Hacker News Reader

Comment by Volundr | Hacker News Reader

This is something that frustrates me a lot when I see people jumping on a piece of writing as AI generated, these AI-isms didn't show up in a vacuum. Yes, AI commonly overuses some patterns, but it got those patterns from it's training data because real people wrote (and continue to write) that way. A few matching sentences here or there is not the smoking gun people make it out to be.

View on HN

The other thing that is characteristic of AI text is messing up the underlying meaning or fine details of what it's saying. It's not just that it's imitating people who write bad contrasting clauses, it also can't write good ones reliably.

edit: Ironically that last sentence is not super clear (but hey writing is hard and I'm constantly exposed to slop). My point is the nature of AI makes its writing worse than its training data, it's not all bad training data.

This is only true if you consider those who write corporate press releases and post on LinkedIn to be real people. /s

Personally I find writing of this type grating whether the source was AI or the amalgamated output of a committee of humans in a PR department.

To be fair to committees, the above quote is explicitly from a single person and I expect that is true of most AD copy simply because having one person write it is cheaper.

Committees are more about stamping out flavour by making what exists more bland, rather than adding endlessly regurgitated idioms.

> This is only true if you consider those who write corporate press releases and post on LinkedIn to be real people.

You make a fair point. Objection withdrawn your honor.

Reply on news.ycombinator.com