I can’t bring myself to read text if I believe it to be AI-generated.
Now, I ask LLMs questions all of the time, and I do read those answers. I frequently use tools like ChatGPT and Claude as replacements for Google for answering specific questions; that’s not what I’m not talking about here. I’m also not talking about reading LLM-generated code. What I mean is, if I’m reading some sort of a document, if I suspect that the document was AI-generated, my motivation to read through it drops down to approximately zero. If I was browsing non-fiction books in a bookstore, and a book was marked as having been AI-generated, I wouldn’t pick it up.
Being honest with myself, I think this point of view is irrational. My personal primary goal for reading any sort of non-fiction document is to advance my understanding of a topic, or put new ideas into my head. In principle, it shouldn’t matter whether the words on the page were emerged from the thoughts of a human being or via an autoregressive stochastic process. In addition, I’m very far from being a perfect detector of AI-generated text, so there isn’t even a way for me to know whether a particular document I’m reading came from a human or a machine. Also, AI generation is a spectrum. No document is completely AI generated: they all start with a prompt was written by a human. Some texts will have been iteratively generated by a collaboration of human and AI. If I knew that an author had used AI as a copyeditor, or to tighten up some of their sentences, that wouldn’t bother me at all. There’s not some magical threshold in my head about how much AI assistance I would consider to be OK, nor would it ever be possible for me to know whether that threshold was exceeded unless the author explicitly told me.
And yet, despite knowing this, I’m just turned off by reading anything that strikes me as being AI-generated. If I’m asked to read a design document, and I suspect the doc was written by AI, I need to fight myself to actually get through it. I feel like a writer should always spend more time generating a document than a reader should spend consuming it, and asking me to spend more time on understanding something that someone else didn’t put the effort into writing feels like a violation of an implicit contract.
As I said, I think this is an irrational response. And I expect the quality of LLM writing to continue to improve over time, so that we stop referring to it as slop. But there’s just something, well, soulless about the idea of writing generated by a machine.
Surfing Complexity 