Those other levels of abstraction that you mentioned are deterministic and predictable. I think that's a massive difference and it justifies why one should be more skeptical towards AI generated code (skeptical doesn't mean "outright dismissing" tbf).

Deterministic tab completion generates horseshit half the time too, at the end of the day you have to know what you're doing in either case imo. You can also run an LLM in deterministic mode, but performance is usually worse.

Fun fact that we have burned those bridges years ago. Every time big codebase updates it's main optimizing toolchain (compiler/linker) ... there is nothing deterministic and predictable. There is only what has been re-tested and what has not been.

> there is nothing deterministic

https://reproducible-builds.org/

> nothing [...] and predictable

Even if the build isn't bit-for-bit-reproducible, any reasonable toolchain will, at least, give predictably equivalent results.

Another level of abstraction that has its real world cost made invisible, from job displacement to environmental damage.

I never saw anyone on HN bemoan the “environmental damage” of data centers until LLMs started popping up. As if all other software and internet tech is necessary and “worth it”?

They did, with crypto. The environmental cost of, like, a website is minuscule in comparison.

Yes, and they found ways to avoid 99.99% of power consumptions by moving away from proof of work, which most of the blockchains use today. But the hate remained.

Well the hate is now mostly because it’s used as a speculative security to run scams.

The energy usage is massive though, and we definitely should be asking ourselves if other solutions from cloud, crypto, streaming etc are worth expanding. The industry’s environmental footprint is on par with aviation, and with AI we’re set to increase our emissions massively when we should be decreasing them.