The only person I think has any sort of reasonable view is Leo de Moura [0]:
> This is amazing. I mean, even if nobody can understand initially ... people would search, breaking it into pieces. People would ... isolate fragments, people would invest a lot of time trying to understand these objects.
To me this is the obvious take on such a result. It's so disheartening to hear the the smartest mathematicians on earth be so cynical and dismissive.
> As I perceive it, the discontent is that we got the proof, but no understanding was gained in the process.
The process has only just begun: people will be studying this proof for years to come, they’ll find ways to reformulate it to be simpler and easier for humans to understand; novel techniques will be identified lurking within it, which will be named, categorised and disseminated.
This is.. a really stupid claim on the face of it.
The proof came out barely over a day ago. Let people have some time to pick it apart before clamoring to say we've learned nothing from the proof. Ridiculous.
Whats prevents the understanding now that the AI has given a solution? Why wouldnt studying a “known good” solution drive understanding faster than continuing to hit your head against the wall as you hope to figure it out?
Ive been really disappointed by what Ive read about the maths community lately. It all seems extremely narrow minded and gatekeepy. Instead of rejoicing that a problem we have struggled with for so long is now solved and moving on to the next challenge, we lament that we didnt solve it “the right way” and that its not fair that people wasted their time trying to so it themselves.
Which, honestly, it sounds like the entire thing wasnt even a huge deal anyways other than the prestige of it. Im a complete foreigner to the math community, but I would have expected more excitement at what the downstream impact is now that we have a proof for this. But it turns out we already assumed it was true and so not much has been unlocked. So that seems like a pretty good indication that people spending so much time on proving this likely were not being terribly productive in the first place in terms of choosing something worthwhile to spend their time on.
interesting point from this video (much more interesting than the sycophancy of 100s of comments in the other, humongous, thread):
part of the promise of AIs is that they'd be able to explain these proofs to us, rather than simply construct massive complex webs of incomprehensible proof.
"AI can make these enormous proofs but can't explain them to us" feels to me like yet another one of those "AI will NEVER [x]" predictions which gets falsified six months later when a larger comes out and does [x] just fine. I think it's silly to be either hopeful or distraught about the possibility of there being oodles of AI proofs that are forever incomprehensible to humans, this current moment is probably a short-lived anomaly.
I didn't say anything like that, I something more akin to "since AIs are good at explaining so many things, where's my explanation for this" and the answer-in-my-head was "because for commercial competitive reasons, anthropic and openai are trying to beat each other to proofs without accidentally helping each other"
What makes you think humans can understand it all? I don’t think that’s the promise of AI. No human living can do these and yet we think humans must understand. (Whether we should or not is a different question)
The idea that advanced machines would be our teachers is older than computing so it’s safe to say that it’s been a common idea.
That said, there’s no notion in philosophy of science, or computational epistemology that there’s a normative direction or limit that machines can effectively share the same level of complex understanding.
In fact i would argue that a system demonstrating a proof that humans can repeat but not understand is precisely what you would expect from a system more capable.
If it looks like magic while being independently reproducable then you have a more capable system than you have the ability to fully evaluate.
The only person I think has any sort of reasonable view is Leo de Moura [0]:
> This is amazing. I mean, even if nobody can understand initially ... people would search, breaking it into pieces. People would ... isolate fragments, people would invest a lot of time trying to understand these objects.
To me this is the obvious take on such a result. It's so disheartening to hear the the smartest mathematicians on earth be so cynical and dismissive.
[0] https://www.youtube.com/watch?v=zqzk43bSdDI&t=137s
As I perceive it, the discontent is that we got the proof, but no understanding was gained in the process.
Bullshit. It's very well structured
> As I perceive it, the discontent is that we got the proof, but no understanding was gained in the process.
The process has only just begun: people will be studying this proof for years to come, they’ll find ways to reformulate it to be simpler and easier for humans to understand; novel techniques will be identified lurking within it, which will be named, categorised and disseminated.
Indeed - hasn't this happened to Wiles's FLT proof ?
This is.. a really stupid claim on the face of it.
The proof came out barely over a day ago. Let people have some time to pick it apart before clamoring to say we've learned nothing from the proof. Ridiculous.
Whats prevents the understanding now that the AI has given a solution? Why wouldnt studying a “known good” solution drive understanding faster than continuing to hit your head against the wall as you hope to figure it out?
Ive been really disappointed by what Ive read about the maths community lately. It all seems extremely narrow minded and gatekeepy. Instead of rejoicing that a problem we have struggled with for so long is now solved and moving on to the next challenge, we lament that we didnt solve it “the right way” and that its not fair that people wasted their time trying to so it themselves.
Which, honestly, it sounds like the entire thing wasnt even a huge deal anyways other than the prestige of it. Im a complete foreigner to the math community, but I would have expected more excitement at what the downstream impact is now that we have a proof for this. But it turns out we already assumed it was true and so not much has been unlocked. So that seems like a pretty good indication that people spending so much time on proving this likely were not being terribly productive in the first place in terms of choosing something worthwhile to spend their time on.
interesting point from this video (much more interesting than the sycophancy of 100s of comments in the other, humongous, thread):
part of the promise of AIs is that they'd be able to explain these proofs to us, rather than simply construct massive complex webs of incomprehensible proof.
"AI can make these enormous proofs but can't explain them to us" feels to me like yet another one of those "AI will NEVER [x]" predictions which gets falsified six months later when a larger comes out and does [x] just fine. I think it's silly to be either hopeful or distraught about the possibility of there being oodles of AI proofs that are forever incomprehensible to humans, this current moment is probably a short-lived anomaly.
I didn't say anything like that, I something more akin to "since AIs are good at explaining so many things, where's my explanation for this" and the answer-in-my-head was "because for commercial competitive reasons, anthropic and openai are trying to beat each other to proofs without accidentally helping each other"
What makes you think humans can understand it all? I don’t think that’s the promise of AI. No human living can do these and yet we think humans must understand. (Whether we should or not is a different question)
The idea that advanced machines would be our teachers is older than computing so it’s safe to say that it’s been a common idea.
That said, there’s no notion in philosophy of science, or computational epistemology that there’s a normative direction or limit that machines can effectively share the same level of complex understanding.
In fact i would argue that a system demonstrating a proof that humans can repeat but not understand is precisely what you would expect from a system more capable.
If it looks like magic while being independently reproducable then you have a more capable system than you have the ability to fully evaluate.
>I don’t think that’s the promise of AI
google would not replace search with AI if AI wasn't already offering what can only be called explanatory text?
>What makes you think humans can understand it all?
what makes you think humans can't understand a lot more about this proof than they do so far?
don't answer, I was just illustrating the opposites phrased in your favorite style hoping it would sound hollow
Are there really no women who do number theory?