Satirizing LLM Math Proofs: Just Rearranging Existing Knowledge?

burny_tech · x · 2026-08-03

The post sarcastically rebuts the recent hype around large language models (like o1) completing complex mathematical proofs.

The author mockingly claims that all the words and numbers used by the model in its proof are simply part of the training data. After reviewing a 125-page summary of the chain of thought (CoT), they claim to have found "not a single novel symbol."

They argue the model merely rearranged previously known mathematical elements into an order that happened to solve the problem. Ignoring the actual mechanics like rearrangement, dependency structure, proof search, and synthesis, they dismissively conclude this is just "memorization."

Original post →

More from Fun

Fun channel →