3 ms·
Perhaps AI generated text should be created with a specific signature in mind _specifically_ to be identifiable?
by lwhi 4y ago
Perhaps AI generated text should be created with a specific signature in mind _specifically_ to be identifiable?
- mattnewton 4y agoisn’t this essentially asking anyone who runs a model to flip the evil bit[0]? People who want to misrepresent model output as human written output will trivially be able to beat this protection by removing the signature or using a version of the model that simply doesn’t add it. [0] https://en.m.wikipedia.org/wiki/Evil_bit https://en.m.wikipedia.org/wiki/Evil_bit
- lwhi 4y agoI think you're correct, but it would promote the idea that what's produced is a basis or source for further work .. and would mean that effort is required. Feels like a good basis tech for something like ChatGPT.
- wongarsu 4y agoThere's a large body of research into invisible text watermarking, so this would certainly be possible. Maybe the simplest to implement in LLMs would be to bias the token generation slightly, for example by making tokens that include the letter i slightly more likely. In a long enough text you could then see the deviation from normal human text characteristics.