For those predicting AI will solve math, why am I not hearing about amazing AI papers in theoretical physics ? It’s less formal so one might think it’s more amenable to natural language.
When an LLM writes a proof in natural language, which a mathematician has to check, its calling the mathematician as a tool. So if you haven't checked the output that's not a "proof". Its a bug in your harness, just like if your LLM failed to properly call the Lean kernel.
The "AI is about the make mathematicians obsolete" view misses that mathematics is not a monolith. There are many areas of math where the progress of AI, even as a research aide has been modest since 2023.
I for one welcome our new AI overlords.
openai.com/index/model-dispr…
