Live data from Hacker News

Reviews have become expensive, rewrites have become cheap

news.ycombinator.com

71–80 of 82 posts

Re: Reviews have become expensive, rewrites have become cheap

#71
post #55
post #47

> LLMs aren’t lazy. They don’t cut corners because a simpler solution feels good enough. If they know how to solve something thoroughly, they will. I don't know why they think this, but no? Perhaps it's badly expressed, but LLMs cut corners all the time. It's sort of their core fault really. Anyway, I disagree with the core premise[1]. Re-writes are not cheap, because 1) code can be so bad it's unclear how to rewrite…

> I don't know why they think this, but no? Perhaps it's badly expressed, but LLMs cut corners all the time. It's sort of their core fault really. I think this is a matter of perspective about what counts as "cutting corners". I think they look like you describe only because they have limited competence; this is on the basis that when I asked one to make a fusion reactor simulator (to see if it could) by using open s…

> I think this is a matter of perspective about what counts as "cutting corners".

The nature of probabilistic sampling practically guarantees that corner cutting is always just a few samples away. Certain sampling strategies can mitigate this, but there's no way to fully eliminate it, without fully eliminating it from the training data and guarding against it during training. Model reasoning can help by giving the model space to draft and review its approach before it executes, but models still aren't guaranteed to follow their own thinking. A mistake or shortcut can always simply slip in during generation and it won't always be caught and corrected.

Re: Reviews have become expensive, rewrites have become cheap

#72
post #65

Earlier quoted context omitted.

The diligence to the absurd that you describe is really just a consequence of cutting corners at planning stages. Consider an organization that relies heavily on formal proofs and specifications to one that uses what fits on the 3 bullets allowed in a PowerPoint slide. The first might do less work over all because the second is cutting corners in planning.

Perhaps, though I find myself disagreeing as the planning stage where it itself told me about WarpX. It just then made stupid decisions when left to itself. Can plan… just not well. Can code… just ok. Can do TDD… just most of the time. Definitely none of these well enough to be a dark factory though.

Yeah, well.. I think a lot of engineers don't have reason to see the bulk of the industry but if you took an exactly median manager, programmer or PM and ask them a few things you also tend to find that they don't even notice that they will contradict themselves in their answer to a second question, etc. Organizations in general are harnesses that try to get some kind of result out of some pretty sloppy attention.

Re: Reviews have become expensive, rewrites have become cheap

#73
post #55

Earlier quoted context omitted.

> I don't know why they think this, but no? Perhaps it's badly expressed, but LLMs cut corners all the time. It's sort of their core fault really. I think this is a matter of perspective about what counts as "cutting corners". I think they look like you describe only because they have limited competence; this is on the basis that when I asked one to make a fusion reactor simulator (to see if it could) by using open s…

> I think this is a matter of perspective about what counts as "cutting corners". The nature of probabilistic sampling practically guarantees that corner cutting is always just a few samples away. Certain sampling strategies can mitigate this, but there's no way to fully eliminate it, without fully eliminating it from the training data and guarding against it during training. Model reasoning can help by giving the mo…

indeed. That being said, there is a psychological effect of reviews that doesn't exists with LLM. One of the thing that make reviews effective is that when we code, we know someone will look at our code and judge it. It might be subconscious, but it's there.

Re: Reviews have become expensive, rewrites have become cheap

#74
post #65

Earlier quoted context omitted.

Perhaps, though I find myself disagreeing as the planning stage where it itself told me about WarpX. It just then made stupid decisions when left to itself. Can plan… just not well. Can code… just ok. Can do TDD… just most of the time. Definitely none of these well enough to be a dark factory though.

Yeah, well.. I think a lot of engineers don't have reason to see the bulk of the industry but if you took an exactly median manager, programmer or PM and ask them a few things you also tend to find that they don't even notice that they will contradict themselves in their answer to a second question, etc. Organizations in general are harnesses that try to get some kind of result out of some pretty sloppy attention.

> Organizations in general are harnesses that try to get some kind of result out of some pretty sloppy attention.

Certainly.

LLMs have superhuman attention, and are so cheap that they can be left running for absurdly long outputs. Despite this, still not high on the competence scale (yet*).

A human doesn't output a few million tokens every day, not even counting our inner monologues, but the AI asymptote to an upper bound of quality somewhat worse than a senior human.

* I would like to have a new job before they reach this point. Got a mortgage to repay.

Re: Reviews have become expensive, rewrites have become cheap

#75
post #55

Earlier quoted context omitted.

> I don't know why they think this, but no? Perhaps it's badly expressed, but LLMs cut corners all the time. It's sort of their core fault really. I think this is a matter of perspective about what counts as "cutting corners". I think they look like you describe only because they have limited competence; this is on the basis that when I asked one to make a fusion reactor simulator (to see if it could) by using open s…

> I think this is a matter of perspective about what counts as "cutting corners". The nature of probabilistic sampling practically guarantees that corner cutting is always just a few samples away. Certain sampling strategies can mitigate this, but there's no way to fully eliminate it, without fully eliminating it from the training data and guarding against it during training. Model reasoning can help by giving the mo…

> practically guarantees that corner cutting is always just a few samples away.

If it were that bad, 100% of chat with AI would look like this comment I'm writing

Re: Reviews have become expensive, rewrites have become cheap

#76
post #75

Earlier quoted context omitted.

> I think this is a matter of perspective about what counts as "cutting corners". The nature of probabilistic sampling practically guarantees that corner cutting is always just a few samples away. Certain sampling strategies can mitigate this, but there's no way to fully eliminate it, without fully eliminating it from the training data and guarding against it during training. Model reasoning can help by giving the mo…

> practically guarantees that corner cutting is always just a few samples away. If it were that bad, 100% of chat with AI would look like this comment I'm writing

Two things:

- Good logits and sampling strategy can make cases like those exceptionally unlikely -- sufficiently so for one to assume it won't realistically happen.

- Once a bad path is sufficiently taken, it tends to be a lot more likely to continue.

This leads to real-world advice:

- If a model refuses your request, do not argue with the refusal; edit your original message or otherwise regenerate -- the presence of refusal tells the model it should continue refusing.

- More generally, don't allow the model context to get contaminated with behavior or commands you don't like -- describing what to do is more effective than describing what NOT to do.

- It's rude to push unreviewed model outputs onto others.

I'm not saying corner-cutting is a thing that is necessarily all over the place (even though there are countless examples in the wild). I'm also not saying it always results in random stops, or that doing one thing bad makes everything else bad. What I'm saying is that bad decisions could be hidden anywhere, at any time, even if everything else looks fine. Such is the nature of current LLMs.

Re: Reviews have become expensive, rewrites have become cheap

#77
post #75

Earlier quoted context omitted.

> practically guarantees that corner cutting is always just a few samples away. If it were that bad, 100% of chat with AI would look like this comment I'm writing

Two things: - Good logits and sampling strategy can make cases like those exceptionally unlikely -- sufficiently so for one to assume it won't realistically happen. - Once a bad path is sufficiently taken, it tends to be a lot more likely to continue. This leads to real-world advice: - If a model refuses your request, do not argue with the refusal; edit your original message or otherwise regenerate -- the presence of…

> - Once a bad path is sufficiently taken, it tends to be a lot more likely to continue.

no they demonstrably self correct with injected bad tokens

> - If a model refuses your request, do not argue with the refusal; edit your original message or otherwise regenerate -- the presence of refusal tells the model it should continue refusing.

wish i could do that with humans :P

> - It's rude to push unreviewed model outputs onto others.

Yes but that's not why.

Same real-world advice applies to stuff random fresh graduates make. I remember being one of those.

The incompetence is why.

Re: Reviews have become expensive, rewrites have become cheap

#78
post #68
post #40

Earlier quoted context omitted.

> “I disabled the test so it’s not run so now all the tests pass” Also: "I implemented it this terrible way because of precedence in the codebase...that I just wrote" "I avoided implementing this correctly because of migration concern for existing installations of this code I'm writing right now" "I deferred this critical feature for the future, so we can deploy quicker" or, my favorite, "I hand rolled an buggy http…

> "I avoided implementing this correctly because of migration concern for existing installations of this code I'm writing right now" This one grinds my gears so bad, probably 1/4 of my job at this point is telling an LLM (either in my own editor or in review comments for someone's MR) "how about we do this right _before_ merging it, eh?". Or: foo = abc if foo != None: ... In my experience giving it "rules" not to do…

All the global memories I have stored are basically just this. Memories appear to be completely useless on their own, but, before a merger or a plan is finalized, having it go read them will definitely help course correct.

I really wish there was an "every x tokens" type trigger for every agent, where I could have it fire off a "Pause, go read the guide to make sure you're adhering fully." To help keep concepts fresh in context.

Re: Reviews have become expensive, rewrites have become cheap

#79

Rewrites are only cheap if the rewrite solves the problem. And the rewrite can only reliably solve the problem if you understand the problem, and even then it's obviously not a guarantee. If you have a huge blob of code that nobody understands then re-generating a new blob of code that nobody understands is unlikely to solve the problem. That said: I've always been a fan of optimistic rather than pessimistic merging,…

Thank god for Strangler Fig. An incomprehensible source of truth is still a source of truth.

True.

Alas in incomprehensible source of false is not a source of truth.

¯\_(ツ)_/¯

Re: Reviews have become expensive, rewrites have become cheap

#80
post #46

Earlier quoted context omitted.

That's not just you imagining things, that's the world around you changing. That's real and it matters.

Please add a /s if appropriate.

I only add /s if it's not obvious, and someone might take it seriously. Although online you never know, too often people take seriously even the most obvious.
Post reply on HN