I think AI has great potential to change as much as the internet. But i dont consider LLMs to be the right type to do that.

Self driving cars and intelligent robotics is the real goldmine. But we still don't seem to have the right architecture or methods.

I say that because self driving cars are entirely stagnant despite the boom AI interest and resources.

Personally i think we need a major breakthrough in reinforcement learning, computer vision (which are still mostly stuck at feed forward CNNs) and few shot learning. The tranformer is a major leap, but its not enough on its own.