News for hackers. Stuff that matters.

A comment on Anthropic's 'Watermark' Text Adulteration in Claude Is a Perversion of Writing. Parent comment

Re:Anthropic's 'Watermark' Text Adulteration in Claude Is a ... by robocat on Sunday August 16, 2026 @09:18PM (#49325555)
What a crazy link:

  So we train a second copy of Claude to work backwards—reconstruct the original activation from the text explanation. We consider an explanation to be good if it leads to an accurate reconstruction. We then train Claude to produce better explanations according to this definition using standard AI training techniques.
Incentives to train a pathological liar. There's no baseline so can only catch out the worst of the lies/errors. Anything (including fabrications) that passes our filters is reinforced?
[ Parent | on HN ]
Re:Anthropic's 'Watermark' Text Adulteration in Claude Is a ... by ACCount37 on Monday August 17, 2026 @03:55AM (#49327671)
What are you even talking about? If there was a "baseline" for how to decode activations into text we wouldn't be having this conversation.

The choice is between "this reconstruction sucks" and "no reconstruction", and we're only now beginning to learn how to make those reconstructions suck less.

[ Parent | on HN ]