[sarchertech]

If you can’t understand the difference between a bug that will rarely cause a compiler encountering an edge case to generate a wrong instruction and an LLM that will generate 2 completely different programs with zero overlap because you added a single word to your prompt, then I don’t know what to tell you.

[petcat]

The point is that expert humans (the GCC developers) writing code (C++) that generates code (ASM) does not appear to be as deterministic as you seem to think it is.

[sarchertech]

I’m very aware of that, but I’m also aware that it’s rare enough that the compiler doesn’t emit semantically equivalent code that most people can ignore it. That’s not the case with LLMs.

I’m also not particularly concerned with non-determinism but with chaos. Determinism in LLMs is likely solvable, prompt instability is not.

[jplusequalt]

Classic HN-ism. To focus on the semantics of a statement while ignoring the greater point in order to argue why someone is wrong.

[anthonyrstevens]

I think it's a perfectly fine point. The OP said (my interpretation) that LLMs are messy, non-deterministic, and can produce bad code. The same is true of many humans, even those whose "job" is to produce clean, predictable, good code. The OP would like the argument to be narrowly about LLMs, but the bigger point even is "who generates the final code, and why and how much do we trust them?"

[sarchertech]

As of right now agents have almost no ability to reason about the impact of code changes on existing functionality.

A human can produce a 100k LOC program with absolute no external guardrails at all. An agent Can't do that. To produce a 100k LOC program they require external feedback forcing them from spiraling off into building something completely different.

This may change. Agents may get better.

[petcat]

I argued the greater point? Software code-generation is not deterministic, whether it's done by expert humans or by LLMs.

[sarchertech]

It has nothing to do with determinism. It's the difference between nearly perfectly but not quite perfectly translating between rigorously specified formal languages and translating an ambiguous natural language specification into a formal one.

The first is a purely mechanical process, the second is not and requires thousands of decisions that can go either way.

[raw_anon_1111]

And that’s no different than human developers

[sarchertech]

The difference is that a human is that a human can reason about their code changes to a much higher degree than an AI can. If you don't think this is true and you think we're working with AGI, why would you bother architecting anything all or building in any guard rails. Why not just feed the AI the text of the contract your working from and let it rip.