[andy99]

Yes, and it's what people seem to ignore when they talk about dethroning GPT4 as the top LLM. It's good data expressly developed for training the behaviors they want that keeps them ahead, all the other stuff (other training and filtering web data) has much less of an impact.

See also "You won't train a better model from your desk: https://news.ycombinator.com/item?id=40155715

[CuriouslyC]

I don't think GPT4 is the top LLM, it's good at coding and good at understanding poorly written prompts but its high level prompt following and creativity are not great. GPT4 likes to answer a particular way and when your question matches up with that it'll seem very smart, but when it doesn't the rails it is on are very obvious.