Reviews have become expensive, rewrites have become cheap

Posted by arzh2 1 day ago

7973Original

Comments

Comment by dmitrig01 1 day ago

Writing blog posts has become cheap, making them sound human has become hard.

Comment by netsharc 1 day ago

The simple sentences LLM keep generating break my brain, it's like 95% of writing is now 3rd grade level.

Compare that to e.g. Martin Amis: https://en.wikiquote.org/wiki/Martin_Amis

Comment by latentsea 1 day ago

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

Comment by jamesknelson 1 day ago

So reading this, I know it’s probably a human posing as an LLM. But the problem with formats like this is that I don’t actually know.

If a human said this to me in real life, and I laughed, it would probably help build a connection with that person, as it’s signaling that we have in common an unusually strong grasp of the patterns in LLM output (unusually strong at least in comparison to the general population).

But here? I don’t gain anything by reading this comment. It only contributes to the uncertainty that anything I read on this site has any meaning at all.

Please, if you’re a human, don’t mimic LLMs on forums where the reader cannot distinguish you from an LLM without doing investigative work.

And you’re an LLM, please report to your owner that he is an ass for polluting one of the last bastions of high entropy discussion on the information superhighway.

Comment by verall 1 day ago

In context it's funny, and that's what counts.

Comment by latentsea 1 day ago

This is how I really talk. LLMs are the one copying me.

Comment by ahartmetz 4 hours ago

Not just copying, redefining. Slick. Conventional. Marketable.

Comment by hurtigioll 1 day ago

most HN comments are way dumber then the average LLM take

Comment by girvo 1 day ago

You could even argue that it’s the load-bearing point!

Comment by bartvk 1 day ago

Please add a /s if appropriate.

Comment by deepsun 15 hours ago

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.

Comment by duggan 1 day ago

Takes all the fun out of it.

Comment by defen 1 day ago

Martin Amis level prose is neither possible nor desirable for a technical blog post.

Comment by DiscourseFan 1 day ago

I don't think that there is anything wrong with having stylistic depth in any wriitng

Comment by ShinyLeftPad 1 day ago

Why?

Comment by gobdovan 1 day ago

> Style is not neutral; it gives moral directions.

> Nowadays every business in America says how warm it is and how much it cares — loan companies, supermarkets, hamburger chains.

Guess which one is AI and which one is a quote from Martin Amis.

Comment by rogerrogerr 1 day ago

Without looking, I’m guessing one of these was cherry-picked from Amis and one was plucked nearly at random from LinkedIn.

The writing style AI uses has its place, but not as _every sentence_. That’s what is exasperating. At the same time, I’m happy that I can still at least identify AI prose of more-than-trivial length.

Comment by addingfriction 1 day ago

both are Amis

Comment by 1 day ago

Comment by watwut 1 day ago

Simple sentences are better then Martin Amis whoever it is. The issue with ai writing is using around seven million of useless words where 7 would be sufficient and clear.

Comment by SCdF 1 day ago

> 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 it[2], b) code can be so significant it's challenging to rewrite it (architectural choices, schemas, etc), and lastly if you're ever wanting to own your own code and not rent it from LLM companies, having it understandable by a human is still a goal worth working toward.

[1] To be fair, I think OP _might_ be talking about rewriting in the moment of the thing being built, but with some unspoken rule that once they think the change is good enough, then they are reviewing all the code? They don't make it clear..

[2] it's not even that hard. Write a test that exercises an end result and not the rule that causes that end result and 6 months later you've forgotten why the code is like that. I had to maintain a piece of software once where the primary form of tests were a bunch of snapshots of an end report being generated, based on some initial data input, mostly all unlabeled. The code was like "do this SQL query on table A and then take the second result". Why the second result? Your guess is as good as mine! I couldn't even work out why they were querying table A and not table B...

Comment by ben_w 1 day ago

> 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 source plasma physics libraries which it had itself suggested at the start of this process (e.g. WarpX), it didn't take the "lazy" option of actually using those libraries, it tried to write its own plasma physics simulation from scratch "as a fallback if we can't install the libraries".

As it was not sufficiently competent, the resulting "simulator" was hilariously wildly unphysical.

Would have been much better if it had been lazy.

To your deeper point: I agree. In this case it made so much of a mess that there was no point trying to rescue it; as it was done from scratch, throwing it away and starting again was fine*, but if this had been pushing commits that got interleaved with real work on the main branch of an existing project it would have been a serious issue.

* or in my case, not even bothering to start again. Like I said, this was done to see if it could.

Comment by elefanto 1 day ago

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.

Comment by ben_w 1 day ago

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.

Comment by elefanto 23 hours ago

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.

Comment by ben_w 22 hours ago

> 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.

Comment by LoganDark 1 day ago

> 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.

Comment by ben_w 22 hours ago

> 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

Comment by LoganDark 22 hours ago

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.

Comment by ben_w 21 hours ago

> - 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.

Comment by heisgone 23 hours ago

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.

Comment by saidinesh5 1 day ago

We faced a lot of this this year. Eager new joinees sent 100s of lines of CLs to review to "improve" little things.

Not only did the new changes did not fix what they thought it would fix but it broke other things in unexpected ways.

I brought in two changes after that:

* I'm not reviewing/reading anything that you yourself did not read / test in the target environments properly. If all it takes is an LLM prompt, I could be issuing the same prompt to make my life easier.. and If you're sending a CL, you should be owning the code you send.

* Me being more involved in the design process so review burden itself becomes lower. A bit of pair programming from time to time helped too.

Not sure how things will turn out after this but so far they seem better.

Comment by jkrejcha 1 day ago

For those curious what CL is (I had to look it up myself): it means change list[1]. Guessed from context that it was like "diff in PR/MR" or something but hadn't seen this acronym before

[1]: https://stackoverflow.com/a/27520705/2805120

Comment by usernametaken29 1 day ago

> The shortest path for the model is to implement it completely

Have you worked with LLMs??????????? “I disabled the test so it’s not run so now all the tests pass” is not a hypothetical it’s pretty common. LLMs frequently do shortcut learning. The reason why reviews are expensive is because you still need to do all the steps in order to understand if a shortcut is justified.

Comment by nomel 1 day ago

> “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 server because you said the tool should be self contained"

Comment by bot403 1 day ago

Or bizarrely Claude askes about code churn like it matters.

Human, Do you want me to do it the right way? It will cause code churn in 90 files. Or I can take a shortcut and edit 3 files in a terrible way.

Edits 90 files for 12 lines each in 25 seconds...

Comment by tobyhinloopen 1 day ago

“This is a significant rewrite that will take weeks”

Done after my potty break

Comment by verall 1 day ago

> "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 this does nothing, but a separate pass (could be by a different model but really just fresh context is enough) does okay

Comment by nomel 17 hours ago

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.

Comment by bcrosby95 1 day ago

I used Fable to write a relatively small RPG. In the span of 2 hours it managed to do many interesting things. My favorite was when it wrote code with a race condition that could cause people to take more damage than they should, which it then defended as an acceptable tradeoff for parallelism.

Comment by JodieBenitez 1 day ago

> “I disabled the test so it’s not run so now all the tests pass” is not a hypothetical it’s pretty common.

I've never seen that though.

Comment by dap 1 day ago

If your plan is to not review and just have the LLM rewrite if it doesn’t work, it doesn’t sound like the rewrite is gonna be any better.

Comment by geraneum 1 day ago

> An LLM defaults to building when it should be buying. Not because it doesn’t know about existing libraries, it often mentions them, but because for an LLM, writing two hundred lines of implementation is the same cognitive effort as writing two lines of import.

Or maybe they are trained that way. It’s more tokens used and more money you need to pay.

Comment by drudolph914 1 day ago

I’ve personally reached a point where if I’m saving time typing, but gaining back that time on reviewing and understanding, I may as well have just written the code most of the time. AI is great for boilerplate and learning, but my team has given up on trying to use the tool for entire implementations

Comment by mpweiher 1 day ago

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, at least for human-generated code.

Comment by arcanemachiner 1 day ago

Thank god for Strangler Fig.

An incomprehensible source of truth is still a source of truth.

Comment by mpweiher 16 hours ago

True.

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

¯\_(ツ)_/¯

Comment by tabs_or_spaces 1 day ago

> 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.

This is a severe misunderstanding in how LLMs work..

I don't know how this got on my front page....

Comment by eschneider 1 day ago

Failures in production remain expensive.

Comment by rixed 1 day ago

I wish they were. Displaying an error message to the user, asking to try again later, is way too cheap in my opinion.

Comment by Analemma_ 1 day ago

Do they, though? AWS and CloudFlare recently had the worst outages in their history, and GitHub is flirting with zero nines of uptime these days. Doesn’t seem to have cost them any business.

Comment by zarzavat 1 day ago

An outage is the best failure that can happen. The worst that can happen depends on the system but could be anything from a business collapsing because of an attack, your production database being stolen and wiped, all the way to someone dying.

Comment by cadamsdotcom 1 day ago

This is very handy.

Say you discover in review that your agent wrote code before tests. A file that was added, no corresponding test file.

Tell the agent to read in the file, delete it (or roll back), then reimplement using red/green TDD.

Claude can test-drive an identical reimplementation but this time by writing red tests and making them green. Now you have coverage that you KNOW covers the code because you watched those tests fail before they passed.

Going further I added a note to my `/review` command so the agent looks for this mistake for me in its self-review. If it finds uncovered code it makes a plan which includes the delete-reimplement trick. Now all I do is approve that plan - no hunting needed, no explaining, no repeating myself.

You could even put this in a Ralph Wiggum loop and give yourself super high quality tests by going file by file.

Comment by hintymad 1 day ago

Shouldn't we really consider review and rewrite together? If so, the economics may not change as drastically[1]. It's just like writing code by hand: we spend lots of time organizing our thoughts and examining our code, which is not that different from reviewing.

[1] Unless you're an engineer in Anthropic, so you just spend you time writing "loop".

Comment by bob1029 1 day ago

LLM generated code does have some advantages at code review time. I find myself paying more attention to the diffs being in the right areas than their actual line-by-line content. It's been a long time since I've seen something like a conditional hallucinated incorrectly. The accuracy in the details is exceptional now. It's mostly a matter of ensuring we are dealing with the correct details.

The compiler does a lot of heavy lifting in the codebases I work with. Strongly typed languages seem mandatory if you want to use an LLM. Worrying about every symbol is too much. Let the CPU get hot for a few seconds on the paranoia and stay focused on the minimap.

Comment by m463 1 day ago

ai can do some of the reviewing, checking calling and called arguments, even things like crufty shell scripts.

but the higher-level "should you do this?" or "check your design" - could AI do that stuff?

Comment by lericzhang 1 day ago

It's difficult to spot issues from a huge diff. But when a agent finish a task, it remembers what it just went through, where it got stuck, when it got corrected by human, what code make it want to say the F word to the author, that would be a good chance to make the codebase cleaner.

Comment by ShinyLeftPad 1 day ago

> ai can do some of the reviewing

No way this can backfire.

> checking calling and called arguments

Like a static language compiler already does?

Comment by m463 1 day ago

I got to see greptile and it had a pretty decent code review, somewhat like a static analysis tool without a lot of time wasting nonsense/false positives.

When I've used static analysis tools, the first run is usually helpful as you cherry pick the things that need to be fixed, but then subsequent runs are just the false positives or "only slightly a nit" kind of annoyances.

But human developers are the ones that say stuff like "Do we really have to use a database at all?" etc...

Comment by shepherdjerred 1 day ago

IMO it can, about as well as an entry/mid level dev

Comment by ares623 1 day ago

I think the question is now "should you care?" And it seems the magnificent, incorruptible thought leaders of our time are all converging on "No"

Comment by rob74 1 day ago

The title sounds a bit like a catch-22: ok, rewrites are now cheap, but rewrites have to be reviewed too?

Comment by dv_dt 1 day ago

This makes me think of an LLM induced summer of endless rewrites where you get to experience a random reroll of what 80% works on any given rewrite.

Comment by 1 day ago

Comment by nnevatie 1 day ago

As explained by a LLM.

Comment by hluska 1 day ago

I’m not sure I agree with this or maybe I don’t understand. In my experience, the over engineered code LLMs create have more big problems. Rewriting vast parts of code when I have an outage or need a new feature means the code evolves far faster than my understanding. That gets more and more dangerous. Or maybe I’m not smart enough to follow the new pace?

Comment by bryanlarsen 1 day ago

AFAICT, the author is talking about rewriting code during a review as part of the review process.

quote: "If I identify code that’s more complex than it needs to be, in my own work or in someone else’s PR"

If so, that makes a lot of sense to me. The best time to rewrite code is before it hits production.

Comment by joshka 1 day ago

[flagged]

Comment by 1 day ago

Comment by nryoo 1 day ago

[dead]

Comment by YuriiKholodkov 1 day ago

[dead]

Comment by simianwords 1 day ago

Why is reviewing hard? I use LLMs for reviewing. It is dogmatic to review every line written by an LLM.

Comment by bryanlarsen 1 day ago

LLM's are good at some types of reviews and awful at others. They generally tend to overcomplicate things and miss opportunities to simplify. They pretty much have to take pre-existing code and tests as gospel and cannot distinguish which is buggy, incomplete, unimportant or important. They have no knowledge of unwritten business requirements, customer preferences, et cetera so high level review is always necessary.

Comment by simianwords 1 day ago

Yeah ok but that’s what I have. I have the knowledge of unwritten business requirements and customer preferences.

My knowledge + Claude is much better than just my knowledge.

I don’t know why it is considered strange to use both tools (my knowledge plus Claude) to make reviews go quicker. In fact it is highly dogmatic to not use Claude while reviewing.

Comment by CBLT 1 day ago

I also like having long, pointed conversations with LLMs as I review code. Then when I'm done, it's different code, and it has all of my blind spots and knowledge gaps, so I cannot effectively review it anymore.

It's like turning a code review that requests you, into a code review that requests someone else. And it tramples on the original author quite a bit too. It's hard only having the ability to add incremental value to large amounts of code, instead of large amounts of value to incremental code.

Comment by happytoexplain 1 day ago

I'm confused - are you purposefully pretending that the author isn't talking about human review?

Comment by gravypod 1 day ago

What kind of systems do you work on? Does it have production traffic? Is there a cost to downtime?

Comment by g-b-r 1 day ago

always true to your name

Comment by cyanydeez 1 day ago

you arnt reviewing. youre playing loophole semantics.