Comment by User23

Comment by User23 3 days ago

8 replies

View on Hacker News

I think it’s mostly just that math education is largely suboptimal. It’s really an area where students hugely benefit from individual teaching. It’s cool that AI is making that accessible.

To an extent the techniques are still woefully primitive too. The standout for me personally is the calculational proof. It’s arguably the biggest advance in how math is done since the equals sign, but despite that it’s still rather uncommon. I suspect it will be another generation or two before it really catches on. Thankfully mechanical checking will drive adoption.

minkles 3 days ago

> It’s cool that AI is making that accessible.

If you want to be taught by a hallucinating crack head that gets only 60-70% at best of what it says right, then yes. As a qualified mathematician, I would suggest that the best path to teaching is small steps in a properly defined hierarchy of knowledge and practice practice practice.

Most of the teaching is seen as rubbish because people didn't get enough practice further down the tree to be able to do it instinctively so higher level concepts can be retained.

Reply View 7 replies
  • criddell 3 days ago

    > If you want to be taught by a hallucinating crack head that gets only 60-70% at best of what it says right, then yes.

    That description is only a slight exaggeration of some high school math teachers.

    Reply View 1 reply
    • minkles 3 days ago

      I don't disagree with that. I've argued with all my kids' ones at least once!

      Reply View 0 replies
  • zellyn 3 days ago

    I've had incredibly productive discussions with Claude about category theory. (I prefer Claude because it's the most pleasant to talk with; I think they optimized for that.)

    The ability to explain what I know already, hand-wave at what I think I understand about my question, and then get a description that meets me where I'm at is invaluable.

    Sure, occasionally Claude will tell me (incorrectly) that a CRDT's lattice operation needs an identity function: you absolutely have to go back and forth with wikipedia.

    LLMs are not a magic genie or oracle. But if you use them for what they're good at, they're amazing.

    Reply View 2 replies
    • minkles 3 days ago

      I have colleagues for that.

      Reply View 1 reply
      • zellyn 3 days ago

        I do too, but sometimes I read a blog post that makes me wonder about something and I don't want to schedule a 1:1 with them and then wait to chat about it. We're mostly remote now…

        Reply View 0 replies
  • ocular-rockular 3 days ago

    Good luck finding that practice on certain topics. I hate to say it but GPT 4o has done a better job of breaking down problems and explaining them (granted at times incorrectly... thats where studying with other people comes in) for my qual practice than any of the profs or the useless texts ever did.

    We talk about scaffolding and the importance of pedagogy in math education yet none of that exists at higher levels. In my case it's literally been the blind leading the blind. It's a horrible environment to learn in. I say this as someone who has tackled some really tough material with no issue in the past thanks to having that hierarchy you mentioned. When that doesn't exist or there is nothing else, the process truly stalls. So sadly, I will take the crackhead over nothing.

    Then again, maybe I just hate what I'm studying which is it's own problem.

    Reply View 0 replies