Various LLM Smells
shvbsle.in
shvbsle.in
A general pattern for LLMs is that they look really good at things you are bad at. What that means is that if you find yourself thinking of its output as significantly better than yours in a particular domain, there's a high chance that you are not equipped to judge that quality effectively.
This makes me think you're only exposing yourself to high quality writing online and from an intelligent circle of friends and coworkers. The average person's reading and writing abilities are _atrocious_ and only getting worse. We're almost at the point where kids are communicating through abbreviations and emojis exclusively. LLM prose is significantly better than what the average person can produce.
At some point you're just making bad excuses for false scarcity.
But that's the real problem with LLMs. Culture is aspirational. It has a consistent goal - find the best, highlight it and distribute it so others can build on it.
LLMs are the opposite - produce as much of everything as possible at the lowest possible barely-acceptable-if-you're-lucky quality.
This was already a problem before LLMs. Mass market content farms - Kindle Unlimited, Wattpad, Spotify, social media in general - give everyone an equal voice, with mass popularity and "likes" as the only metric.
Now LLMs are automating the creation process, so everyone gets more of everything.
Except inspiration. Not so much of the "That's astounding, I wonder if I can learn from that and reach for something in that league."
I think this is open to debate. To me, the code has always been the goal, and the fact that writing it sometimes serves to produce a product is important to others (and what brings the paychecks in), but ultimately not something I've ever been excited about or interested in throughout my career. So I judge a developer based on the beauty and quality of the code he produces, just as I judge an LLM by the same sorts of things.
The fact that AI can one-shot a working CRUD app is not really that interesting to me. If it could make the code beautiful, concise, maintainable, extensible, minimal, performant, readable, and bug-free: a work of art and love that a craftsman would be proud of... that would impress me.
I mean that's certainly one way of looking at it, and both can be impressive technical feats. But most people judge carpenters and artists on their end products, their overall vision, their motifs, their philosophy, and so on. On the other hand, as a trained logician, I definitely see proofs (which, by the Curry–Howard isomorphism, are computer programs) have some degree of beauty-within-themselves, but that's quite hard to achieve. Not everyone is a Gödel, after all.
I also think programming languages, despite being Turing complete (which is frankly not saying much), are far too limiting to truly construct magnificent things with.
> artists on their end products, their overall vision, their motifs, their philosophy, and so on
The main output of programmer's work is their understanding of the system they work with, the rest comes from that. Behind the code there's its author's intention, vision, their tastes, philosophy and experience that makes them tackle problems in specific ways. Code review is, aside of quality assurance, mostly about communication between people, convincing them to your ways of doing things (or getting convinced by others) and communicating needs. It's what keeps projects running and what makes people improve their skills.
You don't need to see magnificence in code to realize that there's more to it than just the syntax tree to compile.
I feel like I need to push back here, because some of the best programmers around: Carmack, Torvalds, Johnathan Blow, even folks that make programming languages like K&R, Rob Pike, etc. are judged on their respective end products, not on minutia found in code reviews. For example, if I asked you "why do you think Stroustrup is a good programmer?"—you wouldn't cite some obscure optimization he came up with, but would rather talk about his overall vision for C++, his ideas of evolving C, his staunch anti-GC takes over the years (and their justification), etc.
Frankly, I have no real idea of how good Carmack, Torvalds or Blow are as programmers, I have never worked with them so I don't really have a way to tell (even though I do contribute to Linux and I've seen some of their code). They're likely past a certain above-average threshold, but they haven't got famous for their programming skills.
That said, if you think Torvalds isn't being judged on "minutia found in code reviews", I'm not sure your take is very serious in the first place - that's the main thing he was being judged on for decades now :)
How?
> you mention things that are very clearly not end products but thoughts and visions behind them that only lead to end products
Thoughts and visions are much more closely intertwined with end products (in fact, likely supercede them) than some random code review is, so I'm not seeing where the contradiction lies.
> that's the main thing he was being judged on for decades now
Linus hasn't written any code[1] in at least half a decade+. To argue that he's being judged on his code misunderstands why Linux became so popular to begin with.
[1] https://linux.slashdot.org/story/20/07/03/2133201/linus-torv...
> Code review is (...) mostly about communication between people, convincing them to your ways of doing things (or getting convinced by others) and communicating needs. It's what keeps projects running and what makes people improve their skills.
The way he does that is exactly what most news stories about Torvalds have been focusing on for many years now. In practice, unless you run a project alone, code review is where thoughts and visions surface up the most. Or, well, should be - not everyone is good at it.
(that said, even though my point is that's he's obviously not being judged on his code, you can easily find code that he wrote as late as this month, so your statement is clearly wrong even if that doesn't really influence the discussion here - code review is still the vast majority of his job, just like he stated there under your link)
Could be both :)
The way I look at it is like this, and you could call this my thesis: I do not categorically think that code in itself is primarily relevant to us looking at a "software engineer" and saying "wow, she's good." The product (the Linux kernel, in Torvalds' case) is, on the other hand, what actually matters. I think we're getting caught up on the idea of a code review; a code review can serve many purposes, as a code review is basically just people talking about the code, the product, their feelings, and so on. Sure, sometimes it's like "this `i` should be a `j`", but other times it's "this should serve feature X, not feature Y."
Overall, I don't think Torvalds is judged by his code quality. And the snippet I cited is the man himself saying "I don't write code anymore" so I took that at face value, even though my conviction stands wether or not he actually does still write code. I don't think anyone actually cared that much about his code quality (maybe with the caveat that the kernel didn't crash).
PS: I could be totally wrong, and this is an interesting & stimulating conversation, regardless.
I write code for a living mostly by hand. In the odd case where I need help I still use google like I always have. I spend more of my time in meetings or staring at the ceiling than writing code. This was also true a decade ago before LLMs. It was also true several decades ago when someone else's ass was in my seat.
Way back in the past (around 30 years ago) I remember reading an article on "how to read a book" or a similar subject. They argued that, you should not skip the acknowledgments, preface and other "personal" related sections of a book, because it was there where you got a glimpse of the person that was writing the book. The idea being that, you should had in mind that the person writing was explaining something through you.
Carl Sagan even has a video where he argues Books/Writing is some sort of communication through time.
Now, this has been the case historically: A person writes some text (even in botched language like my writing, as English is not my first language) with thinking that someone else in the future will read the ideas and reason about them.
But what about text written by an LLM? Does it have inherent intention? When reading LLM text, it feels like looking at those "this is not a person" photos. Yeah, they are words, yeah they form sentences and paragraphs but... they lack "soul".
If so, this seems to be a trivial (still worthy) assertion.
For example, I intend to, say, construct a shed. I make mistakes that I only see because I actually constructed. I revise future endeavours involving sheds.
I admit to not having read this piece, and am merely reacting to the title.
—-
Okay, I got through the first paragraph of Walter’s writings. While I nod to the bitterness (I assent to the existence of it), I do not bow.
What I find to be the normal pattern (by intuition) is that the condensed leading text belies the expansive following text. This is likely lazy (a shortcut) and I am open to correction at your effort. If a call to your effort (I apologize) is unpalatable then I concede.
Maybe? That is one reason to read, but there are a lot of other reasons, too. It doesn't mean you are doing it wrong if you want to read something and don't care at all about the person who wrote it.
At least in the USA: 21% of adults in the US are illiterate in 2024. 54% of adults have a literacy below a 6th-grade level [1].
1: https://www.thenationalliteracyinstitute.com/2024-2025-liter...
54% of adults have a literacy below a 6th-grade level is simply not a some kind of catastrophe and it does not mean those people cant read. There is this idea that 6th-grade level is almost like not knowing how to read, but that is simply not the case at all.
How do you think the author of the page would read this? That sounded pretty asshole-like for me. If it's not for you I'm really sorry for you, you must have to endure really screwed up people.
And sorry about your blog :/ didn't know it was hacked. Looking at the comment section of the hello world though it gets pretty obvious LOL. You should consider removing it from your HN about though.
FWIW, I'm with GP. It's quite easy to get just mind-numbingly tired reading beyond the first two sentences of a typical LLM output, let alone on something I'm familiar with.
Same to you though, have a nice day
The advantage of the writing vs images, is that it takes longer to absorb the whole with text, so its less apparent that the whole thing doesn't quite come together.
My problems was with Claude's prose and ideas is that it kept recycling the tropes and phrases after a while - something that has been observed that these models have very strong statistical biases - when asking for a random number for example, LLMs are far more predictable than even humans, this shows up in unguided writing exercises.
But as for actually crafting text that is both terse and to the point - such as oneliner explanations, or writing summaries - these models are quite bad. The best I have seen is they could turn a given length of prose into an even longer version - with generally some loss in the tonal accuracy or the points made in there.
As such they are a terrible tool for professional communication, but unfortunately, lots of people have started using them for exactly that.
I get that it sounds clever but that's the damn problem!
However, in fiction I’ve found it a useful collaborator. There have been more than a few occasions when, given some notes of how I want a character’s arc to develop in a particular scene, that the LLM gives some excellent pointers and ‘new’ ideas I hadn’t considered.
As far as editing my prose, I use it as a ‘thesaurus of phrases.’ When lazy, I can give it a rough sketch of the paragraph, giving it the gist of what I want, and have it generate a dozen or so versions. I usually can find nuggets of good phrases therein that are useable… much as I would refer to Roget’s to find a more precise word.
That said, one has to resist tbe temptation of using a chunk of generated text verbatim; no matter how good it sounds in isolation, the repetitive grammatical structure and other LLM-smells add up quick and become nauseatingly obvious if used frequently.
In any case, I think LLM’s get a bad wrap for writing… when used correctly, it is incredibly useful. And, it’s tiresome to hear pretentious snobs assume that an author who uses LLM simply lacks the taste to appreciate how bad the prose sounds. Not true in all cases.
And both come from ovens.
This is true for coding, too, which I think, to a large degree, might explain the polarized differences in opinions on HN about the quality of LLM-produced code. You have the 1. "AI produces code better than I could possibly write, one shots things it would take me days to do, and has made me 10X more productive!" camp, and you have the 2. "AI constantly produces poor code needing rework, makes mistakes, has to be babysat, and ultimately costs me time!" camp, with a spectrum in between those. How could the output of the same product be seen so differently? Well, I have bad news for camp 1...
The industry largely has selected for camp 1 long ago.
If you don't get immediate negative feedback camp 1 can go quite a ways before problems surface.
It's bad if they work in a part of the industry where code quality or efficiency matters. That's maybe 10% of the total though.
I am slowly doing more of my own code and cutting out the LLM out of the loop in the unfamiliar territory I am working in.
My main concern is not so much productivity but understanding the code I have written and feeling agency over it.
The LLM is a very good teacher.
The language that I picked for the game runtime is Python. Claude really thought that the best way to validate user submitted Python was to bypass the WASM sandbox and execute it within the application container using shell exec - essentially opening up an RCE vulnerability.
I also find that the quality of Claude Code degrades substantially. Claude really wants to implement every feature in as bespoke way as possible. This is fine when you first generate the project but over time you'll find that every web modal is implemented differently. Every button is different. Business logic is disconnected. It's why agentically produced codebases are MUCH larger than they should be; every feature is developed in a vacuum.
Then I'm trying to shove stuff in my AGENTS.md or CLAUDE.md files like "ALWAYS look for existing patterns within the codebase to keep it consistent." But the harness doesn't always work and it'll generate useless, verbose code anyways.
In some cases it's useful - like if I am shaky on the DSA knowledge needed for a specific operation or optimization then Claude can replace Stackoverflow. But, man, I'm so frustrated with it.
I am trying something similar: basically a LLM managed MUD for people with too little time for roleplaying.
If you want to chat, hit me up at mail-from-ohsohumble@f12n.de
I just had Opus 4.7 build a feature twice, because it didn't close a ticket the first time. (I'm trying to solo-build a fairly large greenfield project, and am at the point where I let it go ham over my codebase because of the scale of things.)
I then spent a couple of hours asking it to compare the features. It argued that they were completely different features for a while, then eventually acquiesced and said that they were redundant.
That's a couple of thousands of tokens and time I'm never going to get back.
Code should grow according to need, not for its own sake. Start small, use it in the real world, and then improve it.
The majority of a project code are written at the beginning or when a major feature is introduced. The daily work is mostly tweaking. And you can’t tweak without a good understanding of the module.
1) he is extraordinarily lucky
2) he is extraordinary brilliant at manipulating LLMs
3) you really are "holding it wrong" and you are hobbling yourself with your failure to properly learn the tools
The first two seem rather unlikely.
Today, it's a kind of chaos magic wherein you summon the beast and try your best to contain him, knowing that someone will probably die in the process. Sometimes literally. It's still a force multiplier in the right hands and domain, and agentic coding is a paradigm that won't retract, at least until something better supplants it.
The problem is that few engineers actually have the discipline available to constrain these models appropriately and instead rely on a hodgepodge network of "skills" aka prompt fragments which are passed around and glued together.
I consider myself as having such discipline, being strongly architecturally-minded, user-first, etc. in both design and implementation. And I still struggle to contain the beast many days. I just got through screaming at Claude for intentionally taking a shortcut that I'd forbidden, leading to a ton of wasted time and tokens.
Sometimes I feel like I saved weeks of R&D with a single ten-minute task handed off to an agent, other times I feel like I'd get better returns playing slots in Vegas at the alarming rate Claude burns through money.
The hard part too is it's not like you can just learn the basics and be able to tell good code apart from bad -- the more you learn to code, the more intricate your understanding of good code is. It's like becoming a good writer; just knowing grammar and spelling doesn't make your writing interesting. Not to mention that there's just a lot of bad advice out there that you can't recognize as bad advice if you're not a regular practitioner. Like, "Clean Code" is IMO a terrible book, but a ton of people follow it because it has the sheen of respectability.. until, hopefully, they learn some new patterns and realize those old ones aren't very good. But you pick these things up with experience and doing the work! Otherwise if you're just reading other peoples opinions, you'll see a bunch of people say "Clean Code is great" and a bunch of other people say it's rubbish, and you'll have no way to know who you should listen to. (If you disagree with me on Clean Code the book that's fine -- I'm just using it to make a point -- sub in a different book/ideology if it suits you)
I think looking at an LLM code and thinking you're now a coder is like watching a someone play guitar and think you can just pick up a guitar and play a song. The truth is, if you want to be good, you have to do the work.
One of the things I hate about AI is that we're going to have a generation of "programmers" that are absolutely shit at programming, create problems for everyone else, and will have absolutely no idea how bad they are. And they'll probably never get better, because you can't get better by just asking claude to do shit for you. And then the LLMs themselves will probably start to degrade because they'll be trained on the slop since it'll heavily outnumber handwritten code..
So many posts here on HN claiming they created another useful tool with AI.
No, you didn't create it. AI did. You only had a supporting role. You're Ringo Starr and the AI is John Lennon.
It's even worse than that: you're the Ringo Starr and the AI is a a "John Lennon" who sucks and is boring and uncreative.
No, AI assembled bits of code written by hundreds of programmers before you.
Then again, I just caught Claude writing setTransparent(!opaque == false), opaque being a bool, on a purely vibecoded project. Which was pretty impressive. ("• You're right, that's nonsense.")
For bigger production ready code, you indeed have to guard the architecture. But for the code, in some corners you can get away with sloppy code, as long as it kind of works.
What I'm saying is, code doesn't always has to be great. You will just have to judge the places where it needs to be high quality, and other places where you can get away with sloppy code.
So there are still only two camps
I have been coding since about 1983 or so. I shipped high quality products that have been used by millions of people. From embedded software to desktop applications to distributed systems.
I don't think I'm in the "don't understand what code should look like camp" (I mean you never know but the evidence seems to show that I do know what I'm doing). I use AI as a tool and it helps me be more productive. I don't "one shot things that would take me days to do". I use it to help me automate things that I could do manually where it is faster and more effective. I review every step and if I don't like something I adjust. There are some specific situations where it basically does as good a job as I would do in running some experiments, doing some analysis or writing some small amount of code. I still know what the changes need to look like broadly, where to make them, and what patterns to follow. It just automates the work and sometimes does have some additional insight that can complement my views. Unlike me it is all knowing about everything in terms of access to "knowledge". It knows all the details of how a certain runtime manages memory, Linux internals and various open source software. I could go look it up myself (which I'd do before AI) but I don't hold it all in my head like AI basically does. It is also "all knowing" in the code base I work in (more so than me, it's a huge code base, I have an outline and a high level picture in my head but not every single code line) where again I can dive into the code but it helps me extract the relevant information faster.
I think the polarization is more on the how you use the tool, what situations you use the tool for, which domain are you operating in (languages, applications etc.). You can also one-shot simple tools and helpers that are not the production software which is another way to accelerate your workflow.
That simple arcade game (without in game transactions) needs to be fun, that website that needs to attract visitors (but not sell them anything or handle sensitive data)?
They don't care about abstract code quality, they care if it works and useful.
So a good coder here means he or she could get to working results according to what the client wants fast. And those things likely make up the vast majority of written code. So no wonder AI gets adopted as it is a powerful tool here to be even faster.
Not all code runs in airplanes, handles financial transaction or sensitive user data - for this you need the best code possible and nothing vibe coded or quick and dirty hacks.
And oh wonder, it is possible to combine both. Because yes, websites often include financial transactions nowdays, but that part can and should be handled with care. People who move slow and check things. And then those who are quick to build things on top of it.
But I strongly object to dividing programmers absolutely in good and bad programmers, when the field is so big and the requirements not the same.
Some optimize in speed, some in quality. And yes, some are just bad in both. And some can do both - but they are very rare, in my experience.
As far as I know basically all of the successful software companies had these quite early. Of course, you need other kinds of people too. And not everybody needs to be like that. But you absolutely need those kind of people.
But if you give me a few examples where this was not the case, and not recently, or during the dot com boom, where hype overwrote everything, then I’ll change my mind.
The most successful projects which I’ve seen closely, all of them had only a few people who mattered, everybody else could be replaced at any given time basically, without a real impact. All of the failing ones were those in which those people didn’t exist, or were too few of them. This is exponentially more important in early phases of projects.
It seems people really want to not believe AI can be useful for strong developers. That's fine. I don't really have a bone in this game, I'm anonymous here, and people can think whatever they choose to ;)
Maybe not.
I'm just waiting for somebody who send me code, which was generated by AI, not overwritten almost completely, and it's not shit. The funny thing is that some people here were so convinced that AI is great, that they recorded how they work with AI. And two things:
- They were slower than manual copy pasting
- They still somehow introduced bugs, and very suboptimal solutions...
Also, it would be good, that anybody could show me anything, that shows, how people became not terrible with code review suddenly. Because before AI, it was a common knowledge, that almost everybody was bad with it, and people rarely did it properly, because it was considered annoying, and not because they were useless. There were jokes about rewriting things, exactly because of the same reasons, and they heavily based on reality. And suddenly, we pretend that this changed.
And somehow you should really would need to convince me that the 100s of thousands of lines of code which I generated with AI in the past years, somehow, it's better than what it is. But I'm sure, that it's easier to assume that I didn't try something, than showing only once what "good" means in this case. Unfortunately, there is nothing similar here, than for example "Groovy is a great programming language", which is a dead giveaway that whoever said that is not just bad developer, but somebody who I would fire immediately from every single project to which I'm related to. Especially if they are tech lead. But there are such people, and some of them would claim the same thing as you. // Obviously interns and juniors can think whatever they want. They are labeled as such, because they cannot know yet.
Groovy ;) You must love Jenkins. At least I can rest at ease you won't fire me for that infraction. I used C and C++ most of my career and more recently Go (which I really loved when it was created but my take is a bit more nuanced these days after seeing a really large code base evolve over years).
I'm confused though. You generated 100s of thousands of lines of code using AI and you think it's crap? The code AI generates for me is not some pinnacle of software engineering. It is repeating existing patterns or fairly simple concepts. I treat AI like an quick intern that scales infinitely (or a junior developer). And yes, juniors and interns don't write the best code but in many organizations there is still a fair amount of code written by them.
The thing is that in a large team/project (and the one I'm on has hundreds of developers of various skill levels) there's an endless backlog of things that can be improved including relatively easy features or refactoring. The constraints are either organizational or time. AI enables these things to get done with very little overhead so that's a net positive. It moves the needle for how much time/effort does it take to address "not that hard" issues and with proper prompting and examples it does a decent job. The bar isn't code that John Carmack would write in a week, the bar is improving a certain crappy area of the code to be more reliable or more performant or a little bit cleaner. This is life for most software projects. Yes, in a perfect world every software project is perfection. And maybe some organizations are able to approximate that.
At least, I can be sure that we are not near the same level. But at least, you hopefully will recognize the same thing with new languages… without seeing them failing first.
> You generated 100s of thousands of lines of code using AI and you think it's crap?
This is a funny question. First of all, there are people whose job is to test LLMs. However, I’m not one of them. I simply tried them, generated, and still generate a ton of code with them, then I rewrite basically every single line of them. Because they use for example outdated patterns, which causes the same problems what you’ve seen with Go.
> It is repeating existing patterns or fairly simple concepts.
Yes, and most of the most popular ones are mediocre the best. Average code from which LLMs are learning are made by beginners, not experienced ones, because their sheer number. So LLMs will use those.
> I treat AI like an quick intern
This is always the funniest sentence regarding this. Before AI, it was quite well known that you don’t ever allow interns near important parts of the code. Now, people who supposed to know this, and the reasons for this, somehow forgot this aspect also, just like the review thing.
> AI enables these things to get done with very little overhead so that's a net positive.
No, it does allow to tick a ticket in Jira. And if you handle this in any other way, then you will fail miserably, as how for example Microsoft quite openly did with this.
> a little bit cleaner
Ah yes, the infamous “cleaner”, about which the exact opposite is quite well known, and it’s quite obviously not true with every single vibe coded projects, without exceptions. If that’s cleaner in any environment, then I have a bad news: you’ve never worked with even medior developers, ever. Seriously, that code quality, especially architecturally, is junior level shit.
My previous boss did these low hanging fruits, he at least would never tell anything more than “it’s better than nothing”. And only regarding non-important code, which can fail without real consequences. And can be shit, obviously. The whole point was that even shit is better than nothing. Not that it’s acceptable quality in any way.
At least, you were obvious at least, that your “success” is magnitudes different, than mine. And not regarding code quality, but when a project/product successful. I completely forgot that I’ve met people who sold that their teams completed the most tickets at a company in a given time frame as success. Probably we are closer than this, but still very far away.
I think you're saying "my bar is so high you even can't understand where it is". I've worked with hundreds if not thousands of software developers, in many companies from startups to well established ones, including producing products that are what I'd call critical infrastructure that work reliably and do what they're supposed to do. I think I have a pretty decent idea of what an "average" software developer looks like and the overall shape of that curve, and similarly the architecture/design curve of various real world projects. I've built software on my own as a team of one and I've worked with teams of more than 100 people. Anyways, if your assertion is what I wrote above then clearly LLMs can't replace the mythical programmer that you are. But that's not what they're aiming to replace. As to "vibe coded projects" I already said that's not how I use LLMs and I agree that can easily end up like a pile of garbage (but still has its place in the new ecosystem).
The only real test of software is whether it does what it's supposed to do: reliably, is maintainable, can be extended and evolve without losing these attributes. If you've shipped systems that are used by many, work well, can evolve to support new features etc. - kudos to you.
> the polarized differences in opinions on HN about the quality of LLM-produced code
Are there strong differences of opinion about the quality? I've seen very few people claim that LLMs write better code than they do.
> one shots things it would take me days to do, and has made me 10X more productive
This is an entirely different claim from the former, and you're conflating them.
The boost from LLM-assisted code isn't _expertise_, it's the power of having an always-on team of reasonable junior developers from every discipline you can possibly imagine willing to do your whim.
Take for example Jesse Vincent / obra[0], who is an exceptional developer, with great taste, and a stack of well-received open-source software to his name. He posts a lot on how he's being made more productive by AI-assisted development. Do you have bad news for him about the quality of his work...?
In my case I see Claude produce code much worse than I would, but it's certainly much quicker and, even after reworking, it makes me finish tasks in less time.
The 'camp 1' people in the pre-LLM days were probably the ones that often just copy+pasted code from SO they didn't really understand, but since the code seemed to work when they ran it, they thought it was all fine and continued on.
Whereas the 'camp 2' people when trying to find an answer to something, discarded 99% of SO and other similar answers, having the knowledge to see how they had broken edge cases, were limited in some way, didn't actually solve the underlying problem, etc.
Nowadays, 'camp 1' people just use the LLM output and it "seems to work" and consider it all fine. Whereas the 'camp 2' people still continuously see all the faults with it.
Let's say you are correct.
You ask an LLM to write something for you, and to you it looks really, really good. So based on your conjecture, that means I am not a very good writer.
Ok, but how does that change what I should do? If I am not a very good writer, that means an LLM IS actually better than me, even if it might not be objectively good to an expert writer.
My two choices are to keep producing my own crappy writing, or use an LLM to create better (but not great) writing.
Wouldn't it make sense to use an LLM?
It seems to me your premise leads you to the same conclusion you would reach even if your premise was false; if me thinking an LLM is good at a task means I am very bad at that task, I am probably better off having an LLM do it. On the other hand, if you are wrong, and I think an LLM is good at something because it actually IS good at that thing, then I should also use the LLM to do the task.
Either way, the LLM is better than me at the thing.
From above, or from below with adequate exposure, it feels facile and hollow. It is good at weaving grammatical structures. It is not good at thinking in words in a way that invites a fellow human along for the journey. Because it doesn't think.
Are you also an LLM or do you have the capacity to learn and grow?
If you're not a very good writer, I'll at least skim your work to see if it contains any good ideas. If it's slop, I'll just close the tab. You already told me it's not worth caring about, so I'll agree with your decision.
I rarely write code, and only once for a living. But I feel like I’m a superhuman and one step away from being a zillionaire when Claude gives me a bunch of code it has written in seconds. I WILL CHANGE THE WORLD!!
And then I remember that Claude can’t write words that don’t make me want to break things and I’m good at writing words but bad at writing code.
So then I delete the code and go back to doing more profitable things than being the next zuckerfuck.
But I think (I think?!) you might have misinterpreted ny comment, because I’m saying “yes, I Dunning-Kruger myself regularly with AI! BUT then I wake up from the psychosis and realise what an idiot I am and go back to my normal life”.
AI makes everyone think they are brilliant. The skill is recognising when you’re just another idiot.
This is why code generation is a disaster waiting to happen. Hunderds of thousands of "programmers" with no idea of what they are pushing to production.
Naah I disagree with this. I think LLM's are good at gas-lighting you into thinking that good writing only comes in one flavor. And LLMs prefer a very "textbook/technical-manual" coded flavor of writing because maybe that way they are more useful to us humans. But human writing is not just about crafting the most elegant sentences. Sometimes great writing is just this doggo-drawing meme:
https://knowyourmeme.com/photos/2160304-the-winner-of-this-c...
- “(The) honest answer:” (again, with colon)
- “The thing to internalize:”
- “The smoking gun:”
(really, sentences that start with “The <tag suggesting the next clause is the key point>:” are a strong tell, but those four are the most prolific)
- “load bearing” (when not talking about architecture)
- “blast radius” (when not talking about actual explosives, but rather the effect of an event/action)
- “smoke test” (esp. when “sanity check” is more apropos)
- Lists of three clauses/adjectives where the third is really just a combination of the first two
- Referring to the “shape” of things figuratively
- Social media posts that end with “Curious if anyone…”
- Stories or anecdotes using. “Oh. Oh.” (where the second “oh” is italicized)
Edit: Yes, some of those last ones are terms that we often use as devs...but I would argue about the actual frequency of their use. Plus, these tells live on in prose generated by the latest models.
Assuming you mean load bearing & blast radius, I'd see those used and use them myself very frequently pre LLM, mostly in online discussions though so its telling where they got their training data. Load bearing itself is/was a pretty normal phrase in the ops world in daily discussion.
Smoke test though, I can't say I've ever see irl usage.
If a repo is bare of CLAUDE.md but mentions a smoke test in a commit in the last year I assume it to be LLM written.
We use it all the time at my employer, and have for decades. They're basic tests to tell you if the app is up or not.
Like: you go to this page, and it shows a big green banner if the app can connect to the database and its disk isn't full. If one of those basic things are wrong you get big red banner or you can't load the page at all.
- And a variant of the above is omitting the subject, "happy to" instead of "I am happy to"
- Codex refers to "the spine" of something
- Claude often says some decision is "locked" (i.e. decided on)
Everything is an escape hatch, try catch is an escape hatch, a cli flag is an escape hatch. It makes no sense, and quickly ended up in my “banned words and phrases” md file
I've always called them exit hatches, entirely unrelated to llms...
Now I wonder if I need to reword the docs... But realistically speaking, llms are the only readers of them nowadays, so I guess it doesn't matter.
I think that the convergence of these tics is just a symptom of distillation.
Also, here’s a link to well-documented patterns by Wikipedia: https://en.wikipedia.org/wiki/Wikipedia:Signs_of_AI_writing
"what really Xes" "is genuinely X" "that actually Xes" "is/makes/does/etc a real X"
The real/genuine/actual cluster of words are wildly overused.
Nope. LLMs are RLHF'ed to the brim.
Simultaneously, because humans subconsciously mimic what we see, we also converge to sound more LLM-ish.
The harsh reality is that no matter what you write and how much research you put into it—especially if you try to be legible to others and not make grammatical mistakes—someone could discount all that and claim you just prompted an LLM. If they want, they can always find some magic “AI checker” that will return a high enough probability. We all know that with a good enough prompt and with round-trip validation against a checker (there are definitely products with this all built in) it will avoid the common tells, it’s just the matter of a few extra tokens.
It’s somewhat demotivating.
The LLM sameness in web design is good. Most sites shouldn't try to be idiosyncratic. The best design for a site with real utility is legibility, and LLMs are better at that than the median developer. Always laying out the same buttons? Always using the same type scales? Good! If it looks good to you, you weren't going to do better on your own, and you were very likely to do worse.
See, I disagree. Having seen plenty of Claude generated websites and slide decks, to me it just screams "no effort whatsoever". AI sloppypasta for content, if you will.
If I can see within a few seconds that your website or slide was obviously AI generated, I will doubt its content, how much effort (if any) you've put into it, if it won't have hallucinations, and (especially for websites) if it's even real or a scam farm.
I'm not saying every website has to be unique, but at least tell your prompt to use a font or colour scheme or something specific to you that will make it seem like you've put in some effort and make the result stand out from the slop.
Maybe here lies the crux; for some of us, the web and by extension the internet is about expression and individuality in a way, but all together all accessible by everyone. Everything looking the same instead looks conformist, and ultimately boring, which I guess is what many of us don't want day after day. We want new ideas, presented by the person/group who came up with it, in the way they would express it. The LLM kind of trashes the parts that add up to something interesting.
It's not just about being the same, in which case the worst is that it has less to stand out with.
It's about looking the same as the other AI generated slops out there, which is a strong signal that the content is probably AI slop too, which doesn't merit time being spent on it and trying to understand if it's slop or a human that just used AI generated UI.
I would feel very weird using LLMs for writing, except for filing out stupid applications. I've had collaborators use LLMs for some technical writing and it's pretty much always borderline nonsense that has the aesthetic of something correct. For creative writing, I feel like heavily using an LLM would defeat the purpose :shrug:
I have not used to review my writing yet. Is it that good?
I'd say that sameness isn't good either from a LLM or solo founder SaaS #171876137.
> Contrastive negation is a rhetorical structure that denies a specific idea in the first half of a sentence and asserts an alternative in the second half.
> It typically follows an "It’s not X, it’s Y" or "not just X, but Y" formula.
Wikipedia also has a great resource which covers many of the common LLM patterns: https://en.wikipedia.org/wiki/Wikipedia:Signs_of_AI_writing
And I especially appreciate the idea that these patterns aren’t the problem. Don’t just fix the patterns. The problem is they point to hastily or carelessly written content.
No ___, no ____. Just _____
"Smooth. Effortless. A perfect fit for your needs".
In any style of informal or persuasive writing this shows up , as if it has to drive the point in.
I kind of wish we'd stop talking openly about what the tells are. It's nice to be able to determine with fair accuracy - but it couldn't last forever.
Least this way it’s out in the open perhaps, since enough users have training enabled labs will naturally learn what annoys us.
Had the same thought though
Labeling each sentence (J)ab and (T)hrust, and using colon ":" to indicate arguments, one gets:
```
J: J. J. T.
J: "J. J. T".
T: T.
T: J. J. T.
```
Also the false not X it's Y is used in a similar way for faux distinctions like a sov cit claiming "it's not driving, it's traveling in a car"
Will be interesting if that holds in other areas when chasing super intelligence.
My favourite one today from today:
“The tax isn't the problem. The mindset is.”
I haven’t tried it with Slack messages because I’m a little scared to read what it says, haha. But the same concept surely applies.
There are a few people at work who are aggressively using Claude to write Slack messages. It’s easy for me to tell because one day they’re writing barely coherent English in multiple messages, and the next they’re sending perfectly coherent prose in a single message.
It actually works quite well
Thought for sure we'd get a critique of Inter overuse. JetBrains Mono is a lovely font, though.
Is it even legal to use it in $RANDOM_WEBSITE?
If JetBrains Mono suits your work you should use it, not avoid it specifically because AI often uses it.
<somebody> just <did something>.
And it changes everything.
Like corporate manager-type emails, of which I get AI generated ones frequently from company ownership. They think LLMs are the best thing since sliced bread.
It's taken corpo-speak to an entirely new level. On the plus side I no longer have to read them, and can just have AI reply on my behalf with more fast food.
In coding, I've noticed a few tropes as well: everything is a "contract" or an "artifact" (clearly trained on like three decades of Java lol), everything is constantly "backwards-compatible" or "versioned" (even if working on a brand new greenfield project), and a few others.
It's been used in an ops context for a long time, pre LLM even. Same with "blast radius" has been a cybersecurity term for as long as I can remember.
It can't get creative, it can only generate based of what's a majority in it's training data.
> "belt and suspenders"
Instead of this:
def add_three_ints(x: int, y: int, z: int) -> int:
return x + y + z
def add_three_ints(
x: int,
y: int,
x: int,
):
return x + y + z
If you have to add arguments, when they're on one line like that, the diff is cleaner, so the reviewer has an easier time kf understanding what's going on. That is, if you still have a human reviewing code, that is.
> By default, Black uses existing trailing commas as an indication that short lines should be left separate, as described in the style documentation. If this option is given, the magic trailing comma is ignored.
https://black.readthedocs.io/en/stable/usage_and_configurati...
It should then keep the lines intact, unless they exceed the line length. I have mine set to 120.
I can also see how it is not necessarily the case, since all models seem plagued by having unique cliche patterns. We all seem to experience the writing smells from the same models. I see 'cleanly' a lot less from GPT 5.5 vs <5.4
Why would you do that in the first place? If you are just starting writing about math, I assume your goal would be to get better at it, and using LLMs is not how you get better at writing.
I grew up seeing this very relevant image on the wall of my grandparents:
https://en.wikipedia.org/wiki/Matthew_7:13#/media/File:Der_b...
Lower quality version with The Eye of Providence not retouced out (the version I remember):
https://commons.wikimedia.org/wiki/File:Der_breite_und_der_s...
free will kiddo
Kind of works.
also, how come jetbrain mono has become one of these tells is insane to me!
I think it makes a lot of sense, it's essentially an inversion. Another related thing I recall is hearing game creators talk in terms of "gimmicks" and "scenarios". Anything that gives a structure, a framework to operate within, I'd say is always useful. This is much the same.
For those curious https://github.com/ryanthedev/oberskills/blob/main/commands/...
sorta like asking an LLM to get more creative with a visual design, that also never works.
Last month it was quiet things being said out loud.
When you use the tools all the time it’s fun to watch these things pop up a week or so later as normies copy and paste the slop out on LinkedIn and their newsletters.
Did Anthropic and/or OpenAI deliberately train their models to produce websites with a specific design language, or did these stylistic preferences emerge naturally as some kind of LLM-selected optimum?
When you bring your own ideas you can get AI to dev pretty nice looking non-generic stuff.
To give an example I'm personally frequently annoyed by, Google's Antigravity will consistently use the word "anthropomorphic" while "thinking" and the end result will consistently have obnoxiously large border radius (kind of like Android's design language).
Codex on the other hand likes to make websites with blue elements on a black background and likes to use emojis for icons for some reason, which is a terrible idea accessibility-wise.
To list out a few personal examples:
1. Several students I knew at GaTech in 2024 taking the algorithm class (which is notoriously hard) started submitting assignments with sub-optimal/brute-force algorithms cuz the TAs kept reporting them for academic misconduct on optimal solutions.
2. I've started avoiding "em-dashes" in all my writing
3. Junior engineers leaving "typos" in their code reviews or submitting code-reviews with absolutely 0 comments (LLMs love to leave verbose comments)
Gotta stop shaming people for using AI fr
/endredditmode
I actually love folks documenting this. I'm all for LLMs producing rough drafts. But rough drafts are, as a rule, slop.
It is strange to read as the topic A has often not been introduced and introducing it by saying what it is not makes very little sense to a new reader.