I think they mean tail end of LLM innovation? As in, all the discovery has been done and so all that is left is just computing power and tweaks rather than brea

I think they mean tail end of LLM innovation? As in, all the discovery has been done and so all that is left is just computing power and tweaks rather than breakthroughs. I believe the OpenAI folks have already said something to that effect a while ago as well. For end users, the results of tweaks/computational power will seem like "breakthrough" to them but it's fundamentally not.