Hacker News Reader

Analyze it using an LLM. Claude was pretty ruthless about this one.

Yeah, for me Claude identified the phrase "this holds with probability 1 over random weight matrices since the null space has dimension"

Treating trained weights as random for the purpose of a proof is immediately discrediting for a paper to me.

"This holds for almost all matrices" is actually something you'd want to know if we're talking about probabilities, no?

sure but it seems spiritually wrong to use an LLM to debug a slop paper. Who knows, maybe claude generated it in the first place?