> that the experts developing the models expect that to continue.

We must be listening to different experts then. One small example, Apple's widely discussed paper on the limits of current approaches: https://machinelearning.apple.com/research/illusion-of-think...