When I read S. Marinelli’s post Stefano, the Human LLM of the 2000s1, I thought “This is exactly how I felt last week, when I was accused I used LLMs for writing a professional piece. My experience is quite similar.”
I was writing a technical article for a client. When I was accused of using LLM, my fist reaction was anger and surprise (fedi thread2). Then I pushed back, which I rarely do. Their supervisor apologized, took responsibility, which made it possible to for me to carry on. He’s a decent person, I would say.
But what exactly triggered the accusation? What kind of writing style?
It was a paragraph that included 8 bullet points and slashes / with synonims. This is not a particularly good style to group concepts and present them to a reader, but I’m using this style since I write technical pieces, that is from 1999.
It was something like:
Examples of <things> are:
- thing 1 / synonym (short explanation)
- thing 2 / synonym (short explanation)
- thing 3....Nothing special, at least to me.
The comment I received was:
“Could you please rewrite these bullet points? Their current structure, and especially the sentence “… and thing 1 / synonym”, makes it quite obvious that they were generated by AI.”
I am not even sure if this was a human or LLM generated comment to the mentioned paragraph. If it was a human comment, I ask myself: what triggered it? Did the commenter read a lot of LLM generated text and the structure was similar to them? I’m genuinely interested in the thought process
If it was LLM generated comment: what is the reliability of LLM tools for detecting LLM generated content? Mixed, I think. Which is … bad. Does occasional successful detection out weights false positives?
Nevertheless, I rewrote those bullet points afterwards into a more readable paragraph. Not because they were generated by LLMs, but because they’re not very pleasant to read for non-technical target groups (as I learned).
What surprised me even more is that some (other, non-related) companies now have a policy for public writing that say (from one of the comments3,4):
…for any public Oxide writing, not only must an LLM not be used to write, but it must be further true that the text not be read as LLM-authored. This is a higher bar, and arguably a weirder one: not only must it be human written, it must be unimpeachably so.
Huh. I’m using ‘my’ style for 25+ yrs. Should I alter it now to avoid the discomfort of readers?
… and what happens if I alter my style? LLMs will learn it and then I have to alter it again? Not sure if I’m ready for that. But on the other hand, it already influenced me, I already altered my style, as I explained above.
Now, after 2 weeks when I look back and after reading similar experiences from artists (e. g. from D. Revoy5), I can better prepare my reactions in the future. But still, I will probably oscillate between these two states:
- Should I be flattered next time when someone accuses me of using gentAI (because LLMs learned from writing styles like mine)?
- Or should I take it as an insult: you writing is shallow because of weirdly empty methaphors?
Not sure. Will wait for the next time.
P. S. one of the commenters suggested that I might be a LLM that ‘thinks’ it is a human. Might be true that we live in a simulation, but I am quite sure that it is not powered by LLMs.
- https://my-notes.dragas.net/2026/10/08/stefano-the-human-llm-of-the-2000s/ ↩︎
- https://mastodon.social/@po3mah/117347498213932137 ↩︎
- https://masto.ai/@bovine3dom/117348727817690490 ↩︎
- https://rfd.shared.oxide.computer/rfd/0576 ↩︎
- https://www.davidrevoy.com/article1164/when-online-commenters-detect-my-art-as-ai ↩︎