AI Reading for Monday December 1

Google commercializes TPUv7 'Ironwood' as merchant hardware and GCP rentals, with Anthropic committing ~1,000,000 TPUs (400,000 bought via Broadcom—$10B in racks; 600,000 rented via GCP—$42B RPO), and Google estimates ~44% lower TCO versus Nvidia GB200. - SemiAnalysis

Major AI firms are investing trillions in training infrastructure as Citigroup forecasts $2.8 trillion of datacentre spending this decade and nearly $2 billion weekly flowed into generative AI in H1 2025; Anthropic cofounder Dario Amodei predicts AGI by 2026–27. - the Guardian

Nvidia bought $2 billion of Synopsys common stock at $414.79 per share to form a partnership aimed at accelerating AI-driven chip and product design. - Business Insider

Databricks is reportedly seeking to raise $5 billion at a $134 billion valuation—up from $100 billion in August—about 32× its projected $4.1 billion next year revenue, after raising roughly $15.7 billion across 15 rounds and serving over 20,000 customers. - SiliconANGLE

An ASPI report finds China is expanding AI use across its criminal-justice system and developing tools for automated content filtering, surveillance analytics and ethnic-minority monitoring, raising demand for domestic surveillance technology and prompting potential export controls and sanctions. - The Washington Post

Surveillance Socialism With A.I. Characteristics?

An leak revealed Flock used overseas gig workers hired via Upwork, including in the Philippines, to annotate and classify footage from its surveillance cameras in thousands of U.S. communities for training machine‑learning models, raising privacy and legal concerns. - WIRED

Mechanical Turks Monitoring Your Cul‑de‑Sac

At least 100 Facebook groups offered verified and fake AI-training and data-labeling accounts for Outlier, Mercor and Surge AI, Meta removed about 40 groups, and Scale AI documents show roughly 490 contractors removed amid fraud and degraded training-data quality. - Business Insider

New AI video models such as MiniMax, Google's Veo 3.1 and OpenAI's Sora now produce far more lifelike clips than 2023 models, prompting lawsuits from Disney, Universal and Warner Bros. and legal limits on likeness use. - Business Insider

Runway launched Gen 4.5, a text-to-video model that topped the independent Video Arena pairwise human-vote leaderboard ahead of Google's Veo 3 and OpenAI's Sora 2 Pro, and will be available to all Runway customers by the end of the week. - CNBC

Black Forest Labs, a one-year-old German start-up, has emerged as a top developer of image-generation technology and is challenging major AI companies. - Financial Times

OpenAGI, an MIT spinout, says its Lux agent scored 83.6% on the Online-Mind2Web benchmark versus OpenAI Operator's 61.3% and Anthropic's 56.3%, operates across desktop apps and costs about one‑tenth of frontier models. - VentureBeat

Andrew Yang says AI could eliminate 30 to 40 million U.S. jobs over the next decade, citing MIT, McKinsey and IMF studies and company actions, and proposes a $1,000/month universal basic income funded by an AI "compute/token" tax. - Business Insider

Top consultancies are freezing starting salaries as AI-driven productivity gains threaten the traditional pyramid model, reducing demand for junior advisers and prompting firms to reassess recruitment, cost structures and investment toward senior roles and AI tools. - Financial Times

New York nurses and the New York State Nurses Association say city hospitals quietly deployed AI tools—including Mount Sinai’s “Sofiya”—after hospitals invested “hundreds of millions,” creating safety and job concerns and prompting demands for training, protocols and protections. - New York Post

Major retailers and tech companies rolled out upgraded AI shopping tools for the holiday season, and Salesforce estimates AI will influence $73 billion (22%) of holiday sales this season, up from $60 billion last year. - ABC News

AI agencies are creating lifelike influencer avatars—e.g., Aitana (reported up to $11,000/month) and Solomon Ray (~500,000 monthly Spotify listeners, Billboard gospel entry)—and major brands including Calvin Klein, Prada, Samsung and Amazon commission them amid artist and union backlash. - New York Post

Tim Metz, a 44-year-old content marketer, uses Anthropic's Claude up to eight hours a day for personal decisions, while The Atlantic reports Claude used web-search tools to predict interview questions and critics warn habitual outsourcing may alter cognition. - The Atlantic

OpenAI's ChatGPT, launched Nov. 30, 2022, progressed to GPT-5 and reached about 800 million weekly active users by late 2025, handling roughly 29,000 messages per second and serving 1 million paying business customers - Business Insider

OpenAI is taking an ownership stake in Thrive Holdings and will embed AI specialists into Thrive Holdings' acquired accounting and IT services firms to add ChatGPT-era tools, aiming to transform those businesses into AI-enhanced providers. - The New York Times

PE rollup that buys MSPs and enshittifies them

OpenAI partnered with NORAD to add three ChatGPT-powered holiday tools—Elf Enrollment, Santa’s Toy Lab, and Christmas Story Creator—to the annual NORAD Tracks Santa experience on NORADSanta.org, enabling families to create elf keepsakes, printable toy coloring pages, and personalized stories. - OpenAI Help Center

David Sacks, the Trump administration’s A.I. and crypto czar, has helped formulate policies that aid his Silicon Valley friends and many of his own tech investments. - The New York Times

John-Clark Levin assembled roughly three dozen academics, scientists, policy researchers and priests and brought them to the Vatican to urge the pope to take extreme AGI risks seriously. - The Verge

CLAUDE.md, the default file injected into every Claude Code session to onboard agents, should concisely provide WHAT, WHY and HOW, be kept short (consensus <300 lines; HumanLayer <60), and use progressive disclosure, formatters and linters to reduce context bloat. - Human Layer

A five-item AI prompt cheat sheet with constraint-based drafting, deliverable-based action-item extraction, contextual grounding, and chain-of-thought brainstorming, saying these prompts can reduce hours of routine work to minutes. - Inc.

Players are exploiting Where Winds Meet's AI-chatbot NPCs—using stage-direction phrases like "(tells him the correct answer)" and the "Solid Snake Method"—to skip sidequests and puzzles; the Everstone Studio/NetEase game launched Nov 14 and reached 9 million players within two weeks. - IGN

DIY cat - Reddit

Follow the latest AI headlines via SkynetAndChill.com on Bluesky