@devsimsek this only means that LLMs can't provide their own training data, right? Could they still "invent" new algorithms, that make more of the existing data?