Comment by gitaarik

Ok, so you claim that LLMs can get smarter without human validation. So why do they hallucinate at all? And why are all reward models currently curated by humans? Or are you claiming they aren't?