will there ever be a 1B param model that’s more capable in every way than GPT-4 is today?

→ View original post on X — @jxmnop