While LLMs can be prompted to write in many different styles, especially if you allow them to edit a text over multiple passes, the default "voice" of ChatGPT is surprisingly recognizable. For instance, the comment I'm replying to was clearly GPT-generated.
Yeah that’s pretty obvious, and yours sounds a bit like it too, especially starting off with “While…”. I do find it hilarious that the ChatGPT default style sounds very much like how I wrote in high school. I guess I was still training my own language model at that point, so it kinda makes sense.
GPT output always reads like a short story or essay, but HN et al. have a distinctly conversational aspect that I don't get with ChatGPT.
You're right about the high school writing style. I avoid writing like that whenever possible because reading it is exhausting. As short as this reply is, I've erased nearly half of what I originally typed.
I agree that the parent is clearly GPT-generated, but was it 3.5 or 4? I notice substantially improved human-like writing in GPT-4 that will probably be hard to detect. Even the current AI detectors struggle (although they have gotten pretty good at detecting 3.5 writing)
