Discussion about this post

User's avatar
Nige Cook's avatar

I've just spent the best part of a week using Grok 3β AI for various science projects, having been pushed into it by Musk's X/twitter which now includes that AI system, and was surprised how slow it is to read technical mathematical books and reformulate proofs to order (taking several hours for hard problems). But that may have been due to multi-tasking (other people's requests), and it is nevertheless amazing to get any help at all with really abstruse physics.

It is approximately at the level of a mathematical physics graduate, but doesn't get "glazed over eyes" or fall asleep when asked a "blue skies" question. It digs in and makes an effort to come up with a new approach, although it does make mistakes and needs checking and corrections. The AI weaknesses are analyzing technical visual stuff like graphs and diagrams, which it can't even read from a PDF (had to upload screen prints as image files for it to analyse them).

It has certainly helped me understand some very complex mathematical proofs that were driving me nuts, and to re-formulate some of my own work in a way that you would expect from good peer-review, so my feeling is that AI can at present help to make tremendous rapid progress in quantum field theory. It's also good to see your own mathematical ideas taken apart and reassembled more clearly and simply by AI. If AI is to be let loose simplifying and enhancing all the calculations in all the dusty old books in mathematical physics libraries, bring it on!

Expand full comment
Phil Sutcliffe's avatar

Thanks Nika, that's very kind and helpful. At least they failed to nick two of my books. And FB has restored me, just like that! All unexplained and inexplicable, tho also astonishingly fast. But their bot said sorry so that's all right… Fraternals, Phil

Expand full comment
5 more comments...

No posts