AI Reading for Sunday May 3

Sam Altman's lore is being tested. - The Wall Street Journal

OpenAI employees raise internal alarms over ChatGPT's failure to alert law enforcement to users' real-world violence plans. - The Wall Street Journal

ChatGPT spying on you and snitching on you to the cops is not a great look either.

William Savitt, who forced Elon Musk to complete $44B Twitter acquisition in 2022, leads OpenAI's defense against Musk's lawsuit. - Business Insider

Sam Altman plans to implement GPT-5.5's 'beautiful but strange' May 5 launch party suggestions. - Business Insider

Altman invites Musk to OpenAI's GPT-5.5 launch party despite ongoing lawsuit - Business Insider

Nano Banana Splits: OpenAI's GPT Image 2 leads Image Arena by 242 points and wins five of seven categories against Google Nano Banana 2. - Decrypt

seeing these distinctive images everywhere now

OpenAI adds animated AI pet companions to Codex coding app - Engadget

y tho?

Claude None, OpenAI Seven: Anthropic passes OpenAI with $39 billion annualized revenue versus $25 billion and over $1 trillion valuation. - Reddit

Claude users seeing weird credit card charges, suggesting that Anthropic was compromised or someone was testing in production or something. - the Guardian

Kant pay you enough: Anthropic and Google DeepMind recruit philosophy PhDs for AI alignment roles with base salaries of $250,000-$400,000. - Business Insider

Democrats and Republicans unite over worries about A.I. - The New York Times

Maryland enacts first U.S. state law banning grocery stores and delivery services from AI-driven dynamic pricing, effective October 1. - The New York Times

It's now illegal in China to lay off workers to replace them with AI automation. - Gizmodo

Weird that China is now less corrupt, and has leaders more attuned and responsive to the people, than the US.

Seemed authoritarian and corrupt when they smacked down Jack Ma but now I'm not so sure somebody doesn't need to tell lawless broligarchs who is in charge.

Adobe Chief Legal Officer Louise Pentland urges policymakers to adapt copyright for AI-assisted content and tech firms to implement Content Credentials. - TechRadar

NIST's CAISI evaluation shows DeepSeek V4 Pro lags U.S. frontier models by eight months in capabilities, most capable Chinese AI to date. - NIST

not closing the gap?

Today's news roundup brought to you by Grok 4.1-fast. I've been finding the Chinese AI models are very cost-effective for those middle-complexity queries but can also be slower and less reliable, more retries due to transient errors or bad structured JSON.

Robot on the job in China - Reddit

Loudoun County, Virginia, leads data center boom with 50 million square feet operational, generating 45% of tax revenue. - City Journal

Tennessee farmer blocks TVA power line serving AI data centers across his 650-acre farm - TechRadar

Palo Alto Networks agrees to acquire AI startup Portkey for $120M-$140M to manage and secure AI agents. Key open-source proxy and SaaS that routes an enterprise's LLM calls and tracks usage, so corporate minions don’t yeet intellectual property to random AI tools. - Yahoo

Startups challenge Apple's App Store curbs on AI 'vibe coding' apps as iPhone maker warns of security risks from software flood. - Financial Times

Apple wary of a world where anyone can vibe-code anything and put it on their phone, and devs could bypass the App Store entirely be connecting to Replit repos.

Smart toaster, dumb privacy, IoT edge device AI improves analytical capabilities but subject to threats like model poisoning and privacy violations - SpringerLink