I've found that people who say this sort of thing rarely change their beliefs, even after being given evidence that they are wrong. The fact is, as numerous people have pointed out, Word and other editors/word processors change '--' to an em-dash. And the "slang version" of an em-dash is "I went to work--but forgot to put on pants", not "I went to work - but forgot to put on pants".
BTW, "humans almost always tend to use" is very poor writing--pick one or the other between "almost always" and "tend to". It wouldn't be a bad thing if LLMs helped increase human literacy, so I don't know why people are so gung ho on identifying AI output based on utterly non-substantive markers like em-dashes. Having an LLM do homework is a bad thing, but that's not what we're talking about. And someone foolishly using the presence of em-dashes to detect LLM output will utterly fail against someone using an editor macro to replace em-dashes with the gawdawful ' - '.
But do you call that latter thing you do “an em-dash”? Do you tell a peer “You should put an em-dash here” when what you mean is a “space en-dash space”?
P.S. Why would someone be "suspicious" of people doing their writing in Word and copying it into a comment field? Suspicious of what, exactly? What crime is being committed? The issue here is AI, not people's workflow methods. I have sometimes written lengthy comments in my editor (emacs) which gives me many more editing features, doesn't throw away my work with the wrong keystroke (not a problem at HN but it is at sites where the comment field is a pop-up), and doesn't randomly freeze or slow down radically (this seems to be a problem with my browser).
I reject everything else about that poorly reasoned "suspicious" response as well.
I've found that people who say this sort of thing rarely change their beliefs, even after being given evidence that they are wrong. The fact is, as numerous people have pointed out, Word and other editors/word processors change '--' to an em-dash. And the "slang version" of an em-dash is "I went to work--but forgot to put on pants", not "I went to work - but forgot to put on pants".
BTW, "humans almost always tend to use" is very poor writing--pick one or the other between "almost always" and "tend to". It wouldn't be a bad thing if LLMs helped increase human literacy, so I don't know why people are so gung ho on identifying AI output based on utterly non-substantive markers like em-dashes. Having an LLM do homework is a bad thing, but that's not what we're talking about. And someone foolishly using the presence of em-dashes to detect LLM output will utterly fail against someone using an editor macro to replace em-dashes with the gawdawful ' - '.