lumpenproletariatart of the deal
obelisk_complexI wanna know where to buy this guy’s keyboard, it must be really rare to have an em-dash key on it.
Fuckin’ em-dash, it’ll get you every time.
You do understand people who write for a living have been using em dashes for longer than AI, yes?
And you don’t need a fancy keyboard to press a few keys to bring up any Unicode character.
People suspecting AI based on the flimsiest logic is getting so fucking tiring with how frequently they are wrong.
Hear fucking hear. AI can pry the em dash from my fucking cold dead hands. Alt-0151 for life. And Alt-0150, too, when the en dash is more appropriate than a hyphen.
The best test for AI isn’t good grammar. It’s presenting relatively simplistic ideas in an overly flowery manner; repeating the same idea not in ways that help build an argument, but which are repeating the same idea in different words despite structurally seeming as though they’re presenting entirely new points; and other structural problems with the text.
yeah the em dash thing is absolute moron bait
like wow mate, you saw a long dash. call the fucking lab. some people are just punctuation goblins and have been since long before the slop machines turned up
- ChatGPT generated comment, people need to know that they wouldn’t be able to spot a well generated bit of text as AI these days. It’s dangerous thinking they can. Was a nice touch when it called itself the slop machine though.
Scientific studies suggest that people usually cannot reliably tell whether a piece of text was written by AI from style alone: accuracy is often only slightly above chance, varies a lot by genre, and drops further for more formal or scientific writing. For example, one 2024 study found average accuracy of about 57% for texts overall, while a 2024 teacher study found that even novice and experienced teachers could not reliably distinguish ChatGPT essays from student essays and were often overconfident in their judgments; likewise, a study of research abstracts found reviewers were largely unsuccessful, with only 38.9% positive identification. There is some evidence that people can improve with explicit feedback and training, but the overall research points to the same conclusion: humans may sometimes pick up clues, yet unaided judgments about whether text is AI-written are generally weak and unreliable, especially when the writing is competent and domain-appropriate. - ChatGPT