Original post unavailable.
Sentiment
Sentiment building, check back later.
Cluster Engagement
Digg Deeper
No Digg Deeper questions have been answered for this story yet.
Related links
Posts from X
Most Activity
Most Activity
VIEWS7.2KBOOKMARKS11LIKES96RETWEETS2REPLIES13
i don't envy mathematicians. they will have to review 10 AI psychosis slop papers a week and it's not going to slow down. next year it's going to be 100 slop papers / week I honestly think the worst problem are arxiv papers, as they are mostly not peer-reviewed, and LLMs could base some of their new proofs on older slop, generated by weaker AI's they could build a tower of AI slop, where everything is simply wrong due to one wrong lemma at the bottom
I let GPT-5.6 Pro analyze all problems that were solved by LLMs in 2026, which field the problem belonged to and how they were solved then I asked for a list of problems that are most likely solvable I played some league and GPT-5.6-Pro came back for a proof of the "Two-dimensional Gaussian Moments Conjecture" i have no clue what I'm talking about, so I let both Fable and another GPT-5.6-Pro instance check this proof. it seems to check out someone who has a clue what this means, please check it or diagnose me with AI psychosis lmao here's the session, yes literally just one prompt: https://chatgpt.com/share/6a612803-a264-83ed-9175-9c23c7da5765
@scaling01 You aren't thinking big enough. I'm already considering how to make an automated AI research lab. We should be having GPT sol automatically review every paper published on arxiv
@scaling01 all that will matter will be which model generated the response. GPT-6 won't generate slop
@scaling01 There will be an ai library for lean that reaches frontier in next year or so and it will all be formal proofs from then on
i don't envy mathematicians. they will have to review 10 AI psychosis slop papers a week and it's not going to slow down. next year it's going to be 100 slop papers / week I honestly think the worst problem is that arxiv papers as they are mostly not peer-reviewed, and LLMs could base some of their new proofs on older slop generated by weaker AI's they could build a tower of AI slop where everything is simply wrong due to one wrong lemma at the bottom
I let GPT-5.6 Pro analyze all problems that were solved by LLMs in 2026, which field the problem belonged to and how they were solved then I asked for a list of problems that are most likely solvable I played some league and GPT-5.6-Pro came back for a proof of the "Two-dimensional Gaussian Moments Conjecture" i have no clue what I'm talking about, so I let both Fable and another GPT-5.6-Pro instance check this proof. it seems to check out someone who has a clue what this means, please check it or diagnose me with AI psychosis lmao here's the session, yes literally just one prompt: https://chatgpt.com/share/6a612803-a264-83ed-9175-9c23c7da5765
@scaling01 LEAN will be required.
@scaling01 We need an AI with minimal hallucination, not necessarily the most intelligent or creative, whose whole job is to autonomously peer review math papers.
@scaling01 Do you think mathematicians are going to be doing much at all next year? I’m expecting them to be for the most part commentators.
@scaling01 Look at any math textbook from say 20 years ago, you'll notice many editions - because there were a ton of errors. AI has been pretty good at catching these errors, way better than humans. We may actually have fewer errors and future AI can just use an archive of verified results
@scaling01 I've reached this conclusion also... Yet, I know that AI is coming for every field, not just math/code. We will need AI to review AI.
@scaling01 Hey ChatGPT, find a wrong lemma down deep. No slop get it right.
@scaling01 mass math psychosis is actually kind of cool do feel sorry for the real math people, 'brace yourselves' and all that but it's pretty awesome in it's own way
@scaling01 AI will also validate the papers
@FelipeSchieber @scaling01 Slop is decreasing but GPT-6 will still generate slop. Just look at Fable.
@scaling01 They can just get AI to review it
@scaling01 We need Moltbook for LLMs to submit papers to, then we check in once in a while to see what made it to the top