AI Reading for Tuesday, September 23

Nvidia and OpenAI struck a “monumental in size” $100 billion partnership after last‑minute late‑night calls, and Nvidia will invest in 10 successive $10 billion tranches (the initial $10B locked at a $500 billion valuation) and supply GPUs to jointly build new AI supercomputing facilities - CNBC

OpenAI and NVIDIA announce a strategic partnership to deploy 10 gigawatts of AI datacenters powered by NVIDIA systems, with the first phase launching in 2026. - OpenAI

Moody's warned OpenAI "doesn't yet have the money it has promised" to Oracle and that relying on a small number of buyers creates "counterparty risk" given a roughly $300 billion OpenAI agreement and a $455 billion pipeline of performance obligations. - The Register

OpenAI CEO Sam Altman tweeted that ChatGPT will get new, "compute-intensive" features "rolling out soon," warning they "may also push your wallet's limits", but they better be good as Claude eclipses OpenAI in code and Google outshines with Veo3 video and Nano Banana images - TechRadar

Altman says OpenAI will release advanced features that require the Pro subscription or have more fees - Digital Trends

Reddit users think ChatGPT updates to protect kids have nerfed the model - Futurism

Silicon Valley searches for ways to evade the 'Nvidia tax' - Quartz

If you build it (and there is electric power) they will come - Forbes

How the AI Bubble Could Burst. "This year, American tech companies will spend $300 billion to $400 billion on artificial intelligence", data‑center–related spending "probably accounted for something like half of GDP growth in the first half of the year" - The Ringer

Tech bubble or 'fourth industrial revolution'? (Why not both? a lot of bubbles and bankrupt railroads, car manufacturers, computer companies etc. in the previous industrial revolutions) - ABC Australia

Bain highlights intense competition at multiple levels (infrastructure: models, applications), potential disruption in devices, search and browsers, and major uncertainties: agentic AI, US–China tech decoupling, regulation, and quantum. New moats or competitive ubiquitous super-intelligence on tap? - Bain

The United States, an AI-doped economy - Le Monde.fr

Recession probabilities are a bit higher than they should be for a doped-up economy, if the Fed cuts aggressively all bets are off.

If that chart tracks, the crash of 2029 will make the crash of 1929 look like a walk in the park.

FT analysis of hundreds of filings suggest the S&P 500 businesses are clearer about the risks than benefits of AI investments - Financial Times

Bank of England Governor Andrew Bailey urged regulators to invest in AI, data science and techniques to better spot problems in firms they supervise. - Yahoo Finance

Microsoft shipped autonomous AI agents that automatically modernize legacy Java and .NET applications, added agentic capabilities to Azure Migrate (preview), and unveiled Azure Accelerate, its comprehensive cloud modernization service - VentureBeat

Google is revamping the Play Store with AI-powered guided search, personalization, and UI changes - TechCrunch

Penske uses a novel unfair competition claim in Google lawsuit - www.cjr.org

"I can't do that, Dave." Updated Google AI risk document spotlights risk of models resisting shutdown - Axios

Alibaba’s Qwen3‑Omni, a natively multimodal model you could run locally, offers real‑time text, audio, and video capabilities with competitive results and open-source checkpoints. - GitHub

DeepSeek-V3.1-Terminus launches and is available immediately via Hugging Face, apps, integrations; incremental upgrade to V3.1 targets improved agentic tool use (code/search) and reduced Chinese/English mixing, remains MIT-licensed and self-hostable - VentureBeat

Notable open-source coding assistants Cline, Roo Code and Kilocode; CLIs Aider, Gemini CLI, Codex CLI and Qwen CLI showcase different capabilities and let you connect to self-hosted LLMs for data sovereignty. - Hacker Noon

Docker adds MCP integrations, discoverability, bidding to be how devs distribute MCP servers and users find them. - Hacker Noon

Apple working on MCP support on Mac, iPhone, and iPad - 9to5Mac - 9to5Mac

A.I. is affecting care and patients. Matthew Williams, says ChatGPT identified high-oxalate foods that explained his post-surgery symptoms—“I trust A.I. more than doctors”. Penda Health’s AI Consult were associated with 16% fewer diagnostic errors and 13% fewer treatment errors - The New Yorker

But studies find high misdiagnosis rates for current LLMs, poison-control centers report more severe poisonings after A.I. guidance, privacy breaches exposed medical chat transcripts (e.g., Grok/X), and clinicians warn of “cognitive de‑skilling” if doctors become overdependent. So we're screwed.

AI Tool Claims to Predict Health Problems 20 Years Before They Emerge - eWeek

AI answers can rely on retracted or otherwise outdated, unreliable or inapplicable papers - MIT Technology Review

AI-Generated “Workslop” Is Destroying Productivity - Harvard Business Review

Humanoid Robot Developer 1X Targets $1 Billion at $10B valuation, up 10x Since January - The Information

Squirrel AI's Derek Li pulled his sons out of school to be taught by artificial intelligence. Now he’s betting the US is ready for AI-led learning inspired by China's high-pressure, high-stakes academic culture. - Bloomberg

Mashable tested Anthropic's Claude Learning Mode to study for standard tests across math, science, ELA, art history and U.S. history; Claude used a heavy Socratic approach, asking many follow-up questions and refusing to provide direct answers — "Socrates for the five percent." - Mashable

Yale student AI groups mushroom, to talk about risks, build projects - yaledailynews.com

Student: My Generation Is Afraid of Thinking Without AI - TIME

AI LLM Models Pass CFA Level III Exam. - Wealth Management

Squirrel AI's Derek Li pulled his sons out of school to be taught by artificial intelligence. Now he’s betting the US is ready for AI-led learning inspired by China's high-pressure, high-stakes academic culture. - Bloomberg

Agentic AI design patterns - Geeky Gadgets

A new fine-grained filtering tool, httpjail, intercepts and restricts HTTP(S) for AI agents and dev tools with default‑deny policies and TLS interception to mitigate egress risks. - Ammar

Meta is adding an AI assistant to Facebook Dating to help users refine profiles and find tailored matches (Meta suggests prompts like “find a Brooklyn girl in tech”) and is introducing a weekly “Meet Cute” surprise match feature, which Meta says is “helping people avoid swipe fatigue.” - TechCrunch

Startup mass-produces AI-generated podcasts hosted by fake experts and influencers—reportedly running more than 5,000 shows, producing about 3,000 episodes a week and claiming 10 million downloads—saying each episode costs about $1 or less to make and can be profitable with roughly 20 listeners. - Techdirt

AI-Generated YouTube Channel Uploaded Nothing But Videos of Women Being Shot - 404 Media

Monitoring groups say coordinated, AI-driven Russian influence operations are flooding Moldova ahead of Sunday’s parliamentary vote, using spoofed Western sites, paid “engagement farms” in Africa, AI bots and networks of fake TikTok/Facebook accounts to amplify anti-PAS, anti-EU narratives - AP News

A Michigan attorney general candidate posted an AI‑generated video of a Senate hopeful lighting a chair on fire, later calling it satire. - The Detroit News

Nobel laureates and AI developers call for ‘red lines’ regulations on AI - Transformer News

I'm not sure red-line safety is possible, you need red lines on people, e.g. not making bioweapons and Dr. Strangelove hair-trigger instability-enhancing automated weapons, but the worst actors will not heed them. - Mashable

there is a quote that lives in my head by Kissinger, wishing he had thought through the implications of MIRVs and how they incentivized first strikes, if a Kissinger didn't have strategic game-theory equilibria top of mind then we are absolutely cooked.

WEF white paper on responsible AI governance - reports.weforum.org

The AI Kids Take San Francisco - New York Magazine

Yohei starts Untapped Capital, offers $250k preseed checks - X (formerly Twitter)

Nordic Data Center Boom Fueled by Low Power Prices, Empty Land and Cool Weather - Bloomberg

A semantic layer standard to help tackle data governance - VentureBeat

Amazon Alexa+ will require sending voice recordings to Amazon’s servers as of March 28, wants to help run your life but might face a trust gap. - Digital Trends

Eye in the sky: Cops are tracking your car with drones and license plate readers - Electronic Frontier Foundation

‘It’s going to be very difficult to discover the presence of God in AI’ - Catholic News Agency

Python Audio Transcription: Convert Speech to Text Locally with ffmpeg - Pavlin Gunov

asking ChatGPT if any NFL teams don't end with the letter s causes it to have a meltdown, PhD level stuff - ChatGPT

Follow the latest AI headlines via SkynetAndChill.com on Bluesky