[jchw]

OK, well, I'm getting very tired of all of the terrible AI-generated articles.

I made a longer reply that discusses why I think this article is bad on top of the fact that the writing is absolutely fucking horrid, but a single person could theoretically pump out hundreds of these per day if they wanted, so having a nuanced critique for each of these is going to be pretty hard.

There's only one solution, and that is flagging and removing all AI generated articles. Fullstop.

[OutOfHere]

Just because some people choose to write poorly using an LLM doesn't in itself make all AI generated content bad. In can in fact be far better than an article written by a human if instructed reasonably.

[jchw]

Yeah, no, if I can tell it's AI generated, that means it has the same garbage writing style that for some reason LLMs can't help but pump out. And if some human person thinks that's fine to post, they either don't care/didn't read their own post, or they have terrible taste.

And again, it's a practical issue anyways. You can have Claude generate hundreds of these. I've already personally seen multiple blogs where there are multiple fully written long articles being posted per day. These even occasionally make it to Hacker News. Did the person who generated these actually read them? Probably not.

Most importantly, it is unacceptable to pass off AI generated prose or images as if it is human expression. It's one thing with code where the primary point of it is to be executed, but I have zero interest in people who can't formulate their own thoughts into writing. I don't see how it is any better to submit AI generated articles to Hacker News than it is to respond to people with AI generated comments.

Humans aren't infallible, but the point of content isn't the content, it is the ideas, and the ideas are valuable because of the work put into them. AI slop articles are a serious problem because they superficially look like something where a lot of effort is put in, as the models will happily make bold claims and justify them into the ground no matter how untrue or unjustified those claims are. There is a feasible future where AI generated content is also valuable because of the effort put into them. It certainly happens on occasion today. It just isn't what we're seeing here right now on Hacker News. And because many people here (certainly myself included from time to time) often skim the articles or sometimes don't even actually read them, it is important that the community put some effort into weeding out slop content, AI or human, that fails to justify its claims. The rising tide of AI generated crap is making this task harder and more annoying.

This isn't something I'm ever going to relent on, either. So I will have to leave it up to HN to decide if they would rather ban us all for complaining or come to reasonable senses and agree that there is no sustainable way to allow blatantly AI-generated content onto the front page. I don't view this as an ultimatum so much as just an observation extrapolating off of what we're seeing today: nobody has really made me feel there is any compelling point to allowing AI generated crap on here. If anything the further it gets the less supported arguments in favor of it are seeming justified.

The explosion in quality from better models is always around the bend.

[OutOfHere]

Like I said, the model quality is not a problem. They can trivially write professionally. Even a two year old model can. If they write poorly, it's because they are asked to appeal to the lowest common denominator of clickbait.

[jchw]

Actually I disagree, it's pretty apparent that modern frontier LLMs are nearly completely incapable of writing good prose for some reason. I'm not sure if it's the RLHF phase or what, but even when you explicitly tell them to try to avoid the cliches it's never enough. They're geared toward writing heavily punchy, low-substance prose, and it shows up everywhere in their output, even in places like documentation and just normal chat replies.

The exact way in which the models are fucked up seems to depend on circumstances, but I think right now one thing I've noticed out of the latest versions of Claude Opus is that it really really likes to use the word "honest" in its summaries. "What Remains (The "Honest" Part)". I figured this was maybe something to do with it just repeating the system prompts but no. It turns out the word "honest" does appear in some fragments of system prompts in Cluade Code, but it doesn't appear to be anywhere where it would've been in the context of my recent runs.

I think this is a tuning issue and that eventually, someone will figure out a good way to prevent models from getting skewed this way.

Still, the bad prose quality is not really the biggest issue. In fact, it's kind of handy that the prose quality is shit because it makes it easier to tell when someone just doesn't seem to care about what they're writing. If the prose quality was really good, yet the amount of effort put in was the same, we would be having an even worse problem right now.

[OutOfHere]

I ask GPT 5.5 to write professionally or in a moderately formal tone and it does. If you are producing garbage output, there are only three explanations: (1) you don't have a system prompt to obtain targeted output, the kind you need (2) you have a badly written system prompt (3) you have a weird bad model.

The world is bigger than Claude and an oversimplified worldview.

If you haven't even read good quality AI written articles, you have either been wilfully blind to them or you have a bigger problem, because you surely haven't avoided them.