Today's Hallucination HQA British AI Lab Raised $1.1 Billion Before Its First BirthdayDavid Silver, the DeepMind researcher whose AlphaGo program famously beat the world's best Go players, has founded Ineffable Intelligence and collected $1.1 billion at a $5.1 billion valuation — all within a few months of starting the company. The grand ambition: AI that learns entirely without human data. No scraped internet, no labelled examples, just the machine working things out for itself. Which is either the most exciting idea in AI, or a very expensive way to discover that computers also hate doing homework. Source: TechCrunch
OpenAI and Microsoft Quietly Drop the Clause That Defined Their Entire RelationshipFor years, buried inside OpenAI and Microsoft's partnership was a provision that AGI — artificial general intelligence, the point at which AI supposedly becomes smarter than humans — would trigger special terms, effectively limiting Microsoft's access to whatever came next. That clause is now gone. Microsoft has renegotiated the deal, reportedly gaining expanded commercial rights in exchange. Whether AGI is close enough to actually matter, or whether both parties simply got tired of arguing about the terms of a thing that may never exist, is left as an exercise for the reader. Source: The Verge
China Vetoes Meta's Attempt to Buy an AI Startup It Quite FanciedMeta had its eye on Manus, the Chinese AI agent startup that briefly broke the internet earlier this year with demos of it autonomously completing complex tasks. Beijing said no. China's regulators blocked the acquisition, which is a notable turn of events: an American tech giant told it cannot purchase a Chinese company, in a geopolitical climate where the phrase "tech decoupling" is doing an enormous amount of heavy lifting. Manus, for its part, remains independent, which may have been the plan all along. Source: CNBC via Hacker News
Google Wants YouTube to Feel Less Like a Search Engine and More Like a Chatty FriendYouTube is testing a conversational AI search feature that lets users ask questions in plain language and receive answers drawn from videos, Shorts, and supplementary text. Google is calling it an "experiment," which is the word large companies use when they want to seem humble about something they are absolutely going to ship regardless. The pitch is that finding videos should feel like a conversation rather than a keyword guessing game — a reasonable ambition, assuming the AI recommends something better than whatever the algorithm currently thinks you want at 11pm. Source: The Verge
Someone Stole 4 Terabytes of Voice Recordings From 40,000 AI ContractorsMercor, an AI hiring platform that connects contractors to data labelling and training work, has suffered a significant breach. Roughly 4TB of voice samples — recordings made by some 40,000 contractors as part of their work training AI models — were stolen. The particular irony here is dense: the humans whose voices helped teach AI systems to speak have had those same recordings lifted by parties unknown. A data breach is always serious; this one has the added quality of being almost structurally poetic. Source: Oravys via Hacker News
|
