Your knowledge is out of date. For example, Claude runs a Linux container with Python and Sympy, so it absolutely can do a good chunk of math now. Still makes mistakes, but it’s good enough to serve as a LaTeX assistant. Similarly for ChatGPT but crappier. If you have the computing power, I think you can also give a local LLM access to Python tools with OpenWebUI.
LLMs are notoriously bad at doing even the simplest math problems
Your knowledge is out of date. For example, Claude runs a Linux container with Python and Sympy, so it absolutely can do a good chunk of math now. Still makes mistakes, but it’s good enough to serve as a LaTeX assistant. Similarly for ChatGPT but crappier. If you have the computing power, I think you can also give a local LLM access to Python tools with OpenWebUI.
Finding a proof is different than solving a math problem.
How so? Particularly in intuitionist logic it would seem to me that they are the same thing.
You’re right. The first is way harder and requires being able to do the second. Which it can’t.
Except this article is literally about the response to AI doing it multiple times.
https://arstechnica.com/ai/2026/06/openais-math-breakthrough-played-to-ais-strengths/
https://www.smithsonianmag.com/smart-news/ai-disproves-a-decades-old-mathematical-idea-the-biggest-conjecture-that-the-tech-has-played-a-role-in-yet-180989189/
https://patmcguinness.substack.com/p/openais-astra-tackles-mathematical
The whole, “LLMs can’t do basic math,” thing is outdated.
They can’t even solve axioms, much less idioms