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 tra

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.