I dont think think my point is getting across. This is in the context of how much world knowledge a model needs to be trained on, not llm vs not llm.