News for hackers. Stuff that matters.

A comment on Hello Dolly: Democratizing the magic of ChatGPT with open models. Parent comment

Re:Hello Dolly: Democratizing the magic of ChatGPT with open... by kurthr on Friday March 24, 2023 @11:24AM (#35290270)
Even easier have a small set of 8-10 character gibberish tokens it's trained on in a particular contexts (eg a non-existent poem). Then feed it one or several poems and see if a gibberish token pops out.
[ Parent | on HN ]
Re:Hello Dolly: Democratizing the magic of ChatGPT with open... by eigenvalue on Friday March 24, 2023 @11:28AM (#35290346)
I think they call these canary GUIDs. If you manage to generate one from an LLM then you can conclude with certainty that the model saw that document during training.
[ Parent | on HN ]