You know what, good on anthropic!
Ah yes, dreaming of an AI novel.
“And what do you want to be when you grow up, Timmy?” “I want to pretend to write stories and then get upset I have to do any work at all.”
Maybe I’m out of the loop, but how would I watermark a text without sounding obvious? By using weird phrases? They get edited. By using e.g. an exact combination of starting letters over a large paragraph? One changed word and it’s broken. And even if not, I could happen to write the same myself, and then?
How am I hiding a signature in Plain text? Anyone got a better idea than my silly ones?
Yeah, it’s the verbose, waffling, psychophantic tone they all have.
Hey Claude, save this text in ASCII 0-127 encoding.
That’s not how the watermarking works. It depends on a pattern deviations from statistics throughout a string of tokens.
Don’t confuse me with facts! (And, thanks… I have since seen the articles about how it works based on word selection from a subset of the words that would have been randomly selected by the “heat” of the model.)
I don’t know much about it but it sounds like it would only work if you have the weights and maybe the running conditions of the model. Otherwise the statistical wright wouldn’t be known.
It also sounds really easy to break by changing random words.
If you dig a little you can get the articles explaining better, but basically: yes, you have to have the weights, and a secret key… Given those, they split the words that might be chosen at each point into two groups and only choose from one group.
Later, you can check for the watermark by splitting the same way and seeing how many times the words in the text match the group that fits with the secret key. Normal text would hit a 50% match, but unlatered AI generated text would hit 100% match, so… somewhere higher than 50 you start to suspect AI involvement, possibly edited, and by the time you get over 99% it’s virtually certain that AI had some involvement - particularly if the sampling of words runs into the thousands.
Thanks :)