[stingraycharles]

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

[thethirdone]

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.

[EGreg]

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

[gaze]

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?