슬롭이 나를 공포에 떨게 한다
Slop Terrifies Me
요약
작성자는 AI 개발이 정체되어 진정한 혁신을 저해하고 품질 저하로 이어지는 "그저 그런" 소프트웨어를 초래할 수 있다는 두려움을 표현합니다. 그들은 사용자들이 불완전한 AI 생성 제품에 안주하게 될까 봐, 그리고 진정으로 훌륭한 소프트웨어를 만드는 데 대한 동기가 줄어들어 쉽게 생산되는 평범한 애플리케이션이 지배하는 미래로 이어질까 봐 걱정합니다. 이러한 소프트웨어의 "테무화"는 단순한 상품화보다 더 해롭다고 간주됩니다.
댓글 (213)
Sounds like the cost of everything goes down. Instead of subscription apps, we have free Fdroid apps. Instead of only the 0.1% commissioning art, all of humanity gets to commission art.
And when we do pay for things, instead of an app doing 1 feature well, we have apps do 10 features well with integration. (I am living this, instead of shipping software with 1 core feature, I can do 1 core feature and 6 different options for free, no change order needed)
A couple of years ago, I worked for an agency as a dev. I had a chat with one of the sales people, and he said clients asked him why custom apps were so expensive, when the hardware had gotten relatively cheap. He had a much harder time selling mobile apps.
Possibly, this will bring a new era of decent macOS desktop and mobile apps, not another web app that I have to run in my browser and have no control over.
There has been no shortage of mobile apps, Apple frequently boasts that there are over 2 million of them in the App Store.
I have little doubt there will be more, whether any of the extra will be decent remains to be seen.
They allow me to do work I could never have done before.
But there’s no chance at all of an LLM one shotting anything that I aim to build.
Every single step in the process is an intensely human grind trying to understand the LLM and coax it to make the thing I have in mind.
The people who are panicking aren’t using this stuff in depth. If they were, then they would have no anxiety at all.
If only the LLM was smart enough to write the software. I wish it could. It can’t, nor even close.
As for web browsers built in a few hours. No. No LLM is coming anywhere new at building a web browser in a few hours. Unless your talking about some super simple super minimal toy with some of the surface appearance of a web browser.
I just enjoy writing my own software. If I have a tool that will help me to lubricate the tight bits, I’ll use it.
This is the browser engine I was alluding to in the post: https://github.com/wilsonzlin/fastrender
Based on the Adobe stock price the market thinks AI slop software will be good enough for about 20% of Adobe users (or Adobe will need to make its software 20% cheaper, or most likely somewhere between).
Interestingly workday, which is possibly slightly simpler software more easily replicable using coding agents is about the same (down 26%).
Agents don’t care about any of Workday’s value-adds: Customizable workflows, “intuitive” experiences, a decent mobile app. Agents are happy to write SQL against a few boring databases.
A society where a large percent have no income is unsustainable in the short term, and ultimately liable to turn to violence. I can see it ending badly. Trouble who in power is willing to stop it?
Absolutely no one.
https://www.penguinrandomhouse.ca/books/719111/survival-of-t...
Brink? This has been the reality for decades now.
>A society where a large percent have no income is unsustainable in the short term, and ultimately liable to turn to violence. I can see it ending badly. Trouble who in power is willing to stop it?
Nobody. They will try to channel it.
I think all signals are pretty inevitably pointing to three potential outcomes (in order of likelihood): WW3, soviet style collapse of the west or a soviet style collapse of the sino-russian bloc.
If the promise of AI is real I think it makes WW3 a much more likely outcome - a "freed up" disaffected workforce pining for meaning and a revolutionized AI-drone first battlefield both tip the scales in favor of world war.
People always say this with zero evidence. What are some real examples of real people losing their job today because of LLMs. Apart from copywriters (i.e. the original human slop creators) having to rebrand as copyeditors because the first draft of their work now comes from a language model.
Sure some things deteriorate, but many things improve. Talking about a net decline (or net gain) is very difficult.
Every age has its own set of problems that need to be solved.
Not to the degree you might originally think. Most of the wealth being captured today is hypothetical wealth (i.e. promises) to be delivered in a hypothetical future. Except we know that future will never come as the masses, as you point out, have almost nothing, and increasing nothing, to offer to make good on those promises. In other words, it is just a piece of paper with IOU written on it, not real wealth.
What that hypothetical wealth does provide and what makes it so appealing, however, is social standing. People are willing to listen to the people who have the most hypothetical wealth. You soon hear of what they have to say. When the hobo on the street corner says something... Wait, there is a hobo on the street corner?
A small group of people having the ear of the people is human nature. In ancient times, communication challenges left that small group of people to be limited to a small community (e.g. a tribe, with the people listening to the tribe leader). Now that we can communicate across the world with ease, a few people rising up to capture the attention of the world is the natural outcome. That was, after all, the whole point — to move us away from "small tribes" towards a "global tribe".
Hypothetical wealth is the attention-grabbing attribute du jour, but if you remove it, it will just become something else like who is most physically attractive, who tells the funniest jokes, whatever. The handling of "Dunbar's number" doesn't go away.
> Trouble who in power is willing to stop it?
China has tried with its Great Wall (meaning the internet one, although perhaps you can find relevance in the physical one too), but is it successful? Maybe to some degree, but I expect many people in China still listen to what Elon Musk has to say, all while completely ignoring the millions of Chinese people immediately outside of their door. It isn't really something a power can do (ignoring that there even being a power contradicts the whole thing). The people themselves could in theory, but they would have to overcome their natural urges to do so.
I feel like long before LLMs, people already didn't care about this.
If anything software quality has been decreasing significantly, even at the "highest level" (see Windows, macOS, etc). Are LLMs going to make it worse? I'm skeptical, because they might actually accelerate shipping bug fixes that (pre-LLMs) would have required more time and management buy-in, only to be met with "yeah don’t bother, look at the usage stats, nobody cares".
When software operators tolerate bugs they’re signaling that they’re willing to forego the fix in exchange for other parts of the feature that work and that they need.
The idea that consumers will somehow not need the features that they rely on anymore is completely wrong.
That leaves the tolerable bugs, but those were always part of the negotiation: Coding agents doesn’t change that one bit. Perhaps all it does it allow more competitors to peel away those minority groups of users who are blocked by certain unaddressed bugs. Or maybe it gets those bugs fixed because it’s cheaper to do so.
Its being compared to that of a slop machine, and billionaires claiming that its better than you are in all ways.
Its having integrity in your work, but the LLM slop-machines can lie and go "You're actually right (tells more lies)".
It all comes down to that LLMs serve to 'fix' the trillion dollar problem: peoples wages. Especially those engineers, developers, medical, and more.
"Terrified" is a strong word for the death of any craft. And as long as there are thousands that love the craft, then it will not have died.
As lots of people seem to always prefer the cheaper option, we now have single-use plastic ultra-fast fashion, plastic stuff that'll break in the short term, brittle plywood furniture, cheap ultra-processed food, etc.
Classic software development always felt like a tailor-made job to me and of course it's slow and expensive but if it's done by professionals it can give excellent results. Now if you can get crappy but cheap and good enough results of course it'll be the preferred option for mass production.
If only it was plywood, at least it'd be solid and sturdy. These days it's particleboard, which is much worse than plywood. Similar concept, but now made out of sawdust and glue instead of woodchips and glue that are alternately laid down in different orientations layer by layer for increased strength.
Particleboard chips much easier, breaks down much faster with moisture, and can't hold screws in. But it's very cheap, can be made very smooth, and is light.
Agree with the general sentiment though.
https://arxiv.org/abs/2510.15061
Our definition of slop (repetitive characteristic language from LLMs) is the original one as articulated by the LLM creative writing community circa 2022-2023. Folks trying to redefine it today to mean "lazy LLM outputs I don't like" should have chosen a different word.
A computer is a person employed to do arithmetic.
The definitions you're operating under are mentioned thus:
> characteristic repetitive phraseology, termed “slop,” which degrades output quality and makes AI-generated text immediately recognizable. (abstract)
> ... some patterns occur over 1000× more frequently in LLM text than in human writing, leading to the perception of repetition and over-use – i.e. "slop". (introduction)
And that's ... it, I think. No further effort is visible towards a definition of the term, nor do the background citations propose one that I could see (I'll admit to skimming them, though I did read most of your paper--if I missed something, let me know).
That might be suitable as an operating definition of "slop" to explain the techniques in your paper, but neither your paper nor any of your citations defend it as the common definition of an established term. Your paper's not making an incorrect claim per se, rather, it's taking your definition of "slop" for granted without evidence.
That doesn't bode well for the rigor of the rest of the paper.
Like, look: I get that this is an extremely fraught and important/popular area of research, and that your approach has "antislop" in the name. That's all great; I hope your approach is beneficial--truly. But you aren't claiming a definition of slop in your paper; you're just assuming one. Then you're coming here and asserting a definition citing "the LLM creative writing community circa 2022-2023" and asserting redefinition-after-the-fact, both of which are extraordinary claims that require evidence.
Again, not only do I think that mis-definition is untrue, I also think that you're not actually defining "slop" (the irony of my emphasizing that in a not-just-x-but-y sentence is not lost on me).
I don't know which of the authors you are, but Ravid, at least, should know better: this is not how you establish terminology in academic writing, nor how you defend it.
It's the societal level impact of recent advances that I'd call "terrifying". There is a non-zero chance we end up with a "useless" class that can't compete against AI & machines - like at all, on any metric. And there doesn't seem to be much of a game plan for dealing with that without social fabric tearing
It's just that many powerful people have a vested interest in keeping the rest of us poor, miserable, and desperate, and so do everything they can to fight the idea that anything can ever be done to improve the lot of the poor without destroying the economy. Despite ample empirical evidence to the contrary.
This isn’t news really. Content farms already existed. Amusing Ourselves to Death was written in 1985. Critiques of the culture exist way before that. But the reality of seeing the end game of such a culture laid bare in the waste of the data center buildout is shocking and repulsive.
Quality. Matters.
It always has, and it always will. If you're telling yourself otherwise, you are part of a doomed way of thinking and will eventually be outcompeted by those who understand the implications of thinking further ahead. [ETA: Unfortunately, 'eventually' in this context could be an impossibly long time, or never, because people are irrational animals who too often prioritize our current feelings over everything else.]
80% of good maybe reframed as 100% ok for 80% of the people. It is when you are in the minority that cares about or needs that last 20% where it is a problem because the 80% were subsidizing your needs by buying more than the need.
This just isn't true. First, cheap tools have always been around. I have a few that I've inherited from my grandfather and great-grandfather. They're junk and I keep them specifically to remind myself that consumer-oriented trash versions of better quality tools have always existed.
Second, Harbor Freight is the only consumer-oriented tool retailer that seems to be consistently improving their product lines. Craftsman, which was the benchmark for quality, consumer-oriented hand tools, dropped off a cliff in terms of quality around the mid- to late-2000s.
If you can afford professional-grade tools (Snap-On, Mac, Wera, Knipex, etc.) great. For the rest of us, Harbor Freight is the only retailer looking out for us. Their American- and Taiwanese-made tools are excellent. Their Chinese-made tools are good. Their Indian-made tools will get the job done, but it won't be pleasant. At least they give the consumer a range of options, unlike Snap-On, which gives you a payment plan.
The job is going to be much less fun, yes, but I won't have to learn from scratch and compete with young people in a different area (and which I will enjoy less, most likely). So, if anything slop gives me hope.
If you substitute "artificial intelligence" with offshored labor ("actually indo-asians" meme moniker) you have some parallels: cheap spaghetti code that "mostly works", just written by farms of humans instead of farms of GPUs. The result is largely the same. The primary difference is that we've now subsidized (through massive, unsustainable private investment) the cost of "offshoring" to basically zero. Obviously that has its own set of problems, but the piper will need to be paid eventually...
Soon, we'll all just be meatpuppets, guided by AI to suit AI.
In my experience coding agents are actually better at doing the final polish and plugging in gaps that a developer under time pressure to ship would skip.
My favorite pre-LLM thing in this area is Flighty. It's a flight tracking app that takes available data and presents it in the best possible wway. Another one is that EU border visa residency app that came thru here a couple of months ago.
Standards for interchange formats have now become paramount.
API access is another place where things hinge on.
Only big creative companies like Disney can play the game of making licensing agreements. And they are ok with it because it gives them an edge over smaller, less organized creators without a legal department.
https://thewaltdisneycompany.com/news/disney-openai-sora-agr...
AI produces code that technically runs but lacks the thoughtfulness that makes software maintainable or elegant. The "90% solution" ships because economic pressure rewards speed over quality.
What haunts me: compilers don't make design decisions. IDEs don't choose architecture. AI does both, and most users accept those choices uncritically. We're already seeing juniors who've never debugged without a copilot.
The author's real question: what if most people genuinely don't care about the last 10%? Not from laziness, but because "good enough" is cheaper and we're all exhausted.
Dismissing this as "just another moral panic" feels too easy. The handcraft isn't dying because AI is too good. It's dying because mediocrity is profitable.
The "AI effect" on the world has many similarities to previous events and in many ways changes very little about how the world works.
> I'm terrified of the good enough to ship—and I'm terrified of nobody else caring.
For almost every product/service ever offered, it was possible to scale the "quality" of the offering while largely keeping the function or outcome static. In fact, lots of capitalistic activity is basically a search for the cheapest and fastest way to accomplish a minimum set of requirements. This leads to folks (including me!!) to lament the quality of certain products/services.
For example, it's possible to make hiking boots that last a lot longer than others. But if the requirement is to have it last for just 20 miles, it's better to pay less for one that won't last as long.
Software is the same way. Most users just absolutely do not know about, care about, or worry about security, privacy, maintainability, robustness, or a host of other things. For some reason this is continually terrifying and shocking to many.
There is nothing surprising here, it's been this way for many years and will continue.
Obviously there are exceptions, but for the most part it's best to assume the above.
Put another way, who here wants a car that costs more than their house? Or shoes that cost 2000$?
Sure, but the OP's concern is whether this chokes off innovation. Is there some better kind of hiking boot, longer-lasting and cheaper and maybe more comfortable, that we've never found because the shoemakers who'd be able to invent it are too busy optimizing Nike production lines?
nitpick: most users don’t care about these things until something goes significantly wrong and it impacts them, e.g. a massive data breach or persistent global downtime.
then they get angry. very angry.
just because people don’t care about it now doesn’t mean they won’t care about it in the future.
edit — these are the hidden requirements.
> For example, it's possible to make hiking boots that last a lot longer than others. But if the requirement is to have it last for just 20 miles, it's better to pay less for one that won't last as long.
until requirements change, or the hidden requirements come out to play … most software engineers can probably recall multiple times when the requirements changed half way through. hell, i’ve done it on solo projects.
now we’re stuck with boots that can only last 20 miles, but we need to go 35.
It's often lamented that the World Wide Web used to be controlled by indie makers, but now belongs to a handful of megacorp websites and ad networks pushing addictive content. But, the indie maker era was just a temporary market inefficiency, from before businesses fully knew how to harness the technology.
I think software development has gone through a similar change. At one point software companies cared about software quality, but this too was just an idealist, engineer-driven market inefficiency. Eventually business leaders realized they can make just as much money (but make it faster) by shoveling out rushed, bloated, garbage software, since even though poor-quality software aggravates people, it doesn't aggravate enough for the average person to switch vendors over it. (Case in point - I'm regularly astounded at how buggy the YouTube app is on Android of all platforms. I have to force-kill it semi-regularly to get it working right. But am I gonna stop watching YouTube because of this? Admittedly, no, probably not.)
Seems an unlikely problem. It'll get better, which may cause it's own problems.
I get the emotion, but factually, the author doesn't need to worry so much.
We've always had cheaper, inexperienced developers shipping code that is the bare minimum to get the product out the door. All of capitalism has been about delivering the minimum quality necessary make money.
Fortunately, economic competition ensures that businesses are constantly trying to deliver a better product than their competition. Which in some contexts means the same thing for cheaper, but in others means better quality at a given higher price.
As long as there is economic demand for higher-quality software, it will continue to be produced. This is structural. It doesn't matter how much starts as LLM output because it's still architected, code gets reviewed, tested, bugs get filed. Because economic competition requires it to be or else you'll go out of business.
And if there's new demand for lower-quality software, for lots of one-off scripts that people can get an LLM to write, or GUI frontends for command-line tools, or whatever it is, then that's amazing too! Even if they're "slop", they're useful and accessible and correctly treated as temporary or disposable. There's nothing wrong with that, and it doesn't mean those tools are going to affect or replace serious software like Photoshop or macOS or whatever.