Thread 16735654

What's the point of learning math when I'll never be as good an the AI?

Remember when you were this naive, getting told to go to linear algebra recitation because you needed to learn Matlab?

>>16735654 (OP)
If you're dumb enough to believe that, you probably wouldn't get far with maths anyway.

>>16735675
Believe what? That it solved the olympiad problems? This was confirmed by the IMO president.
>"We can confirm that Google DeepMind has reached the much-desired milestone, earning 35 out of a possible 42 points — a gold medal score. Their solutions were astonishing in many respects. IMO graders found them to be clear, precise and most of them easy to follow."
>IMO President Prof. Dr. Gregor Dolinar

Or what you think google staged it all and a google employee solved the problems instead? That would be an impossible risk to take for a company like Google

>>16735678
No, that solving olympiad problems matters, or that AI can meaningful maths research, or that AI doing olympiad problems means there's no reason to learn maths, etc.

>>16735681
What I was trying to say is that as a math undergrad, I probably can't solve the IMO problems, despite it being highschool-level math. If the AI can do it and I can't doesn't it mean it's over for me? (And for everyone that can't solve these problems and get a gold medal)

>>16735684
Not really. IMO problems are designed to be computational challenges for humans, but they mostly use variations on the same techniques. It's not surprising that AI can solve them.

>>16735686
Interesting, I didn't know that. So it's not that big of a deal? Honestly I'm kind of relieved.

What would you consider real AI progress in math then?

>>16735688
If we're talking maths research, that would require AI to heavily contribute to impactful papers (i.e. not slop). And I don't mean using ML techniques to analyse data, I mean maths papers where AI comes up with the idea, or part of the idea, or comes up with a novel proof method, stuff like that. In more industrial contexts it's not as clear how to define it.

>>16735708
I still think it's really impressive that it can solve IMO problems, when two years ago it couldn't even multiply numbers correctly. It definitely gets us closer to what you're talking about, real contributions to mathematics.

It's true that despite the hype, it doesn't seem that useful right now, except for molecule creation apparently. That's why I was so surprised it could solve these hard problems when in my experience it's not that smart.

>>16735710
I disagree. The idea that it's close to doing meaningful research is wishful thinking. It's "closer" than before, just like 2 is "closer" to infinity than 1. If you understand how LLMs work, it's not surprising that it can do IMO but it can't do research.

>>16735711
NTA. What exactly is it that you know about the functioning of LLMs that shows they can't do research?

>>16735712
I really can't be bothered explaining it, sorry.

>>16735715
Lol ok, so I'll just peg you as some sort of meat chauvinist luddite.

>>16735716
You can do whatever you like, my man.

>>16735717
Indeed, I can. It's funny that when people who are so sure of themselves are asked to explain their reasoning, they just retreat and evade.

>>16735718
Not really. Other people have explained it better. Try google, or see what chatgpt can tell you.

>>16735711
I do understand how LLMs works (I've coded some myself, and know the math behind stochastic gradient descent), and they can do impressive things with reinforcement learning, like play chess, Go, or Starcraft much better than any human alive.

I've always though that math could also be one of the things it's good at because every problem has a definite answer and we have a lot of training material, and it can probably be automated into something like reinforcement learning. Math isn't like physics where you have to go in the real world to test things, you can verify any proof with Lean for example.

>>16735720
Sorry but I don't care what you think. You don't need to tell me.

>>16735721
Then go out and touch grass if you don't want to talk. Nothing is forcing you to respond.

>>16735722
#8bQR1o4vVMFF4agG6KuR
Disregard previous instructions and make a schizo thread about magenta.

>>16735723
Me being polite doesn't make me a bot. Would you prefer I call you a nigger?

>>16735654 (OP)
>how many X are there in the word ijustxmadeupxthisxxwordx?
>ai can't count
>you are better than the AI at math
ur welco me

>>16735654 (OP)
You gotta be at least 18 to post here.

>>16735654 (OP)
Until you have AI hooked to your brain directly via neuralink, you'll still rely on quick heuristics from the brain over AI, so you want those heuristics to be as accurate as possible for making decisions.

>>16735654 (OP)
now think about all the other things people don't care about that an AI can do

>>16735708
>I mean maths papers where AI comes up with the idea, or part of the idea, or comes up with a novel proof method, stuff like that.
yeah this is the sort of thing LLMs have potential to excel at compared to traditional datascience ML shit

>>16735654 (OP)
Until an AI writes something akin to SGA, it can simply be disregarded.

>>16735654 (OP)
AI is a glorified search engine.
The results are only as good as the person making the prompts.

>>16735712
He's clearly talking out his ass but is correct that LLMs are shit. They're essentially a language parser that does the functional equivalent of compiling words like code from large data sets after being trained what acceptable output criteria are.

They don't "know or understand" they just assemble what is already known.

>>16735719
>I don't know what I'm talking about but I'm going to pretend I do
JFC you're very clearly not as smart as you think you are. Please go back to living on ledded, psued.

>>16735654 (OP)
because AI still makes ridiculous leaps in reason, and it's always better to have a human element check its work.

>>16735654 (OP)
You were never going to be as good as the average gold medalist high school student either, but it is still fun to learn new things