FWIW, LLMs have been getting better so fast that we only barely begun figuring out more advanced ways of applying them. Even if they were to plateau right now, there'd still be years of improvements coming from different ways of tuning, tweaking, combining, chaining and applying them - which we don't invest much into today, because so far it's been cheaper to wait a couple months for the next batch of models that can handle what previous could not.