Manuel Simoni

The fact that Haskell typeclasses work better than OCaml modules shows how lame the best functional programming is compared to even a stunted informal underspecified undocumented version of logic programming.

This idea comes up again and again, but I profoundly disagree that historical quotes about compilers (or high-level languages) have anything to do with the current reaction about what LLMs can or cannot do. Compilers work with formal languages. We have, or can have a semantics for both the input and the output, and we can prove (at least in principle) that the compiler is a semantic-preserving translation. Note that the paraphrased (?) quote was about performance (what is "good"?), and it precisely the fact that the input and output are formal languages that compiler optimization became a fruitful field that produced impressive results. Furthermore hand-crafed assembly still regularly beats compilers. It just became less important. But stepping out a bit, compilers are but a prerequiste of efficient high-level languages (which I suspect John Backus was arguing for). High-level languages are the point, not compilers. And high-level languages are rich in features and utility precisely because they are formal languages with rich semantic properties and we can explain and guarantee those features. LLMs might be useful, even important, but I think it's misguided to compare them to higher-level languages or to compilation. Formal vs. informal language is what precludes this comparison. Not only the input is not a formal language with a know semantics, but the translation mechanism itself is not explainable with respect to ANY semantics. It's not even stable across time.