Comment by unparagoned

What do you mean?

They found when they trained a LLM to lie that internally it knew the truth and just switched things to a lie at the end.