tropes.wtf

Of course I found this in a Hacker News comment. Some background “AI tells” have been a thing for nearly as long as this heLLM-scape has, and they’re just about as bad as the actual machines they supposedly unearth. The whole problem with AI writing is that it imitates—badly—human writing, so I don’t know why anyone thinks that detection is possible. Especially automated detection. I feel like there’s some truism there about P=NP, or how the map isn’t the territory; how can a computer judge a Turing test? tropes.fyi This website offers a few functions to the intrepid web surfer: A directory of tropes that betray LLM authorship A vetter for detecting these tropes on any webpage AI;DR, which purports to reverse-engineer the prompt given to an LLM-written page tropes.md for giving your LLM so that it can evade detection (“let’s play cat and mouse,” it says) I’m going to go through each of these and talk about why this site has increasingly enraged me since I stumbled upon it. the…