2026-05-18
There was a very hot proposal on lobste.rs this week. The proposal was to ban users who regularly post "LLM generated article(s)". It didn't surprise me too much because I feel like, almost every article comment section on that site has some comment to the effect of "this article smells like it was generated by an LLM".
Often when you push back on a comment like that, you get a link to the wikipedia article about signs of AI writing, which even has a quiz associated with it so that you can test your own AI-dar. Interestingly, most people who have taken the quiz and published their score, tend to get more than half right. Neat. Not sure really what that says, since its a sample size of like 20 or whatever, but there you have it.
Anyway, in this article, I'd like to dive into this topic, because I feel like I have a bit of a different take than a lot of people; one which was echoed in a few obscure comments on the giant lobste.rs thread (linked above), but seems to be missed a lot in the conversation.
One has to ask the question: what do you actually care about when you read an article? In other words, why do you even start reading an article on the internet to begin with? For me, the answer to this question usually lies in that I was intrigued to read an article either because it contains new information to me, and so I might learn something from it. Otherwise, I might read an article because it contains some philosophical argument that I am interested in reading because I want to compare it against my current conception of some part of the world.
An example of the first category might be that the article is about a programming language that I have never heard about, so I might be interested to learn about it. An example of the second category might be somebody arguing that functional programming is better than Object Oriented Programming (OOP) (just to use a super cliche example). In this case, I already know what those two concepts are, but I want to hear the authors argument, so that I can compare it with my opinion on the subject, and maybe by the end of the article, I'll have been convinced. The article you're reading right now, by the way, is also an example of the second category.
So the question becomes: in both of these categories, what are we losing of the article is generated by an LLM? Taking the first example, suppose I have a new programming language that I built and I want to share it with the world. Insert excuse for using an LLM here, so I get an LLM to read the git repo and write an article about. I then go through the article and check for inaccuracies, and so on. Suffice it to say that that article is mostly LLM generated. However, the article contains all of the necessary information for somebody to learn about this new language that they had never heard about before.
Now, take the second example. Bare in mind: LLMs suck at logic (trust me, as someone who has asked an LLM to edit a philosophy essay for me, and then found out that it actually subtly changed my argument, I know how bad these things are at reasoning). So in our example suppose I am writing an argument about functional vs OOP. In any case, the LLM is going to need a lot of help from me to get the logic right, but, suppose I work with an LLM to write this article and by the end we can say, over 50% of the text was generated by an LLM, with heavy oversight from me. I'd still call that an "LLM generated" article, but the important thing about the article isn't that an LLM helped generate it. It's that the logic was actually valid. The article contains a real argument that people (or machines?) can debate about.
So in both of these examples the LLM was a tool used to generate potentially many paragraphs of text but the important part wasn't the tool that was used -- whether it's an LLM or just me and my fingers. The important part is the actual content. The new information, or the logic contained within. This is what we should be judging an article on.
This is what gets under my skin when I see a comment (very common on lobste.rs) that is just "this seems AI generated". They're not making any argument against the article with that statement. If the article is poorly written, by an LLM or a human, it should be criticized on it's contents. Slop is a real thing, but it's clear within the first few minutes of reading a slop article, that the article isn't saying much. You would criticize a human as well for just writing a bunch of non-sense fluff (believe it or not that's something we used to do before we started blaming LLMs for bad writing), just as much as you should rightfully criticize an LLM generated piece of garbage.
Another point I want to make here before I conclude is that, yes, there are some obvious things that LLMs do that most native English writers don't, at least not anymore. But, that doesn't mean that no human wouldn't write like that. As we all know, LLMs are trained on human writing. They are literally outputting the thing at the very middle of the bell-curve given all of human English from the internet. Thus, there will be a non-zero set of writers that write that way. It's fair to just not like to read a certain writing style. I have many favorite and not favorite writers of my own. But to accuse someone of not writing like a human. Come on.
So, I guess all I'm saying here is, next time you go to comment section with the intention of complaining that it is slop. Many of us would appreciate if you gave a reason why the content of the article is unsatisfactory. I'm also saying that a blanket ban on "LLM generated" content is ridiculous because of the reasons listed above. That's my two cents on this. Thanks for reading. By the way, I didn't use an LLM to write this, but how would you even tell? I could have just instructed it to write like a me.