@da_667 counterpoint: LLMs are opaque, weird machines (in the Sergey Bratus sense) that reveal surprising things when interrogated. https://www.cs.dartmouth.edu/~sergey/wm/ similar to the execution abstraction in the classic "Next 700 Programming Languages" paper https://www.cs.cmu.edu/~crary/819-f09/Landin66.pdf anyhow, just advocating for why sometimes they're kind of neat.