The last five years have seen dramatic advances in Natural Language Understanding (NLU) and Generation (G), with increasingly large language models trained on large corpuses. These models have moved the needle on academic challenge sets, and are increasingly finding their way into commercial applications.
At the same time there has been discussion of the limitation of this brute-force, statistical approach, and arguments that this approach alone will reach diminishing returns.
I will discuss these issues in general, and their possible ramifications in healthcare.
