Hacker News Reader

Comment by broken-kebab | Hacker News Reader

Speleology aside, just a few days ago in one of HN discussions I was told LLM-produced content is already indistinguishable from human-made. But I'm sure I would have pretty good chances to correctly guess even the exact brand which helped to write this sloppy text. And it's the same with many stories I read.

View on HN

I know what you mean. Flowery and fluffy. I don't see any clear giveaways, but it feels like it and is a bit annoying to read because the information in there isn't very dense.

I saw a great video on YouTube by languagejones, and he asks: do we have to tell the difference between an LLM and an idiot?

The writing is just bad. I think this is clear to everyone, so I guess we could stop at that

the "why it matters" subtitle near the end seems very much like claude to me

That falls into the general category of 'authority groping' (yes, I coin my own terms) where a prediction engine without real reasoning, must apply numerous tricks to convey confidence, authority and substance where there may be less or none. Similar to the constant negative-propositions or "not" statement, where usually one to three things that do not apply are articulated, which has the result of credibility accumulation, I suppose, by showing "hey, I am right about not just one thing, but four!" hence the saturated styles we tend to see.

Stake the claim then. I’ve also read a fair bit of LLM text and this didn’t trip anything for me.

Edit: after reading the author bio and the other articles, I agree with this assessment. Though I will admit, this particular article didn’t jump out as AI to me.

It's Claude. I picked up on it too (I am also a heavy Claude user)

The title alone gives it away. Sentences like this also trigger my daily annoyance:

"Researchers found an elegant way to test it, and the timing was almost too convenient."

Me: "Almost too convenient is expressing an opinion. Rewrite this to be straight forward, succinct information"

Claude: "You're right, and I was wrong. That's the second time you've had to correct me this session. Let me update my memory to adjust for this"

Eye twitch

And the nonsensical, chatty slang

> Let me ...

As if Claude actually needed the human user's permission or cooperation to do so.

"summary, a, b, c, why it matters" + fillers

like the summarization tools based on those models except it's the opposite, it was fluffed up into a long article

Reply on news.ycombinator.com