yeah i absolutely agree that even classifiers can leak, and the autocorrect thing sounds like i was wrong about generative (it sounds like an n-gram setup?)...

yeah i absolutely agree that even classifiers can leak, and the autocorrect thing sounds like i was wrong about generative (it sounds like an n-gram setup?)... although they also say they don't train LLMs (what is an n-gram? still an LM, not large... i guess?)