Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Epoch confirms GPT5.4 Pro solved a frontier math open problem

epoch.ai

Post reply on HN
Print view View original

731–740 of 744 posts

Re: Epoch confirms GPT5.4 Pro solved a frontier math open problem

#731

Post by aspenmartin »

qnleigh wrote: post #597
qnleigh wrote: post #288

I am kind of amazed at how many commenters respond to this result by confidently asserting that LLMs will never generate 'truly novel' ideas or problem solutions. > AI is a remixer; it remixes all known ideas together. It won't come up with new ideas > it's not because the model is figuring out something new > LLMs will NEVER be able to do that, because it doesn't exist It's not enough to say 'it will never be able t…

I might as well answer my own question, because I do think there are some coherent arguments for fundamental LLM limitations: 1. LLMs are trained on human-quality data, so they will naturally learn to mimic our limitations. Their capabilities should saturate at human or maybe above-average human performance. 2. LLMs do not learn from experience. They might perform as well as most humans on certain tasks, but a human…

> 1. LLMs are trained on human-quality data, so they will naturally learn to mimic our limitations. Their capabilities should saturate at human or maybe above-average human performance.

Why oh why is this such a commonly held belief. RL in verifiable domains being the way around this is the entire point. It’s the same idea behind a system like AlphaGo — human data is used only to get to a starting point for RL. RL will then take you to superhuman performance. I’m so confused why people miss this. The burden of proof is on people who claim that we will hit some sort of performance wall because I know of absolutely zero mechanisms for this to happen in verifiable domains.

Re: Epoch confirms GPT5.4 Pro solved a frontier math open problem

#732

Post by sigmoid10 »

Measter wrote: post #496
sigmoid10 wrote: post #241

Earlier quoted context omitted.

We already know exactly what causes these bugs. They are not a fundamental problem of LLMs, they are a problem of tokenizers. The actual model simply doesn't get to see the same text that you see. It can only infer this stuff from related info it was trained on. It's as if someone asked you how many 1s there are in the binary representation of this text. You'd also need to convert it first to think it through, or use…

> It's as if someone asked you how many 1s there are in the binary representation of this text. I'm actually kinda pleased with how close I guessed! I estimated 4 set bits per character, which with 491 characters in your post (including spaces) comes to 1964. Then I ran your message through a program to get the actual number, and turns out it has 1800 exactly.

>I estimated 4 set bits per character, which with 491 characters in your post (including spaces) comes to 1964

And that's exactly the kind of reasoning an LLM does when you ask it about characters in a word. It doesn't come from the word, it comes from other heuristics it picked up during training.

Re: Epoch confirms GPT5.4 Pro solved a frontier math open problem

#733

Post by CamperBob2 »

jibal wrote: post #727
CamperBob2 wrote: post #716

Earlier quoted context omitted.

Indeed, and a lot of humans are very bad at it, as is clear from the comments I'm responding to. This is your idea of "conversing curiously" and "editing out swipes," I suppose. I've had this argument too many times ... it never goes anywhere. So I won't respond again ... over and out. A real reasoning entity might pause for self-examination here. Maybe run its chain of thought for a few more iterations, or spend som…

> Don't be snarky.

ROFL

Re: Epoch confirms GPT5.4 Pro solved a frontier math open problem

#734

Post by xpl »

qnleigh wrote: post #597
qnleigh wrote: post #288

I am kind of amazed at how many commenters respond to this result by confidently asserting that LLMs will never generate 'truly novel' ideas or problem solutions. > AI is a remixer; it remixes all known ideas together. It won't come up with new ideas > it's not because the model is figuring out something new > LLMs will NEVER be able to do that, because it doesn't exist It's not enough to say 'it will never be able t…

I might as well answer my own question, because I do think there are some coherent arguments for fundamental LLM limitations: 1. LLMs are trained on human-quality data, so they will naturally learn to mimic our limitations. Their capabilities should saturate at human or maybe above-average human performance. 2. LLMs do not learn from experience. They might perform as well as most humans on certain tasks, but a human…

> Their capabilities should saturate at human or maybe above-average human performance

LLMs do have superhuman reasoning speed and superhuman dedication. Speed is something you can scale, and at some point quantity can turn into quality. Much of the frontier work done by humans is just dedication, luck, and remixing other people's ideas ("standing on the shoulders of giants"), isn't it? All of this is exactly what you can scale by having restless hordes of fast-thinking agents, even if each of those agents is intellectually "just above average human".

Re: Epoch confirms GPT5.4 Pro solved a frontier math open problem

#735

Post by zar1048576 »

zozbot234 wrote: post #89
snemvalts wrote: post #65

Earlier quoted context omitted.

Math and coding competition problems are easier to train because of strict rules and cheap verification. But once you go beyond that to less defined things such as code quality, where even humans have hard time putting down concrete axioms, they start to hallucinate more and become less useful. We are missing the value function that allowed AlphaGo to go from mid range player trained on human moves to superhuman by p…

This is not formally verified math so there is no real verifiable-feedback aspect here. The best models for formalized math are still specialized ones. although general purpose models can assist formalization somewhat.

[dead]

Re: Epoch confirms GPT5.4 Pro solved a frontier math open problem

#736

Post by array_key_first »

virgildotcodes wrote: post #347
blackcatsec wrote: post #130

Earlier quoted context omitted.

> I often wonder why tech has so many reductionist, materialist, and quite frankly anti-human, thinkers. I think it comes from a position of arrogance/ego. I'll speak for the US here, since that's what I know the most; but the average 'techie' in general skews towards the higher intelligence numbers than the lower parts. This is a very, very broad stroke, and that's intentional to illustrate my point. Because of this…

It's funny because the fundamental thing I'm speaking out against is the arrogance of human exceptionalism. This whole debate about what it means to be intelligent or human just seems like we're making the same mistakes we've made over and over. Earth as the center of the universe, sun as the center of the universe, man as the only animal with consciousness and intellect, the anthropomorphic nature of the majority of…

The need for human exceptionalism doesn't come from the psyche or anything like that, it's just basic survival skills. Humans believe themselves to be special because that's the only belief that isn't self-destructive.

You can choose to believe humans are not exceptional, in the same way I can choose to cut off all my fingers and eat them. Why would I do that?

If what you say about LLMs is true, that's bad for me. And for you. And for our families. Because it means our instrinic value of living just went down a lot. I choose not to believe it because I am not suicidal. And, ultimately, I think the people who do believe it can only ever make their lives worse. Probably my life worse too, but maybe if I'm all the way over here I'll avoid the blast radius.

Re: Epoch confirms GPT5.4 Pro solved a frontier math open problem

#737

Post by qnleigh »

aspenmartin wrote: post #731
qnleigh wrote: post #597

Earlier quoted context omitted.

I might as well answer my own question, because I do think there are some coherent arguments for fundamental LLM limitations: 1. LLMs are trained on human-quality data, so they will naturally learn to mimic our limitations. Their capabilities should saturate at human or maybe above-average human performance. 2. LLMs do not learn from experience. They might perform as well as most humans on certain tasks, but a human…

> 1. LLMs are trained on human-quality data, so they will naturally learn to mimic our limitations. Their capabilities should saturate at human or maybe above-average human performance. Why oh why is this such a commonly held belief. RL in verifiable domains being the way around this is the entire point. It’s the same idea behind a system like AlphaGo — human data is used only to get to a starting point for RL . RL w…

I did mention RL as a valid counterargument in my comment.

I agree that in verifiable domains RL systems should be able to blow past human performance, and this might already be happening. There's another interesting question as to how much RL improves performance on non-verifiable domains. I'm not taking a stance either way, I just think it's an interesting question.

Re: Epoch confirms GPT5.4 Pro solved a frontier math open problem

#738

Post by Validark »

hnfong wrote: post #637
Validark wrote: post #205

Earlier quoted context omitted.

Fair point, however I am simply more interested in how AI can advance frontiers than in how it can transcribe a meeting and give a summary or even print out React code. I know the world is heavily in need of the menial labor and AI already has made that stuff way easier and cheaper. However I'm just very interested in innovation and pushing the boundaries as a more powerful force for change. One project I've been sup…

It's kind of interesting in your original comment you used the words "doubter" and "believer", as if AI was some kind of messianic event of some sort and you are deciding whether to "believe" in it. I mean, if you step back and think about it, there's nothing that requires faith. As you said, current AI can do a lot of things pretty well (transcribe and summarize meetings, write boilerplate code, etc.) Nobody is doub…

"it kinda takes away our fun of discovery"

It might, but that would be an incredibly awesome problem to have, wouldn't it? If we really had the infinite innovation printer, I'd hope we'd have a lot more fun at that point.

By "believer" versus "doubter" I mainly meant I see it as more than a just a next-word-predictor. But the religious language is probably appropriate nonetheless.

Re: Epoch confirms GPT5.4 Pro solved a frontier math open problem

#739

Post by Validark »

keybored wrote: post #329
Validark wrote: post #33

I have long said I am an AI doubter until AI could print out the answers to hard problems or ones requiring tons of innovation. Assuming this is verified to be correct (not by AI) then I just became a believer. I would like to see a few more AI inventions to know for sure, but wow, it really is a new and exciting world. I really hope we use this intelligence resource to make the world better.

> I would like to see a few more AI inventions to know for sure, but wow, it really is a new and exciting world. We already have a few years of experience with this. > I really hope we use this intelligence resource to make the world better. We already have a few years of experience with this.

What has AI discovered more than a year ago?

Re: Epoch confirms GPT5.4 Pro solved a frontier math open problem

#740

Post by int_19h »

gf000 wrote: post #680
int_19h wrote: post #642

Earlier quoted context omitted.

If you treat the human brain as a model, and account for the full complexity of neurons (one neuron != one parameter!) it has several orders of magnitude more parameters than any LLM we've made to date, so it shouldn't come as a surprise. What is surprising is that our brain, as complex as it is, can train so fast on such a meager energy budget.

You are right, but at the same time the human brain does way more stuff (muscle coordination, smell, touch sensing) and all those others take up at least some budget. So interesting question, but I'm not convinced it's only a scale issue. Like finished models don't really learn the same way as humans do - we actually change the parameters "at runtime", basically updating the model and learning is not only for the cur…

It goes both ways though. All that extra stuff is also a part of our "training set" when growing up. And we have already seen that training models on vision etc improves their text outputs as well, even in tasks that aren't directly connected to visual things. That might account for a lot of our advantages.

But yes, of course it's not just a scale issue. Note though that a "finished model" can still be fine-tuned, and you can in fact allow it to fine-tune itself even. It's just that this is prohibitively expensive in practice (once again, the hardware is lagging behind the wetware here).

Post reply on HN