Techbros always exaggerate the claims.

They're solving problems humans haven't

I notice here the distinction between "haven't" vs. "can't".

Claiming that LLMs can be useful in solving some problems that humans simply haven't bothered to spend time looking at more deeply is one thing, but turning around and using those solutions - which may genuinely be low-key exciting and even useful for some practical purposes - is not the same thing as representing evidence that the LLM has reached AGI status, let alone being up to the "very stable senile genius" level.

Apples to oranges, except I note that where hype was needed for profits, somehow statements that vaguely look correct (if you don't delve too deeply) were found and provided. i.e., LLMs might not have reached the level of intelligence of an actual mathematical genius, senile or otherwise, but perhaps it has reached the intelligence level of an average techbro⁉️ (who again only values the appearance of correctness and doesn't let little things like "facts" get in the way of their statements!)

Oh the frontier models definitely can, now.

https://www.indiatoday.in/technology/news/story/openai-says-its-unreleased-astra-model-solved-10-hard-math-problems-anthropic-claims-fable-cracked-5-2962067-2026-08-03