Post by @ZachWeinersmith@mastodon.social

2026-07-29 12:03 UTC

I am personally very interested in mathematicians right now as a kind of vanguard for ai-based replacement of intellectual labor. If you haven't kept up, regular consumer-facing ai has now resolved a number of important open problems, and the pace appears to be picking up.

Replies (4)

@Quoidian@mstdn.ca 2026-07-29 12:12

@ZachWeinersmith@mastodon.social I am reminded of the tale of the blind men and the elephant.

Open ##4196937
@drgroftehauge@sigmoid.social 2026-07-29 12:44

@ZachWeinersmith@mastodon.social Except when the AI exploits bugs in the proof software as we have seen recently...

Open ##4334666
@synlogic4242@social.vivaldi.net 2026-07-29 12:57

@ZachWeinersmith@mastodon.social I've seen LLMs make basic math mistakes many times, in direct interaction. I would never trust them with anything important. LLMs fundamentally cannot think or do math, they just make Bayesian bets and what might make user happy, but will make claims with perfect confidence that has a brazen error in it somewhere. thankfully most of modern math seems trivial and useless haha, and the truly useful parts are already solved and have software libraries or apps for calcs. my impression to date

Open ##4334670
@ardouglass@mastodon.social 2026-07-29 13:09

@ZachWeinersmith@mastodon.social I’m not fully convinced this isn’t the result of training data containing advances academics haven’t seen. https://xkcd.com/664/

Open ##4334671