Do your writing in a public place, then ask Flock to turn over their surveillance videos of you writing it.
post
In programming, we use a tool called "Git" to keep track of changes we make to a piece of software, well, and also to collaborate with other devs.
I've been using Git to have a history of how my text-based creative projects evolve (not just writing 😅). And then I'm publishing those like an open-source code repository. So, others can go through the history and look at the first draft, or the final edits, or everything in between.
Obviously, though, this does mean that your final result is public as well, unless you find a way to only give access to paying customers, I guess.
To give an example, here's a short tech explainer thingy I wrote: https://trem.codeberg.page/rust/macro_rules.html
And here's the complete change history: https://codeberg.org/trem/pages_src/commits/commit/830718aaa788ac00fb7ba9635c405f7c18cc76fb/src/rust/macro_rules.md
Really useful answer, thank you
You can also do that and keep your git private, maybe available on request for proof.
There's no way. Did you use Google search when researching something? You just used AI.
Watch a related video to something you just watched? You just used AI.
The entire concept of being able to detect AI use is flawed from the start. It might have worked—as a concept—if there was just one AI model. Then you could check against its most common outputs and give it half sentences to see if it generates the same output as what the author wrote.
Now we have dozens and dozens of models and they have new releases every few months. You'd have to perform that kind of checking for each and every version that exists!
Let's say you wanted to detect Gemini: Well, they're on 3.8 (Flash) now but just a few weeks ago they were on 3.6. Is the output of 3.8 really that different from 3.6? Yes! Enough that it would make detecting if it wrote some part of a sentence nearly impossible.
Was this sentence modified by Claude, ChatGPT, Gemini, DeepSeek, Grok, Qwen, GLM, Llama, or any number of dozens of versions of dozens of models with these numbers increasing exponentially over time?
It's just not going to happen.
There is no way. Even you filmed yourself writing it long hand, maybe you're just writing down what ChatGPT generated a little while before.
I just think the idea that someone rewrote chatgpt stuff by hand is proof enough that they're putting the effort in: either they're insane enough to copy it out and waste all that time reproducing slop (in which case you can probably identify it from analysing the text) OR they didn't, hence why they evidenced their handwritten workflow.
Rather than trying to convince people that you don't use AI, just write something personal and creative. Develop your voice. The plagiarism machine is unable to generate text that is consistently human-like and unique.
I'm occasionally accused of using AI for my writing. I like short and direct sentences. I use bullet points. I guess people think these things are tells for AI. I don't care what anyone thinks. I use my own head, and I'm completely uninterested in any sort of automated feedback, up to and including spelling and grammar check. If someone like me that never uses AI is accused of it, it must be really difficult to differentiate between human and machine writing.
As to your idea about publishing notes, that's part of the process a good creative or essay writing class will teach you. They may require you to create physical note cards and an outline as a preliminary assignment before the actual writing is due. Its good practice for people that don't know how to organize their thoughts, and its pointless for anyone that does. My process is usually one or more text dumps to create content blocks, merge them together, streamline the narrative through the piece, and then a couple rounds of edits. I work with multiple text files that morph as I go, rarely backed up or versioned. There would be nothing for me to offer as half-done work.
If someone like me that never uses AI is accused of it, it must be really difficult to differentiate between human and machine writing.
It's more than that, I think. Just like the "increase the contrast to spot AI" folks who've never heard of jpeg compression, the AI witchunters (and I will call them that when they start accusing innocent people of using genAI) have gotten so bad that it almost doesn't matter how you write anymore. I haven't yet been accused of sounding like AI but as a lover of the en and em dashes, colons, et cetera, I feel it's only a matter of time because people were absolutely convinced that any em dash was automatically a sign of the devil AI writing.
To whom? It what context?
Grammatical errors are an indication.. You could also speak to people about the topic..
To the reader. To editors or publishers. In case you were facing a hate storm and someone started insinuating your work was just LLM writing.
Right, but I was implying that the most appropriate response will depend on the situation. Better to try to deal with a specific instance than to worry about an amorphous general problem (I suspect you're not likely to face a hatestorm just for LLM use, especially if you don't actually use them in the first place...)
Especially in creative writing, every sentence and word choice matters. You could write a companion piece that does a critical analysis of your writing that explains the choice for every sentence. Yes, a.i. could produce that but it would make no sense at all. And if it ever did made total sense then at that point the machines have passed the test (but this will never happen)
publish it in the 20th century
Ever read Borges’ Tlön, Uqbar, Orbis Tertius?
You could fake a publication date in the 20th century, then have AI generate a bunch of circular fake references and seed them on forums and wikis, making it look like there’s an extended history of other works referencing your fake work.
all 26 comments