[this_user]

Compilers are deterministic and they actually possess domain knowledge of what they are trying to do. AI models are non-deterministic, have no real domain knowledge due to lack of an underlying world model, and their way of "writing" software is to spew out something that looks like something that they have been trained on, then iterate on it long enough until it has reached the level of being barely runnable.

[wgd]

People say "determinism" but I don't think that's actually the property we care about. For instance you could imagine a compiler that makes heavy use of superoptimization with random search and it would still have the ineffable quality that LLM codegen lacks. I think what we're actually trying to say is that the compiler preserves the formal semantics of the source language in its output, whereas English text doesn't have any such formal semantics to preserve.