LLM output is crap. It’s just crap. It sucks, and is bad.
Still don't get it. LLM outputs are nondeterministic. LLMs invent APIs that don't exist. That's why you filter those outputs through agent constructions, which actually compile code. The nondeterminism of LLMs don't make your compiler nondeterministic.
All sorts of ways to knock LLM-generated code. Most I disagree with, all colorable. But this article is based on a model of LLM code generation from 6 months ago which is simply no longer true, and you can't gaslight your way back to Q1 2024.
I think there's a decent chance that folks on the fringes find interesting uses for immutable and programmable distributed ledgers, even if the prevailing culture of crypto is hellish.
> Either that, or live in some futuristic utopia like the EU where banks consider "send money to people" to be core functionality. But here in the good ol' U S of A, where material progress requires significant amounts of kicking and screaming, you had PayPal.
Couldn't agree more on many of these points. There is so much 'whatever' everywhere on the web that I legitimately don't understand people being interested in the platforms that suck everyone's time. Its frustrating as someone who used to enjoy the early web a lot and it's frustrating to see people that I have a lot of respect for buying into these awful systems with their time and attention. Worse still, I'm something of an outsider in many situations for opting out of them.
The author lost me a little on the AI rant. Yes, everything and everyone is shoving LLMs into places that I don't want it. Just today Bandcamp sent me an email about upcoming summer albums that was clearly in part written by AI. You can't get away from it, it's awful. That being said, the tooling for software development is so powerful that I feel like I'd be crazy not to use it. I save so so much time with banal programming tasks by just writing up a paragraph to cursor about what I want and how I want it done.
Broadly agreed with all the points outlined in there.
But for me the biggest issue with all this — that I don't see covered in here, or maybe just a little bit in passing — is what all of this is doing to beginners, and the learning pipeline.
> There are people I once respected who, apparently, don’t actually enjoy doing the thing. They would like to describe what they want and receive Whatever — some beige sludge that vaguely resembles it. That isn’t programming, though.
> I glimpsed someone on Twitter a few days ago, also scoffing at the idea that anyone would decide not to use the Whatever machine. I can’t remember exactly what they said, but it was something like: “I created a whole album, complete with album art, in 3.5 hours. Why wouldn’t I use the make it easier machine?”
When you're a beginner, it's totally normal to not really want to put in the hard work. You try drawing a picture, and it sucks. You try playing the guitar, and you can't even get simple notes right. Of course a machine where you can just say "a picture in the style of Pokémon, but of my cat" and get a perfect result out is much more tempting to a 12 year old kid than the prospect of having to grind for 5 years before being kind of good.
But up until now, you had no choice and to keep making crappy pictures and playing crappy songs until you actually start to develop a taste for the effort, and a few years later you find yourself actually pretty darn competent at the thing. That's a pretty virtuous cycle.
I shudder to think where we'll be if the corporate-media machine keeps hammering the message "you don't have to bother learning how to draw, drawing is hard, just get ChatGPT to draw pictures for you" to young people for years to come.
Love this writing. One paragraph hit very close to home. I used to be the guy who could figure out obscure scripts by google-fu and rtfm and willpower. Now that skill has been completely obliterated by LLMs and everyone’s doing it- except it’s mostly whatever
> I don’t want to help someone who opens with “I don’t know how to do this so I asked ChatGPT and it gave me these 200 lines but it doesn’t work”.
Well they even call them "content creators", is there anything more whatever? It's literally "whatever" content needed to load the ads around it. It's not painters or musicians or documentarians, it's content creators.
Wow, the lack of self-awareness on the part of the writer of this is stunning. To say "I don't go out of my way to dunk on people who use LLMs", and then write a multipage blog post doing exactly that is fascinating.
You do you, and I'll do me. Perhaps spend more time coding which you say you like to do and less time sneering at people who use different tools.
> Quitting programming as a career right now because of LLMs would be like quitting carpentry as a career thanks to the invention of the table saw.
The reaction to that post has been interesting. It's mainly intended to be an argument against the LLM hype! I'm pushing back against all the people who are saying "LLMs are so incredible at programming that nobody should consider programming as a career any more" - I think that's total nonsense, like a carpenter quitting because someone invented the table saw.
Analogies like this will inevitably get people hung up on the details of the analogy though. Lots of people jumped straight to "a table saw does a single job reliably, unlike LLMs which are non-deterministic".
I picked table saws because they are actually really dangerous and can cut your thumb off if you don't know how to use them.
This reminds me of the normie/autistic/sociopath triangle. The idea is that sociopaths can see through normies, normies can see through autists, and autists can see through sociopaths - when there's a sociopath, often the normies in the group will be easily fooled by him, but the autists will be onto him right away. Don't know why that is, but it's true in my experience.
Same with AI. I'm notably more autistic (or more aspie, or whatever) than my friend group, and also I much more easily recognize AI text and images as uncanny slop, while my friends are more easily wowed by it. Maybe AI output has the same "superficially impressive but empty inside" quality as the stuff that sociopaths say.
I read through the essay and really resonated with some parts and didn’t resonate with others, but I think they put some words to the feelings I have had on AI and its effect in the tech industry
> There are people who use these, apparently. And it just feels so… depressing. There are people I once respected who, apparently, don’t actually enjoy doing the thing. They would like to describe what they want and receive Whatever — some beige sludge that vaguely resembles it. That isn’t programming, though. That’s management, a fairly different job. I’m not interested in managing. I’m certainly not interested in managing this bizarre polite lying daydream machine. It feels like a vizier who has definitely been spending some time plotting my demise.
I was several minutes of reading before this paragraph when the idea hit me that this person hates managing. Because everyone I’ve met who hates using AI to produce software describes to me problems like the AI not being correct or lying to them if the model thought that would please you better, and that’s my experience with junior engineers as a manager.
And everyone I’ve met who loves AI at some point makes an analogy to it, that compares it to a team of eager juniors who can do a lot of work fast but can’t have their output trusted blindly, and that’s my experience with junior engineers as a manager.
And then anyone whose been trying to get an Engineering manager job over the past few months and tracking their applications metadata has seen the number of open postings for their requirements go down month after month unless you drop the manager part and keep all the same criteria but as IC
And then I read commentary from megacorps about their layoffs and read between the lines like here[1]
>… a Microsoft spokesperson said in a statement, adding that the company is reducing managerial layers …
I think our general consternation around this is coming from creators being forced into management instead of being able to outsource those tasks to their own managers.
After reading "...the press release and in the fine print it says that now it can count the number of letters in “Mississippi” correctly or whatever" tried to count letters in another word :)
You said:
how many letters are in the lithuanian word "nebeprisikaspinaudamas"? Just give me one number.
ChatGPT said:
23
You said:
how many letters are in the lithuanian word 'nebeprisikaspinaudamas'. Just give me one number.
ChatGPT said:
21
> What are we actually saying here — that even Microsoft has to evaluate usage of “AI” directly, because it doesn’t affect performance enough to have an obvious impact otherwise
I agree with a lot of this at the outset, but don't really like the gloomy outlook. I don't think there's much to gain by writing off all this unfortunate stuff as people being stupid and greedy. I mean sure, that may be true, but you can flip it around and say that it's impressive that we have it as good as we do despite having to co-exist with stupidity and greed. Better yet, you can see it as a challenge to overcome.
And I'm not the only one saying this but - the bit about LLMs is likely throwing the baby out with the bathwater. Yes the "AI-ification" of everything is horrible and people are shoehorning it into places where it's not useful. But to say that every single LLM interaction is wrong/not useful is just not true (though it might be true if you limit yourself to only freely available models!). Using LLMs effectively is a skill in itself, and not one to be underestimated. Just because you failed to get it to do something it's not well-suited to doesn't mean it can't do anything at all.
Though the conclusion (do things, make things) I do agree with anyway.
IMO, this is more ranting about people who meet the metrics of the platform.
You're a platform drone, you have no mind, yada. Yet, we are reading the author's blog.
The author may hate LLMs, but they will lead to many people realizing things they never were aware of, like the author's superficial ability to take information and present it in a way that engages others. Soon that will be a thing that is known. Not many will make money sharing information in prose.
What the author refers to as "LLMs" today, will continually improve and "get better" at everything the author has issues with, maybe in novel ways we can't think of at the moment.
Alternative take:
"Popular culture" has always been a "lesser" ideal of experience, and now that ontological grouping now includes the Internet, as a whole. There are no safe corners, everything you experience on the Internet, if someone shared it with you, is now "Popular culture".
Everyone knows what you know, and you are no longer special or have special things to share, because awareness is ubiquitous.
This is good for society in many ways.
For example, with information asymmetry, where assholes made others their food, it will become less common that people are food.
Things like ad-driven social networks will fade away as this realization becomes normalized.
Unfortunately, we are at the very early stages of this, and it takes a very long time for people to become aware of things like hoaxes.
> Like, just to calibrate here: you know how some code editors will automatically fill in a right bracket or quote when you type a left one? You type " and the result is "|"? Yeah, that drives me up the wall. It saves no time whatsoever, and it’s wrong often enough that I waste time having to correct for it.
I have not yet figured out why anyone would choose this behaviour in a text editor. You have to press something to exit the delimited region anyway, whether that be an arrow key or the closing delimiter, so just… why did the first person even invent the idea, which just complicates things and also makes it harder to model the editor’s behaviour mentally? Were they a hunt-and-peck typist or something?
In theory, it helps keep your source valid syntax more of the time, which may help with syntax highlighting (especially of strings) and LSP/similar tooling. But it’s only more of the time: your source will still be invalid frequently, including when it gets things wrong and you have to relocate a delimiter. In practice, I don’t think it’s useful on that ground.
76 comments
[ 7.8 ms ] story [ 73.9 ms ] threadStill don't get it. LLM outputs are nondeterministic. LLMs invent APIs that don't exist. That's why you filter those outputs through agent constructions, which actually compile code. The nondeterminism of LLMs don't make your compiler nondeterministic.
All sorts of ways to knock LLM-generated code. Most I disagree with, all colorable. But this article is based on a model of LLM code generation from 6 months ago which is simply no longer true, and you can't gaslight your way back to Q1 2024.
The reason behind banning adult materials has to do with Puritanism and with the high rates of refunds on adult websites.
The irony of this rant next to the AI rant.
Progress is not uniformly distributed I guess.
The author lost me a little on the AI rant. Yes, everything and everyone is shoving LLMs into places that I don't want it. Just today Bandcamp sent me an email about upcoming summer albums that was clearly in part written by AI. You can't get away from it, it's awful. That being said, the tooling for software development is so powerful that I feel like I'd be crazy not to use it. I save so so much time with banal programming tasks by just writing up a paragraph to cursor about what I want and how I want it done.
But for me the biggest issue with all this — that I don't see covered in here, or maybe just a little bit in passing — is what all of this is doing to beginners, and the learning pipeline.
> There are people I once respected who, apparently, don’t actually enjoy doing the thing. They would like to describe what they want and receive Whatever — some beige sludge that vaguely resembles it. That isn’t programming, though.
> I glimpsed someone on Twitter a few days ago, also scoffing at the idea that anyone would decide not to use the Whatever machine. I can’t remember exactly what they said, but it was something like: “I created a whole album, complete with album art, in 3.5 hours. Why wouldn’t I use the make it easier machine?”
When you're a beginner, it's totally normal to not really want to put in the hard work. You try drawing a picture, and it sucks. You try playing the guitar, and you can't even get simple notes right. Of course a machine where you can just say "a picture in the style of Pokémon, but of my cat" and get a perfect result out is much more tempting to a 12 year old kid than the prospect of having to grind for 5 years before being kind of good.
But up until now, you had no choice and to keep making crappy pictures and playing crappy songs until you actually start to develop a taste for the effort, and a few years later you find yourself actually pretty darn competent at the thing. That's a pretty virtuous cycle.
I shudder to think where we'll be if the corporate-media machine keeps hammering the message "you don't have to bother learning how to draw, drawing is hard, just get ChatGPT to draw pictures for you" to young people for years to come.
> I don’t want to help someone who opens with “I don’t know how to do this so I asked ChatGPT and it gave me these 200 lines but it doesn’t work”.
But he did invent JavaScript ...
You do you, and I'll do me. Perhaps spend more time coding which you say you like to do and less time sneering at people who use different tools.
> Quitting programming as a career right now because of LLMs would be like quitting carpentry as a career thanks to the invention of the table saw.
The reaction to that post has been interesting. It's mainly intended to be an argument against the LLM hype! I'm pushing back against all the people who are saying "LLMs are so incredible at programming that nobody should consider programming as a career any more" - I think that's total nonsense, like a carpenter quitting because someone invented the table saw.
Analogies like this will inevitably get people hung up on the details of the analogy though. Lots of people jumped straight to "a table saw does a single job reliably, unlike LLMs which are non-deterministic".
I picked table saws because they are actually really dangerous and can cut your thumb off if you don't know how to use them.
Same with AI. I'm notably more autistic (or more aspie, or whatever) than my friend group, and also I much more easily recognize AI text and images as uncanny slop, while my friends are more easily wowed by it. Maybe AI output has the same "superficially impressive but empty inside" quality as the stuff that sociopaths say.
> There are people who use these, apparently. And it just feels so… depressing. There are people I once respected who, apparently, don’t actually enjoy doing the thing. They would like to describe what they want and receive Whatever — some beige sludge that vaguely resembles it. That isn’t programming, though. That’s management, a fairly different job. I’m not interested in managing. I’m certainly not interested in managing this bizarre polite lying daydream machine. It feels like a vizier who has definitely been spending some time plotting my demise.
I was several minutes of reading before this paragraph when the idea hit me that this person hates managing. Because everyone I’ve met who hates using AI to produce software describes to me problems like the AI not being correct or lying to them if the model thought that would please you better, and that’s my experience with junior engineers as a manager.
And everyone I’ve met who loves AI at some point makes an analogy to it, that compares it to a team of eager juniors who can do a lot of work fast but can’t have their output trusted blindly, and that’s my experience with junior engineers as a manager.
And then anyone whose been trying to get an Engineering manager job over the past few months and tracking their applications metadata has seen the number of open postings for their requirements go down month after month unless you drop the manager part and keep all the same criteria but as IC
And then I read commentary from megacorps about their layoffs and read between the lines like here[1]
>… a Microsoft spokesperson said in a statement, adding that the company is reducing managerial layers …
I think our general consternation around this is coming from creators being forced into management instead of being able to outsource those tasks to their own managers.
I am not really sure what to do with this insight
[1] https://www.cnn.com/2025/07/02/tech/microsoft-layoffs-9000-e...
You said: how many letters are in the lithuanian word "nebeprisikaspinaudamas"? Just give me one number. ChatGPT said: 23
You said: how many letters are in the lithuanian word 'nebeprisikaspinaudamas'. Just give me one number. ChatGPT said: 21
Both are incorrect by the way. It's 22
Oh wow.
And I'm not the only one saying this but - the bit about LLMs is likely throwing the baby out with the bathwater. Yes the "AI-ification" of everything is horrible and people are shoehorning it into places where it's not useful. But to say that every single LLM interaction is wrong/not useful is just not true (though it might be true if you limit yourself to only freely available models!). Using LLMs effectively is a skill in itself, and not one to be underestimated. Just because you failed to get it to do something it's not well-suited to doesn't mean it can't do anything at all.
Though the conclusion (do things, make things) I do agree with anyway.
You're a platform drone, you have no mind, yada. Yet, we are reading the author's blog.
The author may hate LLMs, but they will lead to many people realizing things they never were aware of, like the author's superficial ability to take information and present it in a way that engages others. Soon that will be a thing that is known. Not many will make money sharing information in prose.
What the author refers to as "LLMs" today, will continually improve and "get better" at everything the author has issues with, maybe in novel ways we can't think of at the moment.
Alternative take:
"Popular culture" has always been a "lesser" ideal of experience, and now that ontological grouping now includes the Internet, as a whole. There are no safe corners, everything you experience on the Internet, if someone shared it with you, is now "Popular culture".
Everyone knows what you know, and you are no longer special or have special things to share, because awareness is ubiquitous.
This is good for society in many ways.
For example, with information asymmetry, where assholes made others their food, it will become less common that people are food.
Things like ad-driven social networks will fade away as this realization becomes normalized.
Unfortunately, we are at the very early stages of this, and it takes a very long time for people to become aware of things like hoaxes.
I have not yet figured out why anyone would choose this behaviour in a text editor. You have to press something to exit the delimited region anyway, whether that be an arrow key or the closing delimiter, so just… why did the first person even invent the idea, which just complicates things and also makes it harder to model the editor’s behaviour mentally? Were they a hunt-and-peck typist or something?
In theory, it helps keep your source valid syntax more of the time, which may help with syntax highlighting (especially of strings) and LSP/similar tooling. But it’s only more of the time: your source will still be invalid frequently, including when it gets things wrong and you have to relocate a delimiter. In practice, I don’t think it’s useful on that ground.