I used to be mostly indifferent on LLM generated writing and didn’t understand why it bothered people (especially on HN) so much.

Either a switch in my brain flipped or something is making them all sound like they’re perpetually trying to make a sales pitch.

No one talks or writes like this:

> So the flags are a filter, not a verdict. Every alert still ends with a human reading the diff. I'd rather say that than pretend the classifier is smarter than it is.

Especially not for a post ostensibly capturing legalese change in company TOSs.

Did I start noticing it more or is the writing getting worse?

▲

illright 18 hours ago | parent [-]

That bit of text looks pretty unremarkable to me, what makes you think that no one talks or writes like this?

▲

nerdsniper 18 hours ago | parent | next [-]

I see it plain as day as well FWIW. Not just that section, but most of the headings.

The issue I have with it is that I can’t be sure whether a human actually double-checked anything. Writing in a “human” voice rather than robotic one serves as proof that someone actually bothered to really sign their name to the content. Otherwise I need to ground everything myself independently because the author’s reputation becomes meaningless if there’s no evidence they checked it.

I don’t have time or energy to get ground truth on everything I read. That would be exhausting and impractical.

▲

ande-mnoc 18 hours ago | parent | prev | next [-]

I assume you don’t use Claude a lot? If you interact with it on a daily basis, it becomes your second instinct to identify its writing. Anthropic models talk like they are penalized by making sentences flow naturally. "The X I didn’t expect" is basically a mandatory section of a Claude-authored blog post at this point.

	▲	arcfour 8 hours ago \| parent [-]
		Was it 4.5 or 4.6 that started using "load-bearing" all the time? Boy, the first time I saw that I knew immediately that was the Next Big Tic for Claude.

▲

x______________ 18 hours ago | parent | prev [-]

[dead]