Yes, but why would it? "Oops, I got it wrong the first time because I crapped out the first thing that matched in my training data" isn't in the training data. Yet.
So it can't come out of the LLM: There's no actual introspection going on, on any of these rounds. Just using training data.
Yes, but why would it? "Oops, I got it wrong the first time because I crapped out the first thing that matched in my training data" isn't in the training data. Yet.
So it can't come out of the LLM: There's no actual introspection going on, on any of these rounds. Just using training data.