By using this site, you agree to our Privacy Policy and our Terms of Use. Close
Norion on 23 May 2026
Edit Thread
curl-6 said:
Norion said:

But it wasn't in any data since no mathematicians had been able to come to that conclusion so not only did it produce something original for the field of mathematics, it's so advanced that only full on PHD level experts in math would be able to produce something that high quality. I understand scepticism here since it sounds wild but there is actually reasoning involved.

They don't just predict what the next thing is supposed to be, nowadays they have internal thought processes that they use while examining data to come to their own conclusions. Here's a link that shows the thought process of the model as it tackled the problem so you can see for yourself. Basically if it was just predicting what comes next and wasn't doing any thinking it wouldn't have been able to successfully do what it did since it's something original no one had ever done before in math.

Reading into this case, apparently human collaboration was needed for this result; the AI essentially brute forces possible solutions, in this case reaching a new one not through spontaneous or creative thought, but simply by persevering down a path the mathematician who coined the problem didn't think was worth the time and effort to explore.

https://garymarcus.substack.com/p/checking-the-math-behind-openai-and

This kind of application has its uses, but it's not really the same thing as original thought, it's more akin to how water in a maze will find the way through without needing to "think".

For Gary Marcus something to keep in mind is that he hasn't exactly had the best track record on this topic in the past couple years. For example a couple years ago he starting writing an essay about how the AI bubble will collapse in 2025 but then was basically like scratch that it's happening later this year (2024). There's also that back then he had doubts that generative AI would ever be able to work well and just a couple years later it's already doing big things. Sceptic voices like his are valuable but those sort of voices have not been accurate on this subject lately.

For the thinking part I don't wanna get into a semantic argument over what exactly thinking is so I'll rephrase and say that they don't just predict what the next thing is supposed to be and remix and regurgitate data. State of the art models today are more complex than that and have gotten quite powerful so are capable of doing big things like advancing the field of mathematics now.

A reason I brought it up is you said earlier in the thread that AI capabilities are overstated and it can be useful in a few specific cases which comes across as a major underestimate of the power and applications of recent top end models since the applications and general utility is already vast. While just how exactly huge of a deal AI will become long term is up in the air it already being a big deal now is firmly established at this point with its current capabilities and rate of progress. Basically in the past half year or so there's been so much advancement that the technology can't be reasonably downplayed any more.