The headline is broadly defensible, but the qualifications matter. Effect sizes vary by population, the strongest claims rest on shorter trials, and credible voices push back on how it's typically framed.
How much do you use AI? Do you customize the AI to be more effective for what you want? So this is what's really scary is like this is what people call prompt engineering, right? Do you train the AI to give you more effective answers? um do you use the AI for mental health issues? And do you find that the AI's answers are far superior to humans? And these are the these are four of the seven proposed risk factors for like bad outcomes from AI.
Every Sunday: the week’s new conflicts and verdict changes — and nothing else.
Native comments, Twitter mentions, and Reddit threads about this claim — surfaced together so the conversation isn't fragmented across platforms.
Bookmarking — the dossier-vs-overview split is the right call. Most of the time I want overview; sometimes I want receipts.
Would love a "what would change this verdict" RSS feed. Sign me up if it exists.