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