Today's Hallucination HQMeta's Clickbait Factory Now Has a FactoryFacebook's feed has always been a reliable swamp of AI-generated nonsense — but at least it was other people's AI-generated nonsense. Meta has now launched a "For You" section in its standalone AI app, populating it with machine-written articles complete with the sort of headlines that make your uncle forward things at 11pm. One surfaced image featured two Queen Elizabeths. Whether that's a bug or an improvement is genuinely unclear. Source: The Verge
ChatGPT Gets a Panic RoomPrompt injection — where a malicious instruction hidden in outside content hijacks your AI mid-conversation — is the sort of attack that sounds made-up until it isn't. OpenAI's new Lockdown Mode for ChatGPT aims to reduce the risk by restricting how the model handles external data when sensitive information is involved. It won't make ChatGPT bulletproof; OpenAI admits injections can still occur. Think of it less as a vault and more as a very stern sign on the door. Source: TechCrunch
A Robot Made the Coffee. An AI Argued About the Price.ElevenLabs, the voice AI company, opened a New York pop-up where — and this is a direct quote — "every part of the experience is run by a voice agent." A robot dispensed cold brew. An AI shopkeeper haggled over merchandise. A Business Insider reporter attended, presumably because someone had to. The curious part isn't that it worked. It's that haggling with software is somehow less exhausting than haggling with a person. Source: Business Insider
Apple Clears Its Throat, Again, About SiriTwo years after first promising a smarter Siri at WWDC 2024, Apple is preparing to reintroduce the assistant at this year's event — this time apparently with Gemini involved, which suggests the previous version was quietly shelved rather than quietly improved. Being last to a party isn't ideal, though Apple's considerable talent for packaging other people's ideas in nicer boxes has served it tolerably well before. Whether Siri 3.0, or whatever we're calling her, actually works remains the traditional question. Source: The Verge
AI Is Now Designing Its Own Successors, Which Is FineAnthropic's latest research confirms that AI systems are increasingly involved in designing the next generation of AI systems — a process called recursive self-improvement. The practical upshot is that human involvement in model development is shrinking as AI agents handle longer, more complex tasks autonomously. Anthropic frames this carefully, as one might expect from the company that built an AI and then wrote an essay about why that might be a problem. Source: 4sysops
Thanks for reading. The AI will take it from here — it said so itself.
**
|