There is something uniquely difficult to me about reading LLM generated text. I find myself going over it many times until I can understand.
I think text written without actual understanding is very easy to skim, but very hard to understand.
Does anyone else find this?
@minmi I often find myself rereading it on the odd occasion I've had to because its very good at sounding like its going somewhere and then not. This makes me think I've missed something when its just not there to miss.
@ackthrice exactly. I find it very mentally exhausting.
@minmi yes. People who are not bothered by LLM text likely do not read it or have poor reading comprehension skills so, to them, it looks no worse than other things they read. There has probably always been this discrepancy in experience but it's never been as discoverable.
@weyoun6 I think this is exactly the issue
@minmi I’m grateful to say not sure. It’s not something I’ve been exposed to, as far as I know. I’ve worked very hard to avoid it at all costs. How do you know when it’s AI writing?
@mjg I guess I don’t know for sure. But given the source, and how completely opaque it was, and the style, I think so. But I have to comment on this doc for my job and I just want to swim out into the ocean.
@minmi ugh, so sorry!
@minmi I try to avoid it. I don't think I've ever attempted to closely read text I knew to be LLM generated
@scott I really don’t want to but I probably have to in this case and it is so exhausting. It’s like the least fun funhouse ever.
@minmi My personal experience with a lot of LLM text is that there is a specific context and framing that I held in mind, and the LLM just never wrote something that follows the same wavelength, so I need to context switch, and that takes time, albeit maybe a minute or two of going back and forth.
@minmi Absolutely this. On the surface, it usually reads OK. It’s only when you actually try to read it in detail, you notice the random switches in terminology, style and syntax that make it really difficult to parse. It’s also quite capable of logically contradicting itself within the same document. Sure, humans do this too, but there are usually signs that the (human) author isn’t confident, making this easier to spot…