Hacker News new | ask | show | jobs
by nojs 1 day ago
> Em-dash (—) is COMPLETELY banned. It is the LLM's signature stylistic crutch and it is the #1 visual Tell in production tests. There is no "limited use" allowance, no "natural language frequency" allowance, no "in body copy is fine" allowance. None.

I guess the em dash is really dead.

7 comments

Probably not, because now you'll have all of the people who are worried about this ensuring their AI generated text never ever uses em-dashes, and then that will be the new tell.

Or we can just be more concerned with whether something is well written and well presented and accept that sometimes that is going to be AI written text.

My personal "tell" is that I use -- instead of em-dash. If you get a message from me with an actual "—" character it's almost certainly AI generated.

I guess that's kind of my duress code :)

> Or we can just be more concerned with whether something is well written and well presented and accept that sometimes that is going to be AI written text.

Slop is slop, it doesn’t matter if it was human or LLM written. If a code comment or Slack message is 200 words long but still makes no fucking sense then it’s slop. A message can be 10 words long and riddled with typos but if the message is understandable it’s not slop.

I've been posting my writing on the internet since 1990, from a Mac where we all learned how to make en and em dashes, umlauts, accents, degree symbols, proper single and double quotes, and so forth from Day #1 and I have absolutely no plans to stop using em dashes.
Nice try, ChatGPT.

Though I agree with GP: models and harnesses are being updated so hard to avoid any use of em-dashes that soon it will be a tell of human writing. The the pendulum will swing back and forth, forever.

I have been seeing people use it more who I am fairly confident are not otherwise using AI to write public-facing communications. It's almost like there's a minor movement to try and reclaim it. Either that or it's somehow getting past my AIdar.
It's notable that this appears to be used to generate frontends that look more authentic than the typical AI generated site, not as a test for what's acceptable from a human contributor.
This quote is so egregiously stupid, I'm instantly disinterested in this product now. No reason to assume the rest isn't the same applied cretinism as the quoted sentence.
I suppose all the exotic unicode characters are dead now. Anything that's too annoying for a human to type but trivial for AI to generate is probably done for. I used to enjoy using those characters. Sigh.
They have never been annoying to type on a Mac.
And yet they use it on the README