Skip to content

Comment on How AI text watermarking worksparent

Comments

I’d be curious to play with this with some different open weights models. I’d think that a single next token would end up having similar probability distributions so given that this a probabilistic method, you’d be able to get at least some signal.

I suppose if that did work someone would have been able to work backwards and crack their key already, so I must be missing something.

AboutSource Built by g1lg1l

Hackerly is an independent reader for Hacker News, built on the public HN API. Not affiliated with Y Combinator.