The models will not keep betting better. We have pased "peak LLM" already, by my estimate. Some of the parlour tricks that are wrapped around the models will make some incremental improvements, but the underlying models are done. More data, more parameters, are no longer doing to do anything.

AI will have to take a different direction.