Yeah, but there have been papers being published in general LLMs as well finetuned off of GPT4 instead of humans. Even in the narrow space the question remains

Yeah, but there have been papers being published in general LLMs as well finetuned off of GPT4 instead of humans. Even in the narrow space the question remains. If I build a superior model for task X using gpt4 and it’s superior at X, can I think use my new model to train another model in X and continue to see benefits ?