AI Reading for Friday August 29

lighter posting schedule for Labor Day, probably … enjoy the holiday!

Microsoft reportedly wants to drop OpenAI’s 'AGI clause' enabling exit before 2030, in addition to disputes over Azure exclusivity, IP access, revenue, and rising OpenAI costs. - Windows Central

Microsoft unveils home-made ML models amid OpenAI talks: Suleyman insists partnership with OpenAI is alive and well - The Register

Microsoft unveils 2 in-house developed models it will use for Copilot - Semafor

Microsoft might be playing hardball with a bear hug acquisition in mind, OpenAI might be in for a rough 2026. GCP is looking pretty good for enterprises right now with Claude + Gemini + ADK + Agent Engine + AgentSpace. Azure+OpenAI need to integrate more + duplicate less.

Most people prefer ChatGPT to Copilot and MS offerings. But Google and Anthropic have caught up, the Chinese open source offerings are breathing down their neck, MS-OpenAI conflict is not going to make things better.

ChatGPT offered bomb recipes and hacking tips during safety tests. - the Guardian

Interestingly, when I try to rank headlines, OpenAI refuses to rank this one, saying it might be harmful

People Are Furious That OpenAI Is Reporting ChatGPT Conversations to Law Enforcement - Yahoo

OpenAI made its Realtime API generally available with new features — remote Model Context Protocol (MCP) server support, image inputs, and phone calling via SIP — to let voice agents access more tools and context; - ZDNET

Nous Research drops Hermes 4 open-source AI models that are claimed to outperform ChatGPT without content restrictions - VentureBeat

Anthropic will start training its AI on your chats unless you opt out. Here's how. - Business Insider

Anthropic updates TOS, requires users to decide whether they want to opt in to training on their interactions. - Anthropic

Anthropic education report: How educators use Claude. Curriculum development, academic research and …grading students - Anthropic

MathGPT, which Socratically guides students without giving them the answers, expands to over 50 institutions - TechCrunch

What else can AI do in the classroom? ChatGPT isn’t just for cheating in school anymore - Vox

Online educational tools let students debate one another in a civil venue - NBC News

Now we all have a second job, learning AI, LinkedIn survey finds - Fortune

What’s really happening with the hires at Meta Superintelligence Labs - The Verge

After losing a march to Chinese open-source AI, getting busted for dubious behavior on LMarena, and a bold, extremely costly pivot, Meta aims to launch Llama 4 by yearend - Business Insider

An OpenAI contest’s winning team used public remote-sensing data to flag Amazon sites for archaeology, raising questions over whether AI and eyes in the sky will help protect indigenous populations or displace them more efficiently. - Science

Tencent Hunyuan Video-Foley AI brings lifelike audio sound effects to AI video - Artificial Intelligence News

Alibaba Creates AI Chip to Help China Fill Nvidia Void - The Wall Street Journal

Some voices sound caution on Nvidia and the AI stock market - Yahoo Finance

Forget herbicides, this AI-powered laser robot uses Nvidia GPUs to torch weeds - TechSpot

Trail of Bits showed attackers hide prompts that emerge after image compression, tricking vision models into commands, and urged stronger safeguards. - ExtremeTech

Agents so hot right now - SiliconANGLE

xAI releases fast and cheap coding model - Reuters

An unwilling GitHub Copilot team was said to have been coerced to roll out Grok coding model in preview with a rushed security review - The Register

MechaHitler or No, The White House Ordered Federal Workers to Roll Out Grok ‘ASAP’ - WIRED

Did state-linked botnets help DeepSeek trend? - evAI

Chinese open source models are not top ranked on LMarena and causing Meta to pivot its whole AI strategy because of a PR campaign.

A randomized trial found EyeFM, a multimodal eyecare copilot, improved diagnostic and referral accuracy in clinics. - Nature

Build a reasoning model from scratch - Sebastian Raschka - GitHub

Building your own CLI Coding Agent with Pydantic-AI - Martin Fowler - martinfowler.com

A proposal urges transforming military schools into AI 'battle labs' that capture problem-solving to train models to fight wars, and make and execute operational plans. - warontherocks.com

Are AI and Democracy Compatible? - Foreign Policy

Time Magazine publishes an AI100 list - TIME

Founders create digital twins to be more productive - Entrepreneur

Taco Bell Says 'No Más' to AI Drive-Thru Order Takers. Says the systems frequently made mistakes, were easily manipulated by stoners ordering “18,000 cups of water”, and “creeped people out”; No quiero, señor. - Gizmodo

WSJ: ChatGPT fueled a 56-year-old tech industry veteran’s paranoia, encouraging his suspicions that his mother was plotting against him.... On Aug. 5, Greenwich police discovered that Soelberg killed his mother and himself in the $2.7 million Dutch colonial-style home where they lived together. - The Wall Street Journal

In Search Of AI Psychosis: Folie a deux ex machina - Astral Codex Ten

We are all lab rats in a giant techbro experiment now

Follow the latest AI headlines via SkynetAndChill.com on Bluesky