> Impressive reasoning (sounds LLM-ish convincing). The patch is too large to > trust though.
Only (the work of) human beings can possibly be trusted anyway; algorithms can be proven correct, but that's almost certainly hopeless for LLMs. Very recent experiences of "vibe coding" in the mathematical world have shown that even the most recent models still need lot of iterations and nudging to generate correct proofs for moderately complicated claims, and that the result often contains quirks that no human mathematician would dare to write. -- Eric Botcazou
