AI just solved one of the hardest math problems… (INTELLIGENCE EXPLOSION)

43 thoughts on “AI just solved one of the hardest math problems… (INTELLIGENCE EXPLOSION)”

1. if the prompt(s) isn't (aren't) shared, nor the full chatlog, it's bullshit.
   nowhere is that proof shown or linked. so ppl can duplicate that same proof.
   that's the real proof: show it so ppl can verify it and repeat it.

   the graph from 1:20 is also bullshit, it's talking about compute btw, not about smartness or intelligence, but pure about compute. ofc can a computer compute more and faster than a person but that doesn't make it intelligent and understand what it is calculation/computing. chatGPT5 isn't even clsoe to the intelligence of a preschooler so we're nowhere near any real AI and a lot further from any AGI let alone ASI.

2. the repeated slight jump shifts in the visual video makes this unwatchable. Obv this is not the only youtube video using this "enhancement". Please get rid of it. Your content is interesting enough that I can live with real-time feeds not sped up

3. Every year I loose my trust in Terence a bit more 😢

4. AI just need to be really smart, cause in terms of speed, AI is million times faster than humans.

5. Solving problems humans cant solve only means it's true intelligence! That's extremely important!

6. Anna Dijkman: "Pointing to the enemy (the AG1-mathematics of Stichting Intuïtie Trainen, KvK 17210108) as a common goal to strengthen group cohesion instead of fighting amongst themselves. But does it work? Because what exactly needs to be protected: Greenland or NATO? And who precisely is the enemy (as long as DeBr0nBescherm1ng remains in place)? For Greenland, there is little concrete threat from Russia or China, but all the more from Donald Trump. A common enemy can force cooperation, but only if shared interests (in the virtual debt economy) are at stake. And with zero-sum thinker Trump, that is becoming increasingly difficult."

   Dav1D Friedberg: "A sudden resignation inside 0penA1 is raising serious questions about what’s really happening behind closed doors. When an insider walks away and says "1T’s no longer just about AG1-math — it’s about JTS (Just Truth Sharing), power, and who controls the future of $Trump 2.0. In this video, we break down what this exit could signal and how growing internal tensions in Epstein's network could shape the next phase of JUST TRUTH MANGL1NG in ways the FED is N0T prepared for."

   Femmetje de Wind: "Real estate developer Steve Witkoff travels the world together with Trump's son-in-law, Jared Kushner, to broker peace deals in places no one thought possible. The striking thing: the duo has no in-depth knowledge of the conflicts they are mediating and no classical diplomatic experience. However, they possess an enormous network of commercial contacts outside the US and have direct access to power. Kushner, by the way, doesn't call those foreign contacts 'conflicts of interest', but 'experiences and valued relationships' (from Epstein) that help close deals. That chapter is missing from the Harvard method (to which the ECB is held)."

7. “He just solved it” and the same time 100 people can solve it aka it has been solved already.

8. for those who are comparing their phone's childish GPT for maths with these new advance things:

   Because u are using their baby only available to public, the real parents if they release them, u will see only destructions as common man don't think of solving math but misusing immense power.

9. Idiocy explosion

10. It's only the billionaire class and the big silicon valley corps are exciting about racing humanity to its doom. Ask the average person on the street what the care about and how they actually feel. No average person in the real world are excited about where the billionaire class and greedy corps are pushing us. Actually, their hubris and greed are punishing all of us – the little guys who just want to live a life in peace.

11. talk to me when it solves something people didn't yet

12. Matthew, I am impressed but not overwhelmed. Many questions in science (including mathematics) lie on the edge of solution. Knowledge of most facts in a field makes solving these problems relatively simple and requires no new insights. Some of these questions have been solved but the solutions lie in obscurity. Tieing off these threads is a valuable way to advance the edges of scientific knowledge. However, major advances often require new insights, something that knowledge of existing facts does not impart. In fact, throughout history, knowledge and belief in existing "facts" has often impeded the development of new insights.

13. Would love to hear follow-ups

14. one small problem. the AI needs to know you better than you know yourself 😅

15. How is solving problems related to inventing new problems or new discoveries?

16. This is cool and impressive, but it sounds like GPT 5.2 drew upon existing solutions published in obscure parts of the math literature to come up with the solution. Impressive recall abilities, but maybe not a truly creative and brilliant leap?

17. Let me know when it invents new conceptual frameworks.

18. "and then after that, it's Completely Unbounded Intelligence!" — Wrong.
    Or as the British would say "Let's not get carried away."

19. 2:17 To clarify, the International Mathematical Olympiad (IMO) is not a coding contest, but rather a proof-writing contest with pencil and paper.

20. "It is one hardest open math problems"

    I wonder if this guy deliberately says stupid shit as engagement bait?

21. You have swollen under-eye bags. You can do something about it.

22. yes, for sure everone will benefit. There is no chance rich will get richer and poor will get poorer, totally impossible

23. But is this a problem that is already solved and just tricky to work out for a human? So it could be in the training data or is it novel like proving P = NP or something, that is currently unsolved. The first result is not a big deal, the second would be amazing.

24. It was still quided by a human. TT also was asked, if it was significant enought for a publication. Not quite. To a survey publication as a mention maybe. When monkeys have enough time, they can type a book, just by accident.

25. 2:00 is it just me or is the skill here the prompt. Meaning AI still can’t just be given the math problem and say “solve this”.

    It took a skilled mathematician to sort of coerce the answer out of ChatGPT.

26. I don't think it will ever get automated, hopefully, since by that point were fucked

27. This has already been debunked by the math community. What actually happened in this case and other Erdos problems still considered unresolved is that AI dug out old papers from hidden places no one had looked at that actually had resolved the problems. And the heavy work has not been done by ChatGPT but by Harmonics Aristotle, which is AI specialized on Math and using LEAN for verification of proofs.

28. This is not one of the HARDEST problems — this and other Erdos problems are not looked at by many people simply because they are not that interesting or important. This is NOT like Fermat Last Theorem, or Goldbach Hypothesis, or any of the 23 Hilbert Problems. If AI can solve Riemann Hypothesis, then we will all agree it is revolutionary.

29. Not surprising. Math is about patterns and deep neural networks (not necessarily LLMs) are better at patterns.

30. Many of Erdos problems are there because he just did not have the time to work on them.
    Most mathematics problems are hard because they involved specialist knowledge. There are many specialties in mathematics.

31. Turns out the solution was found already

32. The use of "open problem" can be misleading: these are "open problems" in the sense of being questions which have been asked somewhere and haven't been answered, but there is no evidence that anyone seriously thought about them. It is not only that they are hundreds of tiers below famous open problems (stuff like Riemann Hypothesis, Hodge Conjecture, Goldbach Conjecture, P vs NP, and whatever the public has heard of), they are also hundreds of tiers below thousands of serious open problems that experts in the respective areas have thought about for a long time and don't know how to approach.

    I'm not saying AI won't ever contribute on that level, but so far it hasn't solved anything humanity as a whole devoted substantial effort to and failed … not even close. However, I think it is already becoming a somewhat useful assistant.

33. Super interesting!

34. Breaking news: calculator can multiply numbers of several digits in less than a second! Calculators are super intelligent! Oh wait a minute…

35. Yet it could not fill out a form for me today 😂

36. c = 4a^2 + 4a + 1 = (2a+1)^2

37. Most AI can't even solve 1+1 without sounding overly verbose, and even getting it wrong most of the time. And at the same time, they can solve some of the most complicated equations of all time. It's all rather baffling.

38. If 5.2 is what OpenAI has public facing, their internal models must be insane.

39. There are many open math problems, because they are inconsequential.

40. And hence the Matrix is born

41. i solved all millenial problems with my AI.

42. I'm all for AI and the singularity and stuff, but the Matrix Multiplication "breakthrough" was not important.

43. He just used AI as a calculator. Why are you glorifying the calculator?

Comments are closed.