Comment by afspear

Maybe we should find other datasets not generated by humans to train LLMs?