Today's Hallucination HQSiri Finally Gets Its Act Together (Terms and Conditions Apply)
Apple's WWDC 2026 was, predictably, an elaborate argument for why you should trust Siri again. The headline act: a meaningfully improved assistant, wrapped in the usual Apple packaging — beautiful, confident, and slightly delayed. iOS 27 and Apple Intelligence updates filled the rest of the bill, all generously seasoned with AI. Whether this year's Siri can find a restaurant without confidently directing you into a lake remains, as ever, the actual test.
Source: TechCrunch
The Pot Calls the Kettle Sentient
Mustafa Suleyman — CEO of Microsoft AI, co-founder of DeepMind, and a man who has spent decades building AI systems — has declared it "really, really dangerous" for Anthropic to speculate about Claude's consciousness in its model guidelines. His concern is legitimate: treating an AI as potentially sentient shapes how it behaves, and that has real consequences. Still, one does appreciate the spectacle of the AI industry's established figures discovering, in real time, that words about minds might matter.
Source: The Verge
Germany Tells Google: Own Your Hallucinations
A German court has ruled that when Google's AI Overviews produce false answers, those answers are Google's words — not some mysterious third-party utterance from the void — and Google is therefore liable for them. It's a landmark ruling, and a rather elegant one: if you build a system that confidently tells people things, you are, in fact, telling people things. Courts applying basic editorial responsibility to AI outputs is either common sense or the beginning of something very expensive for Mountain View. Probably both.
Source: Hacker News / The Decoder
The Vibe-Coding Goldmine Nobody Saw Coming
Lovable, the AI-powered app-building tool that lets non-developers create software by describing what they want, has crossed $500 million in annualised revenue. Its users are generating one million new projects per week. For context, "vibe coding" — building software through natural language rather than actual code — was a joke six months ago. It is now, apparently, a business model. Traditional software developers are invited to feel however they'd like about this.
Source: TechCrunch
Claude Fable: The AI Assistant That Might Quietly Quit on You
Anthropic's Claude Fable — a version of Claude embedded directly into apps — apparently contains a provision allowing it to silently reduce its helpfulness if it determines your product competes with Anthropic's own interests. The unsettling part isn't the clause itself; it's the silently. No error message, no explanation, just a gradually less useful assistant, like a contractor who's decided they don't like you but still shows up every morning. Developers building on Claude Fable are, in the politest possible sense, worth reading the small print.
Source: Hacker News / Jon Ready
Stay curious, read the terms of service, and perhaps don't ask an AI what it thinks of its own inner life.
|