Its so stupid, because moat people vastly overestimate their ability to identify AI material, especially in shorter passages.
Mark Lawrence has done some (very unscientific) tests regarding this, and even when you're selecting for the type of person who reads an author's blog, the results were basically a coin toss.
Most of the linguistic fingerprints that people use to try and identify AI are not inherent properties of LLMs, but instead due to the default system prompt (e.g. "its not just __, but ___________"). These sort of writing tics are trivially easy to prompt around.
Yeah, I’m wary of this. I also write in the piece that some of my own sentence structures are beginning to look somewhat AI-like, so it’s only a matter of time before people accuse me of having an agent do my writing. I’ll have a look at Mark’s blog now - thanks.
“Most parodies of one’s own work strike one as very poor. In fact, one is apt to think one could parody oneself much better. (As a matter of fact, some critics have said that I have done so.) But there is one which deserves the success it has had, Henry Reed’s ‘Chard Whitlow’’.
I appreciate the author’s putting into words the LLM radar I start to feel when reading some articles.
Its so stupid, because moat people vastly overestimate their ability to identify AI material, especially in shorter passages.
Mark Lawrence has done some (very unscientific) tests regarding this, and even when you're selecting for the type of person who reads an author's blog, the results were basically a coin toss.
https://mark---lawrence.blogspot.com/2025/08/the-ai-vs-authors-results-part-2.html?m=1
Most of the linguistic fingerprints that people use to try and identify AI are not inherent properties of LLMs, but instead due to the default system prompt (e.g. "its not just __, but ___________"). These sort of writing tics are trivially easy to prompt around.
Yeah, I’m wary of this. I also write in the piece that some of my own sentence structures are beginning to look somewhat AI-like, so it’s only a matter of time before people accuse me of having an agent do my writing. I’ll have a look at Mark’s blog now - thanks.
The AI examples read like Judith Butler!
My instance of ChatGPT is peculiarly good at self-deprecation. Better than most humans.
T S Eliot:
“Most parodies of one’s own work strike one as very poor. In fact, one is apt to think one could parody oneself much better. (As a matter of fact, some critics have said that I have done so.) But there is one which deserves the success it has had, Henry Reed’s ‘Chard Whitlow’’.
Excellent essay! Thank you for the mention and link🙏
Keep up the good work, sir.