> They can't be better then the average human though because of training data

Is this based on the belief that an LLM can only represent an "average" human being?