>>53823
>>53824
Yeah, and we've already used almost all of the good available data for training. I think LLMs are going to plateau for a while.