Post

Suppose the model has a 49.9% probability of choosing banana, and 50.1% pineapple. Today, without watermarking, you run the model twice and get two different results. Sure, watermarks modify the probability. But you’re still getting a word it was *already* choosing half the time?

George Black

@george.black

Banana, the “inferior” word, was already showing up in your output before. I don’t think watermarking makes the model produce something it wasn’t already doing. I could be misunderstanding how it works though, it’s all kinda way over my head. 😅

I read through John's very long rant, as well as the supporting documentation and referenced essay and... I think this is a lot of ado about nothing. Nothing practical or perceptible, that is.

In most cases *including the examples John gives*, the diff is largely in preference, not meaning, since the contextual intent itself generally can't be inferred. And even if I'm wrong there, if John really prefers bananas vs. pineapples, *he can just substitute that text if/when he repurposes it!*

And holy forkballs, you know what's REALLY REALLY annoying? The anemic character limit here. I swear, it's rage-inducing and is likely to have me posting here less often because I'm tired of having to split up connected thoughts across multiple posts.

17/8/26, 6:29 am1 repost1 like