Basically, humans are mostly unable to tell apart human and AI text. They’re on average about as good as rolling dice. And they’ll say something is AI for the wrong reasons/cues.
I don’t think that’s the entire story. I heard about the study from some German YouTuber. They spent several days writing the script for a video until the text was perfect (from their perspective). And that made their audience think they turned to AI assistance. So… Seems it both makes people think there’s AI at work: Flaws in it, or being too flawless.
Idk. And from skimming the study, I gather you can train an AI to write exactly with the right amount of contractions etc, so we’re F-ed anyway.
I take something being overly bland, repetitive or smooth in a weird way as indicators as well. After that I’ll look at ChatGPT’s distinctive idiosyncrasies. I think I’m not too bad at it, but sometimes it’s really difficult. Even more so with some styles of images. Like pop art. I wouldn’t know if something was made by AI or an artist. I mean by the image itself. I can probably tell if there’s some layers of meaning in it, or if it’s just some random thing depicted.
I guess “sucks” is relative, but for a normal person a lot of stuff that’s “perfect” actually sucks. It’s overly verbose, with lots of unnecessary adjectives and descriptors and punctuation that bloats the word count. It’s technically good, but masturbatory. The “Look at how smart I am!” school of writing. There are people who write like this from their training, and they will be mistaken for AI.
There’s a study on this phenomenon, regarding text: https://arxiv.org/abs/2206.07271
Basically, humans are mostly unable to tell apart human and AI text. They’re on average about as good as rolling dice. And they’ll say something is AI for the wrong reasons/cues.
Humans can tell if something sucks, and if it sucks they’ll call it AI. That’s not a reliable way to detect AI, though.
I don’t think that’s the entire story. I heard about the study from some German YouTuber. They spent several days writing the script for a video until the text was perfect (from their perspective). And that made their audience think they turned to AI assistance. So… Seems it both makes people think there’s AI at work: Flaws in it, or being too flawless.
Idk. And from skimming the study, I gather you can train an AI to write exactly with the right amount of contractions etc, so we’re F-ed anyway.
I take something being overly bland, repetitive or smooth in a weird way as indicators as well. After that I’ll look at ChatGPT’s distinctive idiosyncrasies. I think I’m not too bad at it, but sometimes it’s really difficult. Even more so with some styles of images. Like pop art. I wouldn’t know if something was made by AI or an artist. I mean by the image itself. I can probably tell if there’s some layers of meaning in it, or if it’s just some random thing depicted.
I guess “sucks” is relative, but for a normal person a lot of stuff that’s “perfect” actually sucks. It’s overly verbose, with lots of unnecessary adjectives and descriptors and punctuation that bloats the word count. It’s technically good, but masturbatory. The “Look at how smart I am!” school of writing. There are people who write like this from their training, and they will be mistaken for AI.