No "AI" of any description is doing novel proofs at the moment. Not o3, or anything else.
LLMs are good for chatting about basic intuition with, up to and including complex subjects, if and only if there are publically available data on the topic which have been fed to the LLM during its training. They're good at doing summaries and overviews of specific things (if you push them around and insist they don't waffle and ignore garbage carefully and keep your critical thinking hat on, etc etc).
It's like having a magnifying glass that focuses in on the small little maths question you might have, without you having to sift through ten blogs or videos or whatever.
That's hardly going to replace graduate students doing proofs with professors, though, at least not with the methods being employed thus far!
No "AI" of any description is doing novel proofs at the moment. Not o3, or anything else.
LLMs are good for chatting about basic intuition with, up to and including complex subjects, if and only if there are publically available data on the topic which have been fed to the LLM during its training. They're good at doing summaries and overviews of specific things (if you push them around and insist they don't waffle and ignore garbage carefully and keep your critical thinking hat on, etc etc).
It's like having a magnifying glass that focuses in on the small little maths question you might have, without you having to sift through ten blogs or videos or whatever.
That's hardly going to replace graduate students doing proofs with professors, though, at least not with the methods being employed thus far!