3 ms·

> Monitor them for "dangerous" thought like we do with LLM We do that?

• yellowapple 3y ago
  Not very well, given the existence of "DAN" and similar prompt hacks, but yes.