Hacker Newsnew | past | comments | ask | show | jobs | submitlogin

> Force all companies training LLMs to add some method of watermarking with a mean error rate below a set value.

How do you watermark plain text?



This is a fairly obvious initial question which I assume nearly everyone who doesn't already have a rough answer in mind would ask, so I'm happy to report that it's fortunately quite clearly addressed in TFA, and in fact makes up a significant part of the (not very long) piece.


"I seem to do fine for a stretch, but at the of the sentence I say the wrong cranberry."


So deliberately bork the output in such a way that users lose all confidence in the product?


no, you modify the output probability so that you sample in a deterministic pseudo-probabilistic way - i.e. save the seed, and insert low SNR bias into the sampling. you can recover the bias afterwards and prove you generated the sequence.

my example was just a reference to Jarvis in the avengers.


You should read the original article... OpenAI say it doesn't downgrade quality.




Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: