Google DeepMind's Alex Imas argues AI-generated writing alienates readers because it lacks the private information needed to justify their attention
Chakrabarty notes that AI text signals skipped cognitive effort.
#449Kawin Ethayarajh@ETHAYARAJHAlso means that as continual learning improves and AI-assisted writing becomes indistinguishable from the author's own writing (or at least requires too much mental computation for most of us), we will need a more robust sense of what is slop and what is kino.
Alex Imas@alexolegimasI think this is just being Bayesian. People use AI in writing for two things mainly: 1) improving their writing after idea generation and 2) as a shortcut to idea generation (eg write me a few paragraphs making argument X and provide support). Most people are likely doing both 1) and 2). When seeing writing, a person needs to decide whether the effort spent reading and thinking about is worth it in terms of learning new information from the writer. If the writer uses AI for 2), then there is little private signal or unique information—it is the same information the reader can generate themselves, probably more concisely, from using the AI model. So it’s perfectly reasonable to detect AI writing (either through detector or just feeling the cringe) and to stop reading. The issue with using AI for 1) is that you are pooling with 2). It is very difficult to signal that you spent time and effort generating the ideas and have unique information to convey. This is all to say this: it is perfectly reasonable to use AI as much as you want for writing, but it’s worth being aware that it is also perfectly reasonable for others to not read it.
#1050Tuhin Chakrabarty@TUHINCHAKRPart of it is also people protecting culture :) When I read something I want to know what the writer is thinking. So obviously when I smell AI writing it signals the writer has not spent time thinking about the topic. Why should I as a reader spend time reading it ?
