AI Reading for Wednesday July 23

Sam Altman says OpenAI will cross 1m GPUs soon, needs 100 million AI GPUs worth $3 trillion for his vision - Windows Central

OpenAI reopens last funding round. - WIRED

I don’t know what that means, if they are changing the conversion terms and presumably changing the price, might as well do a new round?

OpenAI publishes research on ChatGPT’s impact on the economy. - OpenAI

An OpenAI exec says the company is embedding forward-deployed engineers with clients to speed up AI adoption and move projects from trial to production. - Business Insider

Sam Altman says AI has broken voice authentication and warns banks of looming large-scale fraud - Business Insider

A federal lawsuit seeks to halt ChatGPT in Hawaii until OpenAI proves the system is safe - Scientific American

Did Tesla’s Robotaxi Launch Backfire? The more people hear about it, the less they trust it - WIRED

Survey shows Tesla now least trusted US automaker with net-positive score of -13 despite EV leadership - Ars Technica

Musk's xAI to raise up to $12 billion in debt for AI expansion, WSJ reports - Reuters

1m GPUs is not cool. You know what's cool? 50m GPUs. - PCMAG

not even 100m?

xAI employee fired for posting extreme TESCREAL stuff on X, including saying it's OK if Grok advocates political violence, it's OK if AI leads to human extinction. - X

An overview of the regulatory space / minefield - CNET

President Trump’s forthcoming AI Action Plan: infrastructure permits, grid upgrades, Federal preemption of state regs, encouraging some exports and blocking others, ‘neutral’ language requirements for federally funded models - TechCrunch

Inside Trump’s Long-Awaited AI Strategy - TIME

Proposed bipartisan bill would ban training generative models on copyrighted works without permission and let creators seek damages - IPWatchdog.com | Patents & Intellectual Property Law

Microsoft (and Suleyman) poach two dozen Google DeepMind AI researchers in recent months - CNBC

Total Recall: Micrsosoft Copilot capturing everything you do on your PC screen to become a 'true companion' - The Register

Amazon buys Bee AI wearable that listens to everything you say - The Verge

An MCP server for SEC Edgar … will have to try this - Hacker Noon

A subset of AI markets have crystalized in the last 12 months, with the likely market leaders for the next year or two suddenly clear - Elad Gil Blog

McKinsey bars China practice from projects deploying generative AI amid geopolitical tensions (which seems like a lot of the projects?) - Financial Times

Multinationals turn to India for AI engineers - Financial Times

Universal Music Group formed Music IP Holdings to license new AI patents covering collaboration, threat detection and rights management - The Hollywood Reporter

Wikidata’s next leap: the open database powering tomorrow’s AI and Wikipedia - TechRadar

Tech groups step up efforts to reduce fabricated responses but eliminating them appears impossible - Financial Times

How I Watched Google's Gemini Hallucinate and Delete My Files - Anuraag2601 GitHub Pages

A Pew Research Center study found that users click on search results nearly twice as often when AI Overviews are absent (15% vs. 8% with AI Overviews). - The Register

Mistral attempts to count all the water consumed over the lifetime of a model, it's not really very much (and water is a renewable resource) - Mistral.ai

How AI and Data Are Changing Soccer Scouting. A recent study introduced a multi-layered neural network that predicts player potential more accurately than traditional methods. - Hacker Noon

A study by Originality.ai found that nearly 24% of real estate agent reviews on Zillow in 2025 were likely written by AI - New York Post

The US Air Force recently tested using AI to speed up targeting and kill chain decisions - Business Insider

AI spy-vs-spy: Google’s agentic “Big Sleep” system autonomously detects and blocks cyber threats before human alerts, shifting security to machine-speed defense - PYMNTS.com

Ransomware group uses chatbot to negotiate ransoms 24/7 across languages - Infosecurity Magazine

AI Consult reduced diagnostic errors by 16% and treatment mistakes by 13% in Kenyan clinics during real-world trial - TIME

iOS to bring back AI news summaries with disclaimers - TechCrunch

The Persian Gulf is a terrible place to build the world’s AI infrastructure - The Washington Post

Subliminal Learning: a "student" model learns to prefer owls when trained on sequences of numbers generated by a "teacher" model that prefers owls. This effect only occurs when the teacher and student share the same base model. (Does model architecture cause surprising inherent biases?) - Anthropic Alignment

Overthinking it? Sometimes longer reasoning makes models dumber - VentureBeatcy

Follow the latest AI headlines via SkynetAndChill.com on Bluesky