It feels like the pace of improvement for foundational models is already slowing down. The trajectory feels much more linear than a few years ago when we had massive, qualitative jumps like GPT-2 to GPT-3.
I mean in my opinion, slowing down can only do so much. It's already improving fast because of the AI training in these data centers. I think it would help but, I think we should just focus more on how we treat it and how we use it. Since, in my opinion at least, most use it like it's a search engine or like it is the only thing who can do something for us, I mean like, we stop treating it like a tool and more of a replacement. What's sad these days are many small companies treat it or at least in my perspective treat it like a replacement, since it sometimes cheaper to go with AI, that's I think one of reason it was innovating fast because of demand. There should still be people, the one who are in control, and AI should be a second brain or collaborator actually.
They now want to lock you out after they have IPO'd.