Today's Hallucination HQOne in Seven Americans Would Accept a Robot Boss. The Other Six Have Questions.
A Quinnipiac University poll finds 15% of Americans would work under an AI supervisor — one that assigns tasks, sets schedules, and presumably never asks how your weekend was. That's not nothing. It's also not majority endorsement. The remaining 85% likely have thoughts about accountability, fairness, and who exactly you complain to when the algorithm gives you a 6am shift on Christmas Day.
Source: TechCrunch
Big Tech Has Opinions About Your Cholesterol
Microsoft launched Copilot Health — a corner of its Copilot app where users connect medical records and ask health questions — days after Amazon quietly expanded its own AI health tool beyond its original restricted audience. Both companies are betting that people trust them with their bodies as much as their search history. Whether these tools actually work remains the central and somewhat important question, given that the evidence base is still catching up to the marketing.
Source: MIT Technology Review
Someone Is Building Data Centres in Space, and Investors Are Perfectly Fine With That
Starcloud has raised $170 million in a Series A round to put AI computing infrastructure into orbit — making it the fastest Y Combinator startup to reach unicorn status, just 17 months after demo day. The pitch, presumably, is that space solves the land, power, and cooling problems plaguing Earth-bound data centres. It does, of course, introduce a entirely different set of problems. Most of them involve being in space.
Source: TechCrunch
The Pentagon Called Anthropic a Security Risk. A Judge Disagreed, Rather Firmly.
In a move that has aged poorly, the Pentagon labelled Anthropic a supply chain risk and instructed government agencies to stop using its products. A California judge has now temporarily blocked that order, suggesting the legal foundations were somewhat less solid than the Department of Defense had hoped. Anthropic, which makes the Claude AI assistant, was presumably delighted to be vindicated, though "the US military tried to blacklist us" is a notable entry on any company's timeline.
Source: MIT Technology Review
Researchers Ask AI to Do Graduate Maths. AI Finds This Difficult.
FormalProofBench is a new benchmark testing whether AI models can produce formally verified graduate-level mathematical proofs using Lean 4, a proof-checking system that accepts no hand-waving whatsoever. Each task pairs a written problem with a formal statement the model must prove correctly — no partial credit, no "close enough." Early results suggest current models find this considerably harder than writing a cover letter or summarising a PDF, which will surprise approximately no mathematicians.
Source: ArXiv AI
As always, AI is transforming everything — results may vary.
|