Comment by taylorfinley

Maybe we can create some extremely low probability sentences and make sure to include them in our chats. If a future model can re-create the very low probability sequence, we have proof of our "private" conversation being trained on or accessed.