AI agents produce slop when not restricted. Here are 10 things to watch out for

381 points•theanonymousone•3 days ago•241 comments•

241 comments

weakfish3 days ago
I’ve been trying to put my finger on what it is that is happening when the robot writes stuff like “3 campuses, one app” example, I’m glad the author was able to identify it as chat context leaking through.

The other version of it is the robot over-indexing on some part of the prompt and leaving comments places. An example is I ask it to prefer integration tests using TestContainers, it starts adding comments to every new test saying “Real services, no mocks” or something.

And yes, I have a line in my *.MD saying not to do that.

JoshTriplett3 days ago
LLMs love https://tvtropes.org/pmwiki/pmwiki.php/Main/SuspiciouslySpec... . You tell them not to do a thing, they make a point of saying they didn't do the thing.
WorldMaker3 days ago
It's partly a "Don't think of a Pink Elephant" problem. I've worked on several projects where I had to keep telling prompt writers to stop writing negative examples because the more you include the more its "attention" to them is all it has. Like telling a toddler not to do something and being surprised that is now all they can think about and they want to keep doing it. These prompt writers kept getting surprised that I'd delete all their negative examples and harshly worded "Don't do X" and "Never Y" and "NO: Z" sections they spend so much time on and got better results with smaller more focused positive example only prompts.
orwin3 days ago
Huge issue in slop comments.

And btw, if you have slop comments in your PR, you'll get a review from an agent (my company pay for it). I won't bother to read if you didn't bother either.

unholiness3 days ago
I think lots of AI tells these days are leaks from the ai/writer connection into text for the reader. It partly feels like the inevitable result of RLHF with the wrong human's feedback ("I did it! Pick me!") and partly feels like thinking tokens leaking into the main text.

"Here's the argument, in plain terms", "It's not just X, it's Y", "It's worth stating precisely", "The sharper distinction here"... They're things that gesture toward the relationship between the text and the prompt, making it clear where it matched the author's expectations and where it deviated.

Not that humans are immune from coding/writing for their boss instead of their user/reader!

torginus3 days ago
I just used Opus 5.5 for something other than coding, and after about a dozen chat messages (some of which were genuinely amazing), it got LLM brainrot, and started suggesting dumb, irrelevant stuff, ignoring previously established facts, and talking in PR doublespeak.
demibabs3 days ago
Another example (not from this app, just in general) that I don’t see people talking about is: “Your files never leave your browser/device”.

Any vibe coded site that deals with files and works client-side feels an INCESSANT need to inform you of this fact.

Definitely seems like the same kind of context leakage. “I don’t want to spin up a server, can you make it work in the browser?”

czhu123 days ago
At least some of these are just for lack of caring. I vibe coded this landing page recently https://tui2web.com/ and fought off a lot of the defaults the LLM came up with after a few hours of iteration. Feel free to roast it, I'm not really attached, but I thought it did a reasonably good job.
quirino3 days ago
I find this style of "heavy dark borders/shadows" one of the ugliest recent trends on the web. It's rarely done well.

But otherwise the website looks pretty great. Vibe coded websites usually transmit the feeling of: "nobody even bothered to look at this", and that's not the case with yours at all.

In your lists where elements are separated by horizontal lines, I would remove the horizontal line after the last element. That's quite easy in CSS.

czhu123 days ago
You could be right, I haven't seen too many like this yet, but it did some out with this staggeringly fast. The style I've seen the most are the gradients + font changes mentioned in the article.
voxelghost2 days ago
What's interesting to me, is how it selected this heavy dark shadows brutalist style, which is a sort of stylized skeuomorphism , and then it decided to add a background different from page bg, breaking whatever illusion there was.
sinker3 days ago
It's not too readable. It would benefit from a columnated layout. The fold is also too long. It should be an introductory block.

Overall, the "rhythm" is a bit off. The user should know what they're seeing even before they see it, at least conceptually. The design/layout doesn't section off dense sections from skim material. Front pages in general should be built for skimming rather than deep consumption (scanning). Scanning should appear optional visually.

There's a lot of "scan" material, but users who visit your site mostly just want to know what they're looking at - at least at first. So the deep material kinda just gets in the way of actual comprehension.

Another critique: too much visual uniformity. Every major section looks the same (poor "rhythm"). This leads the eyes to glaze over much of the site.

It sounds dumb, but our eyes don't "read" a webpage, we "see/feel" it instead. We see structure, layout, differences in colors, font sizes, patterns, bold headings, etc. We ignore the actual textual content. We only do actual reading when we're hunting for something specific (eg, a specific nav link).

The aesthetic itself is fine, but the layout and structure need work.

WorldMaker3 days ago
All caps text is a roadblock for reading speed/comprehension. Accessibility experts were finally convincing major company design standards to stop doing it when LLMs made it a default again [0] for headers and incidental text.

Not to mention that all caps text is also generally associated with shouting and just feels rude to some readers.

[0] Or more accurately trained on all the bad examples before experts started catching them and has stuck to the mistake.

Anon10963 days ago
This appears to be pretty close to the "brutalist" aesthetic mentioned in the article. Things like drop shadows on boxes are the opposite of the "fingernail" boxes and you very rarely saw them pre-AI. However I think your site looks fine and well crafted, nice work. (the article also said that this kind of design language is less grating)
jdlshore3 days ago
Drop shadows on boxes were a common TUI element in the Norton Commander days.
thiht3 days ago
It's been a trend for years before AI... I remember the Balenciaga website being one of the first big names doing it in like... 2018?
czhu123 days ago
Yeah brutalist is exactly right, that is basically what I prompted for. I didn't realize a bunch of other people were doing this. Actually the only inspiration I had for this was watching the movie (Brutalist) a few years back.
demibabs3 days ago
I don’t know man, still looks exactly like a million vibe coded sites.

Good aesthetic, just overused.

vallerie3 days ago
Love this blog mentioning the issue of implementation detail in prose. Likely my biggest irk of vibe coded apps.

Much of this is it's a simple problem to solve - just prompt agent of choice "Put all user visible text into a translation file for i10n" and then go through and edit said prose as a human.

nullbio3 days ago
OAI's models are notorious for it and always have been.
zinoc3 days ago
This is similar to the rapid early adoption of the Bootstrap CSS framework. A lot of people were annoyed by the sudden spike of websites that looked and felt the same, but I think this was a tremendous step forward in usability and accessibility, and well worth the initial wave of cookie cutter websites.
voidhorse3 days ago
The difference here is that the slop paradigm is not a set of intentionally designed UX choices that work well for users, it's a set of accidental commonplace LLM tendencies, not designed for humans, replicated ad infinitum that are making user experiences worse in 99% of cases. The use of white space is horrendous, the font choices are illegible, the color palettes lack sufficient contrast.
tonic_note3 days ago
Yeah it's problem of sensibility. Like people do not realize that developing a sensibility and good taste here is an actual still that takes time, and is highly undervalued in tech circles with the exception of a handle of truly design driven companies.
tlahtinen3 days ago
Earlier this year I used Claude to design a screen for an Unreal game I work on. At first I thought wow, what a great layout, this is a game changer and gives me a huge advantage.

Less than a year later I am embarrassed how Claude it looks - blinking dot in the corner, double // separating header text, the font choices... now I see the same patterns everywhere, and I'm sure the audience is picking up on it too.

I still use Claude in my work, but a lot more sparingly; I ask for color scheme and font suggestions (no Rajdhani or Share Tech Mono!), but they're mostly for inspiration and I'm not blindly copying them. It feels like I'm being smarter about it, but maybe in a year I'll realize I still haven't climbed out of the trap :)

Read the full thread on Hacker News →

Related stories