[sgt101]

There is a lot of retrieval in the behaviours of LLM's, but I find it hard to characterize it as overfitted. For example, ask ChatGPT to respond to your questions with grammatically incorrect answers.