I think there is still a lot of the 'dancing bear' about many LLM applications.

We are so impressed by the fact that the bear dances at all, that we tend to forget to ask ourselves whether it dances well.

And whether a human dancer might not be more effective and humane.

With the bear, we know that it will never achieve the grace of a human. With AI, who knows? But how much of our available resources, humanity and social underpinnings might we sacrifice in the determination (of a vested minority interest) to find out? We may well ultimately be the losers either way.