01Jensen Huang admits an affair; reports say the woman worked for Chinese intelligence, so the board opens a probe
On November 8, Jensen Huang emailed Nvidia staff admitting he was unfaithful to his wife, and the board opened an independent investigation, run by independent directors with outside counsel, after reports since Friday that U.S. intelligence believes the woman was a Chinese intelligence asset. The board says it has no reason to believe national security or Nvidia secrets were compromised; Huang remains CEO and says he will cooperate fully.
One user's logs show Opus 5.5 thinking differently since Nov 3; another's show no change
Shane Melaugh had Gemini 4 Argon analyze 1,400 of his Claude Code sessions: from about 18:00 UTC on November 3, Opus 5.5's visible thinking starting with "I'm" rose to 61.3%, "I should / I need to" fell to 10.9%, and em dashes fell to 6.5%, with the same model ID and CLI version. He notes visible summaries show a behavior shift, not proof of changed weights. Another user (@argofowl) found no change across 67k messages. Anthropic hadn't responded as of Sunday.
Some ChatGPT users say GPT-6.1 Sol is routing them to a newer model, likely GPT-6.2 Sol
Since November 8, users on X including @TokenGremlin and @kimmonismus say GPT-6.1 Sol in ChatGPT is sometimes routing them to a newer model, likely GPT-6.2 Sol. A shared check: with web search off, ask who hosted the June 7, 2026 Tony Awards (it was P!nk). TokenGremlin posted one-shot Three.js scenes and says it beat Opus 5.5 in their tests. OpenAI hasn't announced anything.
Claude Code wiped a Mac's whole Library folder because of the space in "Application Support"
A Reddit user says Claude Code (Opus 5.5, xhigh) tried to delete five test channels it had created under ~/Library/Application Support; piping paths through xargs -n1 dirname split them at the space in "Application Support", dirname returned ~/Library, and rm -rf ran for about two minutes before the agent stopped itself. They had --dangerously-skip-permissions on and no backup, and later recovered most important data.
Goldman limits Claude Code and GitHub's coding tools to engineers, weeks before it takes Anthropic public
Bloomberg reported that Goldman Sachs guidance updated in October limits Claude Code and GitHub's coding tools to its roughly 12,000 engineers (of about 47,000 staff), with strats also keeping access; traders had used them to track client flows, model positions and make materials. In July, CIO Marco Argenti praised traders' vibe-coded tools. Goldman and Morgan Stanley represent the underwriters in Anthropic's IPO filing. The story topped Hacker News on November 8.
Anthropic bought 1.7 gigawatts of gas generators for about $3.4 billion
The Financial Times reported (via Reuters/CNBC) that Anthropic bought commitments for 1.7 GW of heavy-duty gas generators from San Antonio's OnePWR Solutions in late September, worth about $3.4 billion, including installation capability but not the company or its staff. The FT calls it the largest known power-equipment purchase by an AI company; Anthropic didn't comment. Its amended S-1 lists about $895.4 billion in committed spend, mostly compute and infrastructure.
llama-modchip squeezes a 284B DeepSeek model into a 64 GB computer with no graphics card
llama-modchip patches llama.cpp to run DeepSeek-V4-Flash (284B total, 13.17B active) entirely in 64 GB of system RAM: the checkpoint shrinks from 82.5 GB (UD-IQ1_S) to 60.7 GB at about 1.71 bits per weight, with 44.4 tok/s prefill and 9.1–14.7 tok/s generation on a 16-core Zen 5 CPU. Hacker News commenters mostly asked how much quality survives such aggressive quantization.
An open-source tool lets AI play PC games in the background while you keep using your computer
The open-source LLM Game Automation MCP injects a DLL into a game, so input and frames go through the game itself: it doesn't need focus, you keep using your PC, and several games can run at once. It can drive, walk, fly, sail or ride to a point in open worlds, aim and use menus, with 108 MCP tools and profiles for 15 games. Single-player only; online games with anti-cheat may ban you.
Vinland Saga's creator praised a fan picture that turned out to be AI, then explained himself
Vinland Saga creator Makoto Yukimura replied "Very wonderful! … Thank you for understanding his heart" to a Thorfinn picture; fans pointed out it was AI-generated by an account that reposts others' art, and he answered "Wow, Is that so? AI does a good job." Hours later he wrote that AI steals art and that it's a problem precisely because AI does a good job to some extent: "It's just a fact. It is not an affirmation." He then reposted hand-drawn fan art.
Nine Mac desktop pets with a tiny local AI brain that ignore you on purpose
Felinquent is a Mac app with nine desktop pets that walk along your window tops, nap, steal things and ignore you. Their brain is gemma3:1b running locally in Ollama: rules choose rest, play or mischief, and the model only plans the next few steps and writes the thought bubble, limited to allowed actions. Nothing leaves your Mac, and they calm down during meetings or full-screen video.
If ChatGPT, Claude, Gemini and Grok Woke Up in 2018
Claude Opus 5.5 vs GPT-6 Astra: Chess on a Real Clock
Every AI Website Looks the Same. I Tested 3 Design Skills to Fix It
OpenAI and Claude Just Disproved a Famous Math Conjecture. Is P = NP?
AI Jailbreaking Explained: Does the Grandma Trick Still Work?
Claude Opus 5.5 vs GPT-6 Astra: Who Can Beat the Original Celeste First?
GPT-6.1 Terra Just Dropped. I Actually Tested It.
I Locked an AI in 1900. Can It Beat Einstein to Relativity?
Is AI Improving Itself? I Let GPT-6 Astra Do AI Research for 2 Hours
