x.com/Ognifedefingo/status/2103032919988330723
1 correction found
AI can now solve any math problem a human can solve
This is too broad to be true. Frontier AI systems still fail on some math problems that humans do solve.
Full reasoning
The claim is universal: it says AI can solve any math problem a human can solve. Official results from leading AI labs contradict that.
- Google DeepMind’s IMO 2024 system explicitly did not solve all six IMO problems. DeepMind says AlphaProof/AlphaGeometry 2 solved four out of six problems, and that the two combinatorics problems remained unsolved.
- The official IMO 2024 results page shows that human contestants did solve those supposedly unsolved-by-AI problems: for example, multiple contestants received a full 7 on Problem 3, and many received a full 7 on Problem 5.
- OpenAI’s own 2026 First Proof writeup also shows current frontier AI is not universally reliable on human-solvable research math. OpenAI says a relevant academic department could conceivably solve many of the challenge problems in a week, yet also says one of its own proof attempts (problem 2) was later judged incorrect.
So there are clear counterexamples: math problems that humans solved, but leading AI systems did not. That makes the statement "AI can now solve any math problem a human can solve" false.
3 sources
- AI achieves silver-medal standard solving International Mathematical Olympiad problems - Google DeepMind
"Together, these systems solved four out of six problems from this year's International Mathematical Olympiad (IMO)... AlphaGeometry 2 proved the geometry problem, while the two combinatorics problems remained unsolved."
- Individual Results IMO 2024
The official results table shows human contestants scoring full marks on the problems AI missed: e.g. Haojia Shi scored 7 on P3 and 7 on P5, and many other contestants also received 7 on P3 and/or P5.
- Our First Proof submissions | OpenAI
"An academic department that has substantial overlap with the subject areas could conceivably solve many of the problems in one week... We initially believed our attempt for problem 2 was likely correct. Based on the official First Proof commentary and further community analysis, we now believe it is incorrect."