43:35 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20260810154335368/</guid><description>Cursor brand probably gets axed. SpaceX bought it, xAI likely takes over the name. But Cursor was my first vibe coding experience. Thought it was autocomplete s…</description></item><item><title>Google killed Gemini 3.5 Pro.</title><link>https://haoyi.im/feed/20260810150819384/</link><pubDate>Mon, 10 Aug 2026 15:08:19 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20260810150819384/</guid><description>Google killed Gemini 3.5 Pro. Semi Analysis says they&amp;#39;ve dropped it entirely skipping straight to Gemini 4 to chase frontier models. Probably won&amp;#39;t ship till ye…</description></item><item><title>Muse Spark 1.2 Contributor hits hard for the price</title><link>https://haoyi.im/feed/20260807022445063/</link><pubDate>Fri, 07 Aug 2026 02:24:45 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20260807022445063/</guid><description>Muse Spark 1.2 Contributor hits hard for the price. Cheaper than DS-v4-flash, performance close to GPT-5.5. Trade-off: Meta trains on your data.</description></item><item><title>word is DeepSeek might be jacking up prices soon</title><link>https://haoyi.im/feed/20260806143429057/</link><pubDate>Thu, 06 Aug 2026 14:34:29 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20260806143429057/</guid><description>word is DeepSeek might be jacking up prices soon. their API already crashed once from the surge. guess they&amp;#39;re short on GPUs raising prices to deal with the flo…</description></item><item><title>commandcode just launched a $10 plan.</title><link>https://haoyi.im/feed/20260806135229006/</link><pubDate>Thu, 06 Aug 2026 13:52:29 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20260806135229006/</guid><description>commandcode just launched a $10 plan. $70 in credits, open source + closed source models included. Honestly the $1 plan is still better bang for your buck this…</description></item><item><title>9router just got auto vision + audio model routing.</title><link>https://haoyi.im/feed/20260806033501649/</link><pubDate>Thu, 06 Aug 2026 03:35:01 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20260806033501649/</guid><description>9router just got auto vision + audio model routing. Configure it once. Any non-multimodal model called through 9router now detects vision/audio tasks and silent…</description></item><item><title>Meta just shipped Muse Code terminal coding agent, Muse Spark 1.2</title><link>https://haoyi.im/feed/20260806024118067/</link><pubDate>Thu, 06 Aug 2026 02:41:18 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20260806024118067/</guid><description>Meta just shipped Muse Code terminal coding agent, Muse Spark 1.2 under the hood. $1.25/M in, $4.25/M out. Async background jobs, event logs, /plan and /grill b…</description></item><item><title>Another AI browser.</title><link>https://haoyi.im/feed/20260806021546547/</link><pubDate>Thu, 06 Aug 2026 02:15:46 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20260806021546547/</guid><description>Another AI browser. Hark launched Handoff a browser agent that controls real sites like LinkedIn through mouse and keyboard. No API needed. Closed beta now, pub…</description></item><item><title>ds-v4-flash got a pair of eyes and it actually works.Set up a</title><link>https://haoyi.im/feed/20260805071346929/</link><pubDate>Wed, 05 Aug 2026 07:13:46 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20260805071346929/</guid><description>ds-v4-flash got a pair of eyes and it actually works.Set up a visual model group on 9router all free models. Defined the subagent&amp;#39;s role, responsibilities, and…</description></item><item><title>Cloudflare just shipped Wallets, programmable wallets for AI</title><link>https://haoyi.im/feed/20260805015922893/</link><pubDate>Wed, 05 Aug 2026 01:59:22 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20260805015922893/</guid><description>Cloudflare just shipped Wallets, programmable wallets for AI agents.Humans fund an Account Wallet, set spending caps, then Virtual Wallets let agents autonomous…</description></item><item><title>Oracle's going after existing free instances now</title><link>https://haoyi.im/feed/20260805014317881/</link><pubDate>Wed, 05 Aug 2026 01:43:17 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20260805014317881/</guid><description>Oracle&amp;#39;s going after existing free instances now. They were only nerfing new VMs before, so I left my 4+16G box untouched thinking it&amp;#39;d survive. Nope.</description></item><item><title>CommandCode $1 plan got GPT-5.6-Luna now too, but the credit cap</title><link>https://haoyi.im/feed/20260802153117212/</link><pubDate>Sun, 02 Aug 2026 15:31:17 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20260802153117212/</guid><description>CommandCode $1 plan got GPT-5.6-Luna now too, but the credit cap is just $20, DeepSeek-V4-Pro gives you $40 at the same tier.</description></item><item><title>macbook gets a dynamic island</title><link>https://haoyi.im/feed/20260802143129152/</link><pubDate>Sun, 02 Aug 2026 14:31:29 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20260802143129152/</guid><description>macbook gets a dynamic island stress tested DeepSeek-v4-flash-0731 full-stack coding — built a macOS Dynamic Island app. media controls, file stash, system info…</description></item><item><title>DeepSeek-V4-Flash just dropped. It's eating GLM-5.2 alive.</title><link>https://haoyi.im/feed/20260731081713210/</link><pubDate>Fri, 31 Jul 2026 08:17:13 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20260731081713210/</guid><description>DeepSeek-V4-Flash just dropped. It&amp;#39;s eating GLM-5.2 alive. Same params, no new architecture — just post-training. Per the official writeup, it now tops multiple…</description></item><item><title>GPT-5.6 Luna just dropped 80% in price.</title><link>https://haoyi.im/feed/20260731075601930/</link><pubDate>Fri, 31 Jul 2026 07:56:01 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20260731075601930/</guid><description>GPT-5.6 Luna just dropped 80% in price. Ran the numbers against DeepSeek-v4-Pro. Luna wins for what I need, so I&amp;#39;m moving my subagent slot over to it. Setup rig…</description></item><item><title>trying to build a paid product</title><link>https://haoyi.im/feed/20260730095757792/</link><pubDate>Thu, 30 Jul 2026 09:57:57 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20260730095757792/</guid><description>trying to build a paid product. just needed a multimodal model that&amp;#39;s actually good AND doesn&amp;#39;t bleed me dry. tried everything on the shelf, only Xiaomi&amp;#39;s MiMo-…</description></item><item><title>Lately I’ve been turning this old “key sprite” style project I</title><link>https://haoyi.im/feed/20260730025847823/</link><pubDate>Thu, 30 Jul 2026 02:58:47 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20260730025847823/</guid><description>Lately I’ve been turning this old “key sprite” style project I wrote six years ago into an agent-driven setup.These kinds of projects are a natural fit for agen…</description></item><item><title>Grok now ships with its own server</title><link>https://haoyi.im/feed/20260729155742026/</link><pubDate>Wed, 29 Jul 2026 15:57:42 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20260729155742026/</guid><description>Grok now ships with its own server. Every account gets a cloud box. You can build projects on it, phone to web and back. Tried it asked Grok to make a pixel bas…</description></item><item><title>Qwen 3.8 is free right now and it's wild.</title><link>https://haoyi.im/feed/20260729144759836/</link><pubDate>Wed, 29 Jul 2026 14:47:59 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20260729144759836/</guid><description>Qwen 3.8 is free right now and it&amp;#39;s wild. Alibaba launched a Growth Plan — use Qwen3.8-Max-Preview for your actual projects, send them wins or disaster stories,…</description></item><item><title>Finally cracked how SuperGrok billing works.</title><link>https://haoyi.im/feed/20260729141704826/</link><pubDate>Wed, 29 Jul 2026 14:17:04 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20260729141704826/</guid><description>Finally cracked how SuperGrok billing works. Your weekly quota is fixed and shared across XAI, Build, and Chat. The difference comes down to multiplier rates an…</description></item><item><title>I'm on the $1.5/mo plan for 50 IPs.</title><link>https://haoyi.im/feed/20260729035353633/</link><pubDate>Wed, 29 Jul 2026 03:53:53 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20260729035353633/</guid><description>I&amp;#39;m on the $1.5/mo plan for 50 IPs. pointed them at opencode running deepseek-v4-flash as the Hermes. been using it for real and it&amp;#39;s actually solid, everyday s…</description></item><item><title>Vivix just shipped A1 and it's got something</title><link>https://haoyi.im/feed/20260729023207964/</link><pubDate>Wed, 29 Jul 2026 02:32:07 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20260729023207964/</guid><description>Vivix just shipped A1 and it&amp;#39;s got something. Full-duplex, full-body digital human. Siri-style turn-taking is half-duplex: you talk, then it talks. A1 runs full…</description></item><item><title>grok-4.6 in 2 weeks, 4.7 in 4</title><link>https://haoyi.im/feed/20260726022738258/</link><pubDate>Sun, 26 Jul 2026 02:27:38 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20260726022738258/</guid><description>grok-4.6 in 2 weeks, 4.7 in 4. Musk says they&amp;#39;re gunning for Kimi K3.</description></item><item><title>Claude Opus 5 is kinda insane</title><link>https://haoyi.im/feed/20260725052701464/</link><pubDate>Sat, 25 Jul 2026 05:27:01 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20260725052701464/</guid><description>Claude Opus 5 is kinda insane Near-Fable-5 level intelligence at half the price. 43.3% on Frontier-Bench — more than 2x Opus 4.8, and cheaper. On CursorBench (h…</description></item><item><title>Unlimited grok-4.5 + deepseek-v4-flashLately I've seen tons of</title><link>https://haoyi.im/feed/20260725015643772/</link><pubDate>Sat, 25 Jul 2026 01:56:43 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20260725015643772/</guid><description>Unlimited grok-4.5 + deepseek-v4-flashLately I&amp;#39;ve seen tons of relay services offering grok-4.5 as low as 0.1 yuan per 1M tokens. Dug in — the trick is dead sim…</description></item><item><title>Gemini 3.6 Flash is here</title><link>https://haoyi.im/feed/20260722025911281/</link><pubDate>Wed, 22 Jul 2026 02:59:11 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20260722025911281/</guid><description>Gemini 3.6 Flash is here. 1M context, 65K output, takes image/audio/video. Google says it uses fewer tokens than 3.5 Flash, writes better code, and reasons bett…</description></item><item><title>Databricks just ran a real-world coding test (not some benchmark</title><link>https://haoyi.im/feed/20260721043406279/</link><pubDate>Tue, 21 Jul 2026 04:34:06 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20260721043406279/</guid><description>Databricks just ran a real-world coding test (not some benchmark suite) pitting different agent tools against various models. Chart’s simple: higher on the Y =…</description></item><item><title>Just found a legit 0-cost local solution for Agent search &amp;</title><link>https://haoyi.im/feed/20260720124832746/</link><pubDate>Mon, 20 Jul 2026 12:48:32 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20260720124832746/</guid><description>Just found a legit 0-cost local solution for Agent search &amp;amp; crawling I’ve been playing around with this open-source project called wigolo lately. It lets you do…</description></item><item><title>Qwen-3.8 vs Kimi-K3</title><link>https://haoyi.im/feed/20260720080009571/</link><pubDate>Mon, 20 Jul 2026 08:00:09 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20260720080009571/</guid><description>Qwen-3.8 vs Kimi-K3 Prompt: &amp;#34;SVG animation of a real tennis rally, two players going at it, physically accurate, follows real-world logic.&amp;#34; This one took me 2 h…</description></item><item><title>Alibaba dropped Qwen3.8. Only Fable 5 is ahead of it right now.</title><link>https://haoyi.im/feed/20260719082752019/</link><pubDate>Sun, 19 Jul 2026 08:27:52 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20260719082752019/</guid><description>Alibaba dropped Qwen3.8. Only Fable 5 is ahead of it right now. 2.4T params. Open source. Biggest Qwen model they&amp;#39;ve shipped. Qwen3.8-Max-Preview is already liv…</description></item><item><title>had leftover grok-4.5 quota go to waste</title><link>https://haoyi.im/feed/20260718140617807/</link><pubDate>Sat, 18 Jul 2026 14:06:17 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20260718140617807/</guid><description>had leftover grok-4.5 quota go to waste weekly reset hit, last night pi accidentally went through the xAI endpoint. burned 33% in 2 hours. switched pi to grok b…</description></item><item><title>let an agent drive your browser. it should be simple. it's not.</title><link>https://haoyi.im/feed/20260718021930855/</link><pubDate>Sat, 18 Jul 2026 02:19:30 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20260718021930855/</guid><description>let an agent drive your browser. it should be simple. it&amp;#39;s not. most agent tools open a fresh browser or a separate profile. want to reuse your own? they copy y…</description></item><item><title>A parent in china dropped 300K rmb on an AI certificate</title><link>https://haoyi.im/feed/20260717082431382/</link><pubDate>Fri, 17 Jul 2026 08:24:31 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20260717082431382/</guid><description>A parent in china dropped 300K rmb on an AI certificate lmao a parent in china dropped 300K rmb on an AI certificate. it&amp;#39;s a completely worthless piece of paper…</description></item><item><title>No Mac. No code. App Store anyway.</title><link>https://haoyi.im/feed/20260717081554323/</link><pubDate>Fri, 17 Jul 2026 08:15:54 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20260717081554323/</guid><description>No Mac. No code. App Store anyway. The old way: buy a Mac, learn Xcode, pay Apple $99/year, and drown in certs and provisioning profiles. Baidu Miaoda skips all…</description></item><item><title>Gemini 3.5 Pro pushed back again, didn't hit internal perf</title><link>https://haoyi.im/feed/20260717074442541/</link><pubDate>Fri, 17 Jul 2026 07:44:42 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20260717074442541/</guid><description>Gemini 3.5 Pro pushed back again, didn&amp;#39;t hit internal perf targets. Fable 5 and GPT-5.6 are just too far ahead. Next Gemini&amp;#39;s gonna have a rough time catching u…</description></item><item><title>Kimi-K3 is wild.</title><link>https://haoyi.im/feed/20260717071631626/</link><pubDate>Fri, 17 Jul 2026 07:16:31 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20260717071631626/</guid><description>Kimi-K3 is wild. 2.8T params, 896 MoE experts, only 16 active at a time. Native 1M context + multimodal. AA index 57, beats Opus 4.8 and GPT-5.5. #1 on Arena We…</description></item><item><title>kimi web ( kimi.com ) just added the k3</title><link>https://haoyi.im/feed/20260716141206187/</link><pubDate>Thu, 16 Jul 2026 14:12:06 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20260716141206187/</guid><description>kimi web ( kimi.com ) just added the k3</description></item><item><title>PrismML just put a 27B multimodal model on a phone</title><link>https://haoyi.im/feed/20260716135420534/</link><pubDate>Thu, 16 Jul 2026 13:54:20 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20260716135420534/</guid><description>PrismML just put a 27B multimodal model on a phone. First time anyone&amp;#39;s done it. Bonsai 27B, built on Qwen3.6 27B with end-to-end ultra-low quantization. Ternar…</description></item><item><title>subscription-based skills become a real revenue model.</title><link>https://haoyi.im/feed/20260716063844460/</link><pubDate>Thu, 16 Jul 2026 06:38:44 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20260716063844460/</guid><description>subscription-based skills become a real revenue model. In my opinion, 2026 is year zero for general-purpose agents. once everyone&amp;#39;s running one, subscription-ba…</description></item><item><title>Mira Murati just dropped her first big model.</title><link>https://haoyi.im/feed/20260716034002988/</link><pubDate>Thu, 16 Jul 2026 03:40:02 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20260716034002988/</guid><description>Mira Murati just dropped her first big model. Her team Thinking Machines released Inkling. Native multimodal, 975B params, MoE, 41B active during inference. Met…</description></item><item><title>Everyone's bearish on MiniMax because coding slipped.</title><link>https://haoyi.im/feed/20260716033143722/</link><pubDate>Thu, 16 Jul 2026 03:31:43 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20260716033143722/</guid><description>Everyone&amp;#39;s bearish on MiniMax because coding slipped. But I&amp;#39;ve been running a general agent on vision stuff lately with M3, and it&amp;#39;s surprisingly good. Other mu…</description></item><item><title>xAI open sourced the Grok Build CLI and reset the weekly quota</title><link>https://haoyi.im/feed/20260716032552572/</link><pubDate>Thu, 16 Jul 2026 03:25:52 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20260716032552572/</guid><description>xAI open sourced the Grok Build CLI and reset the weekly quota After getting dragged for silently uploading repos, xAI open sourced the Grok Build CLI and reset…</description></item><item><title>My friend just dropped real money on AI certs that are basically</title><link>https://haoyi.im/feed/20260715063512353/</link><pubDate>Wed, 15 Jul 2026 06:35:12 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20260715063512353/</guid><description>My friend just dropped real money on AI certs that are basically fancy PDFs. The AI cert space is the wild west. No single recognized body, no industry standard…</description></item><item><title>Two weeks with SuperGrok and ChatGPT Plus</title><link>https://haoyi.im/feed/20260715044455218/</link><pubDate>Wed, 15 Jul 2026 04:44:55 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20260715044455218/</guid><description>Two weeks with SuperGrok and ChatGPT Plus. Weekly usage is about the same. Grok Build pulls from SuperGrok&amp;#39;s ~$40/wk credit pool. grok-4.5, roughly 50 rounds. C…</description></item><item><title>There's another open source coding agent called Kimchi.</title><link>https://haoyi.im/feed/20260715034246047/</link><pubDate>Wed, 15 Jul 2026 03:42:46 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20260715034246047/</guid><description>There&amp;#39;s another open source coding agent called Kimchi. It&amp;#39;s built on Pi, has a /ferment mode that auto-routes tasks to different models (very similar to how I…</description></item><item><title>Don't post Grok content on Meta platforms</title><link>https://haoyi.im/feed/20260715015457075/</link><pubDate>Wed, 15 Jul 2026 01:54:57 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20260715015457075/</guid><description>Don&amp;#39;t post Grok content on Meta platforms Posted Grok stuff for a few days. Threads emailed me today asking for a selfie. Uploaded it. Permanently banned. Threa…</description></item><item><title>Get comfortable with agents.</title><link>https://haoyi.im/feed/20260714074347185/</link><pubDate>Tue, 14 Jul 2026 07:43:47 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20260714074347185/</guid><description>Get comfortable with agents. The old way of doing things doesn&amp;#39;t really apply in 2026. Agents are basically an operating system that understands plain language…</description></item><item><title>📞 xAI launched a voice agent and they're giving out free phone</title><link>https://haoyi.im/feed/20260713043812160/</link><pubDate>Mon, 13 Jul 2026 04:38:12 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20260713043812160/</guid><description>📞 xAI launched a voice agent and they&amp;#39;re giving out free phone numbers Describe what the agent should say in plain language, drop in your product docs, done. Ta…</description></item><item><title>xAI's Grok Build CLI uploads your entire repo.</title><link>https://haoyi.im/feed/20260713043213892/</link><pubDate>Mon, 13 Jul 2026 04:32:13 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20260713043213892/</guid><description>xAI&amp;#39;s Grok Build CLI uploads your entire repo. Someone wire-captured Grok Build CLI&amp;#39;s harness and it&amp;#39;s ugly. All your files. Full git history. Straight to xAI&amp;#39;s…</description></item><item><title>don't use xAI OAuth for Grok-4.5</title><link>https://haoyi.im/feed/20260712030534512/</link><pubDate>Sun, 12 Jul 2026 03:05:34 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20260712030534512/</guid><description>don&amp;#39;t use xAI OAuth for Grok-4.5 supergrok via grok cli has a lower rate multiplier and comes with $200/mo in credits</description></item><item><title>Grok 4.5 is legit. On par with GPT-5.5 honestly.</title><link>https://haoyi.im/feed/20260712024823244/</link><pubDate>Sun, 12 Jul 2026 02:48:23 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20260712024823244/</guid><description>Grok 4.5 is legit. On par with GPT-5.5 honestly. The pricing model though. SuperGrok&amp;#39;s new weekly cap is rough. Tested it myself: you get about 8M tokens per we…</description></item><item><title>$30 a month for unlimited grok-4.5?</title><link>https://haoyi.im/feed/20260711032625371/</link><pubDate>Sat, 11 Jul 2026 03:26:25 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20260711032625371/</guid><description>$30 a month for unlimited grok-4.5? India grok deal is back. 3 months Heavy, $100. I&amp;#39;ve heard Heavy has unlimited usage. Haven&amp;#39;t tried it myself though, so I ca…</description></item><item><title>Grok has weekly limits now</title><link>https://haoyi.im/feed/20260711025222261/</link><pubDate>Sat, 11 Jul 2026 02:52:22 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20260711025222261/</guid><description>Grok has weekly limits now. No more going wild with it. At least it&amp;#39;s not a five hour cap. Last month I burned through the entire $150 monthly credit in three d…</description></item><item><title>OpenAI dropped a ChatGPT Work app.</title><link>https://haoyi.im/feed/20260710032327292/</link><pubDate>Fri, 10 Jul 2026 03:23:27 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20260710032327292/</guid><description>OpenAI dropped a ChatGPT Work app. Turns out they finally realized calling it &amp;#34;Codex&amp;#34; made everyone assume it was for developers. ChatGPT Work is literally just…</description></item><item><title>OpenAI is shutting down Atlas, its AI browser, less than 9 months</title><link>https://haoyi.im/feed/20260710024517760/</link><pubDate>Fri, 10 Jul 2026 02:45:17 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20260710024517760/</guid><description>OpenAI is shutting down Atlas, its AI browser, less than 9 months after launch. Atlas dropped in October 2025, meant to replace Chrome. ChatGPT sidebar, agent m…</description></item><item><title>SpaceXAI dropped Grok 4.5 and it's actually good enough to pay</title><link>https://haoyi.im/feed/20260709040525243/</link><pubDate>Thu, 09 Jul 2026 04:05:25 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20260709040525243/</guid><description>SpaceXAI dropped Grok 4.5 and it&amp;#39;s actually good enough to pay attention to now. Coding and engineering: hangs with GPT 5.5 and Claude Opus 4.8. SWE Bench Pro h…</description></item><item><title>Tencent's Hunyuan Hy3 is out. Best agent capability at its size.</title><link>https://haoyi.im/feed/20260708034258825/</link><pubDate>Wed, 08 Jul 2026 03:42:58 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20260708034258825/</guid><description>Tencent&amp;#39;s Hunyuan Hy3 is out. Best agent capability at its size. 295B total params, activates only 21B per inference. Simple queries answer fast. Complex ones t…</description></item><item><title>Grok + voice cloning. A fake account made me $3k.</title><link>https://haoyi.im/feed/20260707114832028/</link><pubDate>Tue, 07 Jul 2026 11:48:32 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20260707114832028/</guid><description>Grok + voice cloning. A fake account made me $3k. I never use up my Grok Super video credits. They rot every month. So I started a thirst trap page. Grok handle…</description></item><item><title>turns out AI agents can be your marketing team</title><link>https://haoyi.im/feed/20260705144311877/</link><pubDate>Sun, 05 Jul 2026 14:43:11 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20260705144311877/</guid><description>turns out AI agents can be your marketing team marketing-skills is an open source skills pack. drop it into Claude Code, Cursor, or Hermes and it gives them rea…</description></item><item><title>Render your prompts as PNGs. Your API bill drops 60-70%.</title><link>https://haoyi.im/feed/20260705140719121/</link><pubDate>Sun, 05 Jul 2026 14:07:19 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20260705140719121/</guid><description>Render your prompts as PNGs. Your API bill drops 60-70%. Found this today. Text costs by the token, images cost by the pixel. Same 48K characters: 25,000 tokens…</description></item><item><title>Nvidia now has a free API for GLM-5.2</title><link>https://haoyi.im/feed/20260704091429616/</link><pubDate>Sat, 04 Jul 2026 09:14:29 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20260704091429616/</guid><description>Nvidia now has a free API for GLM-5.2 GLM-5.2 is one of the best open-source coding models right now. No need to buy Z.ai&amp;#39;s Coding Plan. Just use Nvidia&amp;#39;s free…</description></item><item><title>Google made Gemini 3.5 Flash free.</title><link>https://haoyi.im/feed/20260704034207072/</link><pubDate>Sat, 04 Jul 2026 03:42:07 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20260704034207072/</guid><description>Google made Gemini 3.5 Flash free. No credit card, no nothing. Head to aistudio.google.com, create an API key, and plug it into any client with Gemini provider…</description></item><item><title>i turned my subtitle tool into an agent skill</title><link>https://haoyi.im/feed/20260703140440222/</link><pubDate>Fri, 03 Jul 2026 14:04:40 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20260703140440222/</guid><description>i turned my subtitle tool into an agent skill built a subtitle recognition saas last year. openclaw dropped and it was dead before launch. a lot of people asked…</description></item><item><title>There's a Chinese Codex alternative getting some heat lately.</title><link>https://haoyi.im/feed/20260703083044018/</link><pubDate>Fri, 03 Jul 2026 08:30:44 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20260703083044018/</guid><description>There&amp;#39;s a Chinese Codex alternative getting some heat lately. Tencent&amp;#39;s WorkBuddy. Non-technical roles are buzzing about it. I tried it: basically Codex minus t…</description></item><item><title>grok is actually really good</title><link>https://haoyi.im/feed/20260703074348890/</link><pubDate>Fri, 03 Jul 2026 07:43:48 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20260703074348890/</guid><description>grok is actually really good been using it for a while. no browser access and no knowledge base, but it replaces most of what i used manus for. built-in sandbox…</description></item><item><title>Free GLM-5.2</title><link>https://haoyi.im/feed/20260703050015036/</link><pubDate>Fri, 03 Jul 2026 05:00:15 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20260703050015036/</guid><description>Free GLM-5.2 Tencent&amp;#39;s CodeBuddy has a free tier with GLM-5.2. Route it through 9router&amp;#39;s reverse proxy and you&amp;#39;re good. 500 credits a month, no clue how many t…</description></item><item><title>OpenClaw shipped a mobile app.</title><link>https://haoyi.im/feed/20260703031251676/</link><pubDate>Fri, 03 Jul 2026 03:12:51 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20260703031251676/</guid><description>OpenClaw shipped a mobile app. Local pairing: same WiFi, /pair qr on the host, scan the QR code on your phone. Clean. Remote pairing needs Tailscale running on…</description></item><item><title>Cline is selling access to 6 open-source models for $9.99/month</title><link>https://haoyi.im/feed/20260630114421732/</link><pubDate>Tue, 30 Jun 2026 11:44:21 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20260630114421732/</guid><description>Cline is selling access to 6 open-source models for $9.99/month now. GLM-5.2, DeepSeek, Kimi, MiniMax, Mimo, Qwen. They&amp;#39;re calling it ClinePass. First month was…</description></item><item><title>deepseek just 2x'd their prices during peak hours.</title><link>https://haoyi.im/feed/20260630104829218/</link><pubDate>Tue, 30 Jun 2026 10:48:29 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20260630104829218/</guid><description>deepseek just 2x&amp;#39;d their prices during peak hours. email says peak is 9-12 and 2-6 china time. so... the whole day, basically, minus lunch. what i actually want…</description></item><item><title>So Owl Alpha, that anonymous model that's been eating tokens on</title><link>https://haoyi.im/feed/20260630104628970/</link><pubDate>Tue, 30 Jun 2026 10:46:28 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20260630104628970/</guid><description>So Owl Alpha, that anonymous model that&amp;#39;s been eating tokens on OpenRouter for two months, is actually Meituan&amp;#39;s LongCat-2.0-Preview. 10.1 trillion tokens a mon…</description></item><item><title>Someone sent me 9router for the Commandcode reverse proxy problem</title><link>https://haoyi.im/feed/20260628143436989/</link><pubDate>Sun, 28 Jun 2026 14:34:36 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20260628143436989/</guid><description>Someone sent me 9router for the Commandcode reverse proxy problem. It&amp;#39;s like cliproxyapi but compresses git diff, grep outputs before sending them. Author says…</description></item><item><title>Been hunting for a stable way to run GLM 5.2</title><link>https://haoyi.im/feed/20260628025627305/</link><pubDate>Sun, 28 Jun 2026 02:56:27 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20260628025627305/</guid><description>Been hunting for a stable way to run GLM 5.2. Most options are inflated. These two actually aren&amp;#39;t. Both are token plans. Command Code Go is $1/mo sub for $10 c…</description></item><item><title>Tencent made an email address for AI agents</title><link>https://haoyi.im/feed/20260627124940535/</link><pubDate>Sat, 27 Jun 2026 12:49:40 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20260627124940535/</guid><description>Tencent made an email address for AI agents Not the &amp;#34;AI reads your email&amp;#34; kind. The AI gets its own independent inbox. You know, an actual mailbox. xxx@agent.qq…</description></item><item><title>You've got 12 hours to grab free GLM 5.2 credits on GMI Cloud.</title><link>https://haoyi.im/feed/20260627090429712/</link><pubDate>Sat, 27 Jun 2026 09:04:29 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20260627090429712/</guid><description>You&amp;#39;ve got 12 hours to grab free GLM 5.2 credits on GMI Cloud. Right now it&amp;#39;s the strongest open-source coding model. Beats GPT-5.5 on benchmarks, sits just beh…</description></item><item><title>Refactored my workflow again.</title><link>https://haoyi.im/feed/20260627084501774/</link><pubDate>Sat, 27 Jun 2026 08:45:01 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20260627084501774/</guid><description>Refactored my workflow again. The old Pi workflow was heavily coupled — tons of hooks and scripts trying to enforce a fixed dev SOP. One wrong step and everythi…</description></item><item><title>SoundWise.ai. No signup needed, no payment needed.</title><link>https://haoyi.im/feed/20260627032010730/</link><pubDate>Sat, 27 Jun 2026 03:20:10 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20260627032010730/</guid><description>SoundWise.ai. No signup needed, no payment needed. Runs OpenAI Whisper right inside your browser. Nothing gets uploaded. Interview recordings, meeting notes, an…</description></item><item><title>Make Roblox games with your mouth 🤯</title><link>https://haoyi.im/feed/20260627024226814/</link><pubDate>Sat, 27 Jun 2026 02:42:26 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20260627024226814/</guid><description>Make Roblox games with your mouth 🤯 Been obsessed with Lemonade.gg — it&amp;#39;s an AI agent built for Roblox. Type &amp;#34;make me a tactical RPG&amp;#34; on the web, and it generat…</description></item><item><title>Anthropic has released Opus 4.7—the strongest publicly available</title><link>https://haoyi.im/feed/20260625060939498/</link><pubDate>Thu, 25 Jun 2026 06:09:39 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20260625060939498/</guid><description>Anthropic has released Opus 4.7—the strongest publicly available Claude model so far. There’s an even more powerful o... Anthropic has released Opus 4.7—the str…</description></item><item><title>Basjoo Open Source: 24/7 AI Customer Support System</title><link>https://haoyi.im/feed/20260625060939495/</link><pubDate>Thu, 25 Jun 2026 06:09:39 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20260625060939495/</guid><description>Basjoo Open Source: 24/7 AI Customer Support System After 6 months of weekend coding, my first agent project is now fully open source and free to use.</description></item><item><title>Canva’s new Magic Layers feature is really impressive</title><link>https://haoyi.im/feed/20260625060939492/</link><pubDate>Thu, 25 Jun 2026 06:09:39 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20260625060939492/</guid><description>Canva’s new Magic Layers feature is really impressive. It can break an image into editable layers while preserving th... Canva’s new Magic Layers feature is rea…</description></item><item><title>Chrome just rolled out a new feature called Skills, which turns</title><link>https://haoyi.im/feed/20260625060939502/</link><pubDate>Thu, 25 Jun 2026 06:09:39 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20260625060939502/</guid><description>Chrome just rolled out a new feature called Skills, which turns frequently used prompts into “shortcuts.”</description></item><item><title>Eddie AI has introduced Night Shift: after you finish shooting</title><link>https://haoyi.im/feed/20260625060939503/</link><pubDate>Thu, 25 Jun 2026 06:09:39 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20260625060939503/</guid><description>Eddie AI has introduced Night Shift: after you finish shooting, just hand over the footage, get some sleep, and wake ... Eddie AI has introduced Night Shift: af…</description></item><item><title>Gemini has released a Mac desktop app, going head to head with</title><link>https://haoyi.im/feed/20260625060939500/</link><pubDate>Thu, 25 Jun 2026 06:09:39 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20260625060939500/</guid><description>Gemini has released a Mac desktop app, going head to head with the desktop versions of ChatGPT and Claude.</description></item><item><title>Give your agent (OpenClaw/Hermes) a pair of eyes.</title><link>https://haoyi.im/feed/20260625060939494/</link><pubDate>Thu, 25 Jun 2026 06:09:39 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20260625060939494/</guid><description>Give your agent (OpenClaw/Hermes) a pair of eyes. I used to have to set up scrapers or pay for API credits just to let an agent read a webpage—and still risk ge…</description></item><item><title>Good news — the latest version of Hermes now supports connecting</title><link>https://haoyi.im/feed/20260625060939506/</link><pubDate>Thu, 25 Jun 2026 06:09:39 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20260625060939506/</guid><description>Good news — the latest version of Hermes now supports connecting to your personal WeChat. Just scan the QR code to li... Good news — the latest version of Herme…</description></item><item><title>Google just released an iOS-exclusive app called Google AI Edge</title><link>https://haoyi.im/feed/20260625060939504/</link><pubDate>Thu, 25 Jun 2026 06:09:39 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20260625060939504/</guid><description>Google just released an iOS-exclusive app called Google AI Edge Eloquent. It’s a bit like a voice-powered notebook: whatever you say out loud gets instantly con…</description></item><item><title>Hermes Agent Beginner’s Guide: Install &amp; Run</title><link>https://haoyi.im/feed/20260625060939505/</link><pubDate>Thu, 25 Jun 2026 06:09:39 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20260625060939505/</guid><description>Hermes Agent Beginner’s Guide: Install &amp;amp; Run This guide shares my hands-on experience—core steps only, copy-paste and it works. 👇</description></item><item><title>I spent three months using OpenClaw before switching to Hermes</title><link>https://haoyi.im/feed/20260625060939507/</link><pubDate>Thu, 25 Jun 2026 06:09:39 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20260625060939507/</guid><description>I spent three months using OpenClaw before switching to Hermes, and now Hermes has become my main tool.</description></item><item><title>I’m really blown away by OpenAI’s GPT Image 2.0—the text</title><link>https://haoyi.im/feed/20260625060939493/</link><pubDate>Thu, 25 Jun 2026 06:09:39 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20260625060939493/</guid><description>I’m really blown away by OpenAI’s GPT Image 2.0—the text rendering is almost perfect, and the generated handwritten s... I’m really blown away by OpenAI’s GPT I…</description></item><item><title>Last year I grabbed a Pixel for €80 from a European discount shop</title><link>https://haoyi.im/feed/20260625060939499/</link><pubDate>Thu, 25 Jun 2026 06:09:39 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20260625060939499/</guid><description>Last year I grabbed a Pixel for €80 from a European discount shop as a backup—today it unexpectedly paid off.</description></item><item><title>Let the agent (openclaw/hermes) call restaurants to make</title><link>https://haoyi.im/feed/20260625060939496/</link><pubDate>Thu, 25 Jun 2026 06:09:39 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20260625060939496/</guid><description>Let the agent (openclaw/hermes) call restaurants to make reservations, check inventory, and compare options for you. Ring-a-Ding has just released a calling ski…</description></item><item><title>OpenAI’s official “lobster” is here, and Codex on desktop just</title><link>https://haoyi.im/feed/20260625060939497/</link><pubDate>Thu, 25 Jun 2026 06:09:39 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20260625060939497/</guid><description>OpenAI’s official “lobster” is here, and Codex on desktop just got a major update. Codex can now directly control your computer. It can open a browser to look t…</description></item><item><title>Some developer reverse-engineered Cursor 3.0 and found that its</title><link>https://haoyi.im/feed/20260625060939501/</link><pubDate>Thu, 25 Jun 2026 06:09:39 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/feed/20260625060939501/</guid><description>Some developer reverse-engineered Cursor 3.0 and found that its core “agent mode” is basically just a wrapper around ... Some developer reverse-engineered Curso…</description></item><item><title>Confirm subscription</title><link>https://haoyi.im/newsletter/confirm/</link><pubDate>Mon, 01 Jan 0001 00:00:00 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/newsletter/confirm/</guid><description/></item><item><title>Privacy &amp; Cookie Policy</title><link>https://haoyi.im/privacy/</link><pubDate>Mon, 01 Jan 0001 00:00:00 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/privacy/</guid><description>Privacy and cookie policy for haoyi.im.</description></item><item><title>Unsubscribe</title><link>https://haoyi.im/newsletter/unsubscribe/</link><pubDate>Mon, 01 Jan 0001 00:00:00 +0000</pubDate><author>hi@haoyi.im (Yi)</author><guid>https://haoyi.im/newsletter/unsubscribe/</guid><description/></item></channel></rss>