Today's Hallucination HQMark Zuckerberg Is Building a Mark Zuckerberg
Meta is training an AI clone of Mark Zuckerberg — his voice, image, mannerisms, and tone — to attend meetings and give feedback to employees on his behalf. This raises several questions, chief among them: how will anyone tell the difference? The project reportedly aims to scale Zuckerberg's presence across a company of 70,000 people, which is either visionary leadership or the plot of a moderately unsettling sci-fi film. Possibly both.
Source: The Verge
OpenAI Would Like to Help You Spend Your Money. After Spending Quite a Lot of Its Own
OpenAI has acquired Hiro, an AI-powered personal finance startup, signalling that financial planning is the next capability being folded into ChatGPT. The irony of a company that has burned through billions in pursuit of artificial general intelligence now offering budgeting advice is, we feel, worth sitting with for a moment. No acquisition price was disclosed, which is probably for the best.
Source: TechCrunch
OpenAI Sends Four-Page Memo Reminding Everyone That Competitors Exist
Chief Revenue Officer Denise Dresser sent employees a Sunday memo — four pages, which is two more than most people read — outlining OpenAI's strategy to retain users and grow its enterprise business, with Anthropic named explicitly as a threat. The timing is notable: OpenAI is simultaneously expanding into finance, cloning rivals' customers, and apparently spending weekends sending strategic literature. Rest, it seems, is for the competition.
Source: The Verge
Stanford Confirms That AI Experts and Normal People Are Not Having the Same Conversation
Stanford's 2025 AI Index finds a widening gap between how AI insiders view the technology and how the general public feels about it — with rising public anxiety around jobs, healthcare, and the economy. Experts tend toward optimism. Everyone else is rather less convinced. This is, historically, not a dynamic that ends with the experts being right and everyone else saying "fair enough, well done."
Source: TechCrunch
An AI Has Aced the LSAT. Law School Applications to Follow, Presumably
Researchers have documented the first language model to achieve a perfect score on an officially administered LSAT — the exam required for US law school admission. Controlled tests across eight reasoning models confirmed the result holds under varied prompts and shuffled answer choices, meaning it wasn't a fluke. Lawyers, who spent three years and considerable money learning to think like this, are invited to feel however they'd like about that.
Source: ArXiv AI
As always, treat this newsletter as informational. Do not ask it to manage your finances, attend your meetings, or sit the bar exam.
|