The dashes thing is really only a tell for high schoolers and obvious illiterates –writers can and do use them. Next we need to teach GPT about interrobangs.
thank you for saying this. I'm not a writer, just vaguely literate, and I find em-dashes to be extremely useful as a way of approximating conversational speech without a string of comma splices.
Additionally, the observation about AI using dashes really only applies (if it applies at all) to informal text conversation, not published articles. Here's a random NYT article from 2014 that uses 4 em-dashes within the first 3 paragraphs. https://www.nytimes.com/2014/12/21/fashion/the-year-of-taylo...
It seems like everyone watched some video essay about how em-dashes are a sure sign of AI, and they're just parroting it without question. The thing that apparently goes over people's heads is that the reason AI tends to use a lot of em-dashes is because the text in their training data uses a lot of em-dashes, and the training data is largely published articles and books, so it's a terrible heuristic for whether a published article or book was written using AI.
The thing that actually is somewhat more telling is that many people will use hyphens instead of dashes, eg on a computer, I typically type '--' instead of an em-dash, partly because my xcompose setup is inconsistent, partly because I write most of my text in editors that use monospaced fonts and the distinction between -, – , and — is extremely subtle in most fixed-width fonts, for obvious reasons. But on macs and many phones, as well as in google docs and similar, by default a hyphen will be autocorrected to an en- or em-dash depending on context, so it's not really a tell that the entire thing was written with AI, just that there was possibly non-human involvement. But also, a lot of people actually just know how to type and use dashes. It's not really that hard.
Additionally, the observation about AI using dashes really only applies (if it applies at all) to informal text conversation, not published articles. Here's a random NYT article from 2014 that uses 4 em-dashes within the first 3 paragraphs. https://www.nytimes.com/2014/12/21/fashion/the-year-of-taylo...
It seems like everyone watched some video essay about how em-dashes are a sure sign of AI, and they're just parroting it without question. The thing that apparently goes over people's heads is that the reason AI tends to use a lot of em-dashes is because the text in their training data uses a lot of em-dashes, and the training data is largely published articles and books, so it's a terrible heuristic for whether a published article or book was written using AI.
The thing that actually is somewhat more telling is that many people will use hyphens instead of dashes, eg on a computer, I typically type '--' instead of an em-dash, partly because my xcompose setup is inconsistent, partly because I write most of my text in editors that use monospaced fonts and the distinction between -, – , and — is extremely subtle in most fixed-width fonts, for obvious reasons. But on macs and many phones, as well as in google docs and similar, by default a hyphen will be autocorrected to an en- or em-dash depending on context, so it's not really a tell that the entire thing was written with AI, just that there was possibly non-human involvement. But also, a lot of people actually just know how to type and use dashes. It's not really that hard.