Hacker News new | past | comments | ask | show | jobs | submit login

This ticks 9/10 boxes on my detector for typical LLM generated SEO content spam. :\



OP here.

Honestly, I got similar feedback when I got this reviewed internally. At this point I am not sure how to write so that it doesn't seem LLM generated.

Would be helpful if you could share why you thought this was LLM generated. The suggestions I have gotten so far has been to remove bullet points and sections - which I feel breaks readability.


I don't think it's so bad, but if I had to guess, it's from the division / breakdown of sections and lists, which reads a lot like the formulaic approach you get from an LLM (which is not necessarily bad, just common in the output). E.g. "Docker and Docker Compose can simplify the process of installing and managing services. They allow you to:" etc etc. This may sound like an LLM covering all its bases rather than a human explaining subject matter.

That's just my take, again I don't think it's that bad. The article would be a useful breakdown for beginners.

(Also, I'm sure you know, LLM content sounds that way because the LLM was trained on content just like this, so it's not really surprising that a guide generated by an LLM would sound like the kind of guide that was used to train an LLM...)


Not parent commenter, but I've been trying to verbalize why it feels LLM-like.

- h2 titles feel as basic as possible, just "what self-hosting, who self-hosting, why self-hosting, ..."

- SEO spam often overuses keywords; on this page, it feels like "self-hosting" is used a bit too often, even if it's well-intentioned

- the text ends in a classic LLM warning "remember to be careful"

- predictable sentence patterns

Some of these things are good for readability. I guess this article feels a bit too plain? I think tech company blog posts add a unique style and voice these days, because otherwise they'll blend in with the average SEO/LLM content.

Also editing nits:

    > self hosing
    > Self Hosting 
    > atleast
Good self-hosting tips, though. Thanks for sharing.


Thanks. This is really helpful.

The overuse of "Self Hosting" is fair. Better H2 titles would have made it less frequent. Will be more thoughtful about this the next time.

The unique style and voice is where I am struggling with. Have always been instructed to write in a plain tone and simple English so that its easier to read through.


I tried reading the article with the GP's comment in mind. For most of the sections it didn't feel like there was anything that would flag it as LLM generated for me.

But when I got to "How to Start Self-Hosting?", which is the section I was most interested in, I got a strong sense of déjà vu.

Reading this section felt exactly like I feel when I hit a bad prompt on ChatGPT. I feel I'm being given a huge dump of keywords but nothing that lets me make any progress. Reading it I felt the same frustration I do with ChatGPT as I have to prompt it again with "Can you elaborate on bullet point 6" to get anything useful out of it.

With ChatGPT the reason is usually a prompt that was either too broad/open-ended or a difficult topic for ChatGPT to answer. And it has a tight limit on how long the answer can be, which is understandable. For an article though it feels a bit jarring and there is no immediate way to ask for details.

I think the rest of the article is fine really. Sure the word of caution is exactly what LLMs do but unlike LLMs, which usually state the obvious, it has a lot of useful information.


Normally I'd not pay too much attention to these comments but the assessment here is spot on. I'd say LLMs articles in general are:

  1- Always longer than necessary with a lot of fluff

  2- Favor lists and hierarchy
Because they're trained on mostly SEO spam and buzzfeed-style articles

I asked Gemini to "write an article about self hosting" and the output structure and content is eerily similar

Here is a side by side comparison: https://i.postimg.cc/kXXpWgnZ/why-it-look-like-LLM-generated...




Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: