Yea, the pervasiveness of this analogy is annoying because it's wrong (because a compiler is deterministic and tends to be a single point of trust, rather than trusting a crowdsourced package manager or a fuzzy machine learning model trained on a dubiously-curated sampling of what is often the entire internet), but it's hilarious because it's a bunch of programmers telling on themselves. You can know, at least at a high level of abstraction, what a compiler is doing with some basic googling, and a deeper understanding is a fairly common requirement in computer science education at the undergrad level

Don't get me wrong, I don't think you need or should need a degree to program, but if your standard of what abstractions you should trust is "all of them, it's perfectly fine to use a bunch of random stuff from anywhere that you haven't the first clue how it works or who made it" then I don't trust you to build stuff for me

I think you're mistaken on that. Maybe me and the engineers I know are below average on this but even our combined knowledge of the kinds of things _real_ compilers get up to probably only scratches the surface. Don't get me wrong, I know what compilers do _in principle_. Hell I've even built a toy compiler or two. But the compilers I use for work? I just trust that the know what they're doing.

I'd wager a lot of money that the huge majority of software engineers are not aware of almost any transformations that an optimizing compiler does. Especially after decades of growth in languages where most of the optimization is done in JIT rather than a traditional compilation process.

The big thing here is that the transformations maintain the clearly and rigorously defined semantics such that even if an engineer can't say precisely what code is being emitted, they can say with total confidence what the output of that code will be.

Not in any great detail. Gold vs ld isn't something I bet most programmers know rigorously, and thats fine! Compilers aren't deterministic, but we don't care because they're deterministic enough. Debian started a reproducible computing project in 2013 and, thirteen years later, we can maybe have that happen if you set everything up juuuuuust right.

They also realize that adding two integers in a higher level language could look quite different when compiled depending on the target hardware, but they still understand what is happening. Contrast that with your average llm user asking it to write a parser or http client from scratch. They have no idea how either of those things work nor do they have any chance at all of constructing one on their own.

> Then, when it fails [...], you can either poke it in the right ways or change your program in the right ways so that it works for you again. This is a horrible way to program; it’s all alchemy and guesswork and you need to become deeply specialized about the nuances of a single [...] implementation

In that post, the blanks reference a compiler’s autovectorizer. But you know what they could also reference? An aggresively opaque and undocumented, very complex CPU or GPU microarchitecture. (Cf. https://purplesyringa.moe/blog/why-performance-optimization-....)

I phrase I use is spooky action at a distance. Quantum entanglement but with software.