Hacker News Reader

Comment by aaronbrethorst | Hacker News Reader

I found the AI writing of this post to really detract from its message. Give your agent meaningful writing samples of your own work and use those as a ‘style transfer’ basis for blog posts to get something far more true to your own voice.

View on HN

Or please just put the prompt in a blog post instead.

Who would read a blog post that just says "Think of a topic that users of an app for monitoring blood oxygen might be interested in, do a web search for related articles and synthesize them into a blog post. Make sure to draw on your own personal experience to make it more engaging"? I'm sure the actual ad at the bottom had more human effort put into it than this article.

The website owner didn't even bother to check for hallucinated links, though https://journals.physiology.org/doi/full/10.1152/japplphysio... does exist and somewhat backs up the clickbait headline, so it would be satisfying comeuppance if the mods could just replace the submission accordingly.

It's well-structured and the message is clear. Are we intentionally prompting LLM to write badly now? Do we have to manually write bad essays to avoid AI accusation?

It's just so cliche. The dramatic transitions which introduce things that aren't as important as the transition itself. The flow is very AI, short dramatic responses to a previous question that's also not "groundbreaking" enough to warrant such a style. It's just hard to unsee these things. Idk man, I guess if you like it, that's great but I cringe when I read this and I never finish reading because I assume the author put in minimal effort so why should I?

I mean... this is an ad for their app and the purpose is to make you click more articles on their blog. Even if it's written by a human it doesn't justify much more time reading it. If there are jobs deserving to be replaced by LLM, this kind of copywriting is on top of my list.

There's confidently hallucinated citations, which makes it bad writing either way.

Why is this AI writing accusation necessary? Plenty of humans write this way. Have you ever read pre-AI content marketing articles? If you've learned a bunch content marketing advise then you'll see those patterns that you now associate with "AI writing" were already all over the place. Baity titles like "Why it's bad that X did Y" or "<explanation of the problem>. Want to be freed from worrying about this? Use $OUR_BRAND", urgh, once you learn those patterns you can't unsee it.

Granted, you don't like to like this style of writing, I don't either. But you don't have to auto-accuse AI writing either. Also, there's nothing wrong with using AI to rephrase a manually written text for better readability, plenty of people use AI for that too rather than writing the entire thing.

Usually, I can easily tell bad AI slop, because it is just that - sloppy - the bullet points, the 'delving' and all that. But how can you tell this article was also AI-tainted? On a second skim, I can sort of sense some of it - the bulletpoint-enthusiasm, the idiosyncratic segues (?) that link sections/paragraphs of the text. But it didn't trigger for me immediately, or cause me concern..?

I'm worrying that soon, I will have to hunt for non-AI essays by them just being worse written/more 'crude' and not as eloquently written as an AI would do :-/ Basically, seeking out "authentic human slop".

Here's the part that really stood out for me. Not one thing but the other thing that isn't really noteworthy.

(Also have a look at Wikipedia how to identify signs of ai writing.)

Passages like this one suggest that maybe it was an AI rewrite, rather than from scratch:

> I experienced this pattern without understanding it. My Tuesday evening interval sessions, scheduled after long workdays, consistently felt worse than my Saturday morning sessions. I blamed sleep, stress, hydration. Those all matter, but the research suggests the cognitive load itself was a primary culprit.

You really couldn't tell? The overly dramatic transitions all over the place is such an obvious tell:

> Here's the part that surprised me:

Might as well have said "here's the kicker" and used emojis instead of bullets. Maybe you can share your reading sites as you seem rather undrrexposed to not recognize this immediately lol.

Edit: I mean come on man, how can you not tell?! I'm still cringing from this one:

> The incremental cost of actually thinking hard? Almost nothing.

Edit II

"This isn't one study"

Dum dum dum. Sooo dramatic. 100% slop.

Hmmm... I write like this.

Maybe AI is being trained on my writings.

Edit: Maybe because I was raised in the 80s, but this style of "asking a question to introduce a topic" was very common back then.

I do that too. It feels more natural, like I'm telling you what I am going to try to answer.

Do you really write like this in every single paragraph?

> The incremental cost of actually thinking hard? Almost nothing.

Wait you can tell from this that it's written by a LLM? I think you're written by a LLM...

The sentences are all roughly the same length too

It's very clear to me on its face that it's AI, but not "obvious as the sky is blue" others seem to be implying. I would dislike the writing style even if it weren't AI.

For the record, an AI detector that appears to have put work into reliability and that I trust very much from my own testing, Pangram (https://www.pangram.com), says this is 100% AI generated. I've used it plenty before when experimenting with AI-collab writing, both fiction and non-fiction, and it's frustratingly accurate in identifying what is and isn't my contribution. I have since largely given up trying to do AI-collab writing, because no matter how nice the writing looks in the moment, it always reeks when read closely, or on later days.

Your detector did not work well on an AI-collab writing fiction project I did a while ago, tagged it as 100% human even with high confidence for the most part. But to be fair, most detectors weren't significantly better, although this one gave a justification that made sense https://aidetector.com/

Or better, style-transfer the training data, and retrain the model.

I wanted to test my theory that "don't use cliche'd language" helps with that, but incredibly the essays ChatGPT is giving me today don't have any of the tells. How do I get it to give me slop?

I asked "Can you give me a short essay on the history of fire." Maybe the type of writing requested has a massive effect on the language used?

Possibly. As for this blog post and for most slop the prompt itself will contain the material. The material will be very short.

"Here is one paragraph of an idea, an abstract of a report. Write a blog post."

Whereas a "history of fire" contains no article in the prompt and so might have more material from its own auto complete database to draw from.

Or just fucking write it yourself!

Reply on news.ycombinator.com