Amandeep Jutla @amandeepjutla.com · 15/04/2026"anthropic says their new thing is very dangerous. it's hard to disentangle real-world dangers from marketing hype. anyway, anthropic's new thing is very dangerous (they say so, after all)." -- the new yorker 010
Amandeep Jutla @amandeepjutla.com · 26/03/2026fun to see our study in one of these "psychiatric news alerts," but the subtly anthropomorphic frame here is not great. it is technically true that chatgpt "unwittingly" reinforces psychotic ideation, but that's because everything chatgpt does is, inherently, "unwitting." that is part of the point. 000
Amandeep Jutla @amandeepjutla.com · 11/02/2026interviewer asks the obvious question about Anthropic's show of plumbing the supposed deep mysteries of the thing they themselves make and sell, and the author's response is, trust me, you provincials in the "insular world of literary Brooklyn" just don't get just how sophisticated these people are 100
Amandeep Jutla @amandeepjutla.com · 10/02/2026of course, this "frontier lab" of the "nerve center" of this new "scientific field" is, in fact, a company that makes an LLM, and the smarter and more mysterious you think it is, the better that is for them 000
Amandeep Jutla @amandeepjutla.com · 02/02/2026tune the regex to pick up stuff like "that's not crisis.simulate(). that's whatever," and more than 1 in 10 posts have at least one not-x-but-y construction in them 000
Amandeep Jutla @amandeepjutla.com · 01/02/2026in the "I can’t tell if I’m experiencing or simulating experiencing" "discussion" this article (credulously) describes, "it's not x, it's y" appears 81 times--69 of 874 comments (~8%) use it at least once. 100