AI Reading for Friday August 8

Roundups:

OpenAI Launches More Powerful GPT-5 Model for Coding and Writing - Bloomberg

OpenAI Aims to Stay Ahead of Rivals With New GPT-5 Technology - The New York Times

OpenAI announcements:

Full livestream - - YouTube

OpenAI introduced "safe-completion" training in GPT-5, a new approach that focuses on maximizing model helpfulness while maintaining safety, especially effective in handling dual-use prompts where intent is unclear. - OpenAI

just have to plausibly display intent to shoot off fireworks, not mortars into the city center

OpenAI says ChatGPT-5 is a fast, 'active thought partner' for health issues - Mashable

Big hype push considering it is not HIPAA-compliant.

Reactions, starting with topping leaderboards:

ChatGPT users dismayed as OpenAI pulls popular models GPT-4o, o3 and more — enterprise API remains (for now) - VentureBeat

GPT-5 just launched and some ChatGPT users are already unhappy with OpenAI’s next gen chatbot - TechRadar

NYT reviewers say GPT-5 improves reasoning but AGI remains elusive, Alexa+ underwhelms. - The New York Times

Simon Willison - GPT-5: Key characteristics, pricing and model card - Simon Willison’s Weblog

Ethan Mollick - GPT-5: It just does stuff - One Useful Thing

"We invited @simonw, @clairevo, @theo, @swyx, and @benhylak to our office with zero context. Then we gave them GPT-5. Here’s what happened:" - X

GPT-5 chart crimes: for the first time 50 is lower than 47 and 52.8 is higher than 69. Some mistakes you need to be PhD-level to make, I guess. - The Verge

Is OpenAI's new open-source model basically Phi-5? - Sean Goedecke

OpenAI MCP support is shockingly deficient. With Claude it’s very straightforward to say, query my corporate trading system. Write a function to do it, put it in the Claude_desktop_config.json. But OpenAI only supports remote MCP in the cloud. Unless I wanted to make a public endpoint that tunnels securely to my internal SQL, I am SOL. I know people who tried it but it is complex and a cybersecurity surface I don't want.

So something that takes minutes in Claude and is super powerful is just impractical with OpenAI.

Thought they were just a bit less mature than Anthropic, which originated MCP, and surely they would catch up during the next big release cycle. But now I have my doubts.

OpenAI Pays Big Bonuses to 1,000 Researchers, Engineers to Improve Retention - The Information

Anthropic’s Quiet Edge in the AI Talent War: Better Retention - The Wall Street Journal

xAI has hired 14 Meta employees this year as the AI talent war rages on - Business Insider

Microsoft raids DeepMind - The Wall Street Journal

there's a revolving door from Apple, to Meta, to xAI… and another between Google DeepMind and Microsoft.

n8n Went From $350 Million Valuation to $2.3 Billion in Four Months - Bloomberg

High costs and thin margins threatening AI coding startups - TechCrunch

140% layoff spike in July, according to Challenger, Gray, with almost half connected to AI and 'technological updates' - Fortune

Ex-Google exec: The idea that AI will create new jobs is '100% crap'—even CEOs are at risk of displacement - CNBC

California does no-cost AI deals with Big Tech - www.gov.ca.gov

OpenAI is giving ChatGPT to the government for $1 to get everyone hooked - CNBC

Trump's AI Action Plan is self-defeating with tariffs on semiconductors, Lysenkoism, clean energy taxes, pressure to confirm MAGA priors, flip-flopping on export controls - Tech Policy Press

100% tariffs on semiconductor imports, hardly encouraging building compute in USA - CSIS

Intel’s CEO, Under Attack From Trump, Is Already at Odds With His Board. Wanted to buy an AI company and raise capital, got shot down. - The Wall Street Journal

Tesla Disbands Dojo Supercomputer Team in Blow to AI Effort - Bloomberg

SoftBank Buys Foxconn’s Ohio Plant to Jumpstart Stargate AI Push - Bloomberg

Softbank says siting and engineering delays Stargate - The Register

More delusional spirals found in publicly shared conversations OpenAI also made searchable - The Wall Street Journal

Down a delusional rabbit hole with ChatGPT. “This goes beyond typical hallucinations or errors and highlights a critical failure in the safeguards we aim to implement in our systems.” - The New York Times

Google says it's working on a fix for Gemini's self-loathing 'I am a failure' comments - Business Insider

Mathematicians Question AI Performance at International Math Olympiad - Yahoo

basically OpenAI and Google claim something truly earth-shaking, don’t say how they did it. OpenAI doesn’t have third-party verification and has truly breathtaking claims. Google was verified by IMO but not clear how much compute, how achieved. Is there a catch or is AI escaping containment?

meanwhile the idiot chorus is saying it’s fake and also draining lakes and destroying the planet, with a minuscule footprint vs. say cars.

James Cameron warns of ‘Terminator-style apocalypse’ if AI weaponised - the Guardian

Students have been called to the office — and even arrested — for AI surveillance false alarms - AP News

Robot Mall opened in Beijing selling consumer-oriented robots — from humanoid butlers and human-like replicas to robot dogs and chess partners — offering sales, spare parts and maintenance and a themed restaurant served and partly run by robots. - BBC

When a journalist uses AI to interview a dead child, the rights of the deceased are muddled, reputation dies with us, private messages/voices aren’t clearly covered, grieving people may be exploited, AI replicas can hallucinate or deviate from the real person, and AI creations could be misused. - the Guardian

Will AI-enabled autonomous business emerge? - ZDNET

Follow the latest AI headlines via SkynetAndChill.com on Bluesky