The problem is, they're nothing like transistors, and never will be. Those are simple. Work or don't, consistently, in an obvious, or easily testable, way.

LLM are more akin to biological things. Complex. Not well understood. Unpredictable behavior. To be safely useful, they need something like a lion tamer, except every individual LLM is its own unique species.

I like working on computers because it minimizes the amount of biological-like things I have to work with.