You make a good point, that maybe the models won’t perform much better with those improvements, or at least not enough to get people to pay more. I’m curious a

You make a good point, that maybe the models won’t perform much better with those improvements, or at least not enough to get people to pay more. I’m curious about the Llama3 bit - do you have a source for that? I’ve been hearing they trained using OpenAI outputs (not sure how that would work).