4 ms·
> AFAIK there is no fingerprint in any model’s output. It seems like there easily could be. What if some of the data they trained it on didn't exist anywhere
by stametseater 4y ago
> AFAIK there is no fingerprint in any model’s output.
It seems like there easily could be. What if some of the data they trained it on didn't exist anywhere else except in the training set, and was put there specifically for this purpose? For instance they could have taught it a few poems that don't exist anywhere else. If you can coax the LLM of unknown origin into reciting those poems back to you, you know where it came from.
- kurthr 4y agoEven 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.
- eigenvalue 4y agoI 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.