alt Hacker NewsAnd there might be no way to prove you really wrote something Post-2022. I wrote a long article, all by hand. I never used any LLMs, even for searching. I checked it with a couple of AI detection tools and they confidently said that 60% of the article was written by AI.
Replies
I track every change in the fiction I write in git. Its not hard proof but it at least shows how the prose evolved over time and is something like a proof of work.
Every time I attempted to convince people to not use Pangram on HN I got downvoted.
Most "AI detection" tools are BS, report AI writing for all writing. The current issue with AI writing is that it has a very generic, easy-to-spot style if you spend even a bit of time working with it. If everyone in the world spoke in the same manner, used the same punctuation, spoke in exactly the same catchphrases, the world would lose its richness, and that is the problem with AI writing and communications in general - it has 0 personality, and humans by nature engage with strong and unique personalities. Over time, people will realize the futility of using AI in creative writing, or AI will get really, really good at not just sounding human but being human.
Yet, from what I can see, AI writing is mostly used by people who don't know a thing about writing, and because they have bad taste, they do not see what's wrong with AI writing and put it out there.
At the end, you write for a purpose: for marketing copy, etc., you would require a different type of writing talent than something like writing a fiction book. But AI doesn't understand this nuance; it has only a default type of communication, which is highly optimized for being a chatbot. It is possible to write a very good text using AI if you have taste and you know what you're doing, but most people don't.
Similarly, a lot of vibe-coded apps are garbage, but because the people creating them lack software domain knowledge and don't even know what they don't know, they think it's good and put it out.
We have a massive problem here that's not just limited to writing - the promise of AI for the mainstream market is that you can replace domain-specific knowledge and have world-class execution in any vertical with just AI, but that's very overhyped imo and doesn't stop the people who don't have domain experience to try out stuff with AI and not realize what they made is a steaming pile of shit in reality.
The thing is, llm token frequency was derived from human writings like yours, and rlhf for good writing practices, like the emdash. So getting ’detected’ on good writing it’s unfortunately to be expected. My broken ESL english is much safer for now