No, I'm talking about using a LLM to generate LLM training data, ungrounded (with no selection by a non-DL system). AlphaGo game data is actual game data, it isn't generated by a LLM.
LLM-generated training data differs from actual game data like AlphaGo
By
–