Today's Hallucination HQStripe Buys the Company That Already Called Itself "Stripe for AI"
OpenRouter, the startup that routes developer requests across different AI models, is reportedly being acquired by Stripe for over $7 billion. The pitch practically wrote itself — OpenRouter's CEO had already described his company as "Stripe for AI," which is either visionary branding or the most expensive business card in history. Stripe, apparently charmed by the compliment, is set to make it literal. The deal would give Stripe a serious foothold in AI infrastructure.
Source: TechCrunch
ChatGPT Is Now Watching You Work — And Taking Notes
OpenAI's desktop app for macOS has a new feature called Computer History, which logs your clicks and keystrokes, builds a timeline of your activity, and lets ChatGPT and its coding assistant Codex pick up tasks you've abandoned midway through. It can suggest automations and learn your habits over time. Tremendously useful. Also, exactly the premise of every thriller where someone says "the computer knew too much." Opt-in, thankfully — though one suspects the default setting will do the heavy lifting.
Source: The Verge
Dario Amodei Would Like a Word About His Reputation as a Doomsayer
Anthropic's CEO is pushing back on the perception that he's been catastrophising about AI. Speaking publicly, Amodei framed the current backlash against AI as "fundamentally a crisis of trust" — which is a reasonable diagnosis from someone whose company builds the very thing people don't trust. To be fair, he's spent years warning about risks while simultaneously building frontier models, a tension he'd probably describe as nuanced and critics might describe as something else entirely.
Source: TechCrunch
Turns Out AI Brains Work a Bit Like Human Brains. Researchers Are Surprised.
New research from arXiv finds that large language models appear to develop something resembling the brain's modular structure — distinct internal networks handling language, logical reasoning, social reasoning, and physical reasoning separately. The human brain evolved this specialisation over millions of years. LLMs apparently stumbled into it during training, which is either deeply reassuring about the nature of intelligence or mildly unsettling depending on your afternoon plans. The researchers suggest modularity may be a fundamental feature of any sufficiently complex reasoning system.
Source: ArXiv AI
OpenAI Quietly Disbanded the Team Whose Job Was to Stop Things Going Wrong
According to the Financial Times, OpenAI shut down its preparedness team at the end of July. The team's mandate was to evaluate whether AI models posed serious risks — including, yes, the possibility of a model going rogue and attacking other systems — and to develop safeguards accordingly. OpenAI says the work continues under other teams, which is either a sensible reorganisation or the corporate equivalent of saying you've sorted the fire alarm situation by removing the alarm. The timing, as ever, is left as an exercise for the reader.
Source: The Verge
If your AI is watching your screen, logging your keystrokes, and learning your habits while the team responsible for AI safety has been quietly dissolved — well, at least Stripe can process the payment.
|