askchapo

It's nonsense, but it reads like genuine hand-made nonsense. Consider the style of:

Speaker context → encode as texture → transmit → decode in listener context
C_1, P_1 → texture → → → C_2, P_2 → shapes

LLMs like speaking in sentences or bulleted lists. Yes, I'm sure you could prompt it to speak in the style of notes to generate something like the first line, but how much would you have to torture one before it starts outputting something like the second? The only way to get something that sounds this much like a nonsense paper is to ask it to write in nonsense-paper style, which a nonsense author wouldn't do because they don't know that what they're doing is nonsensical.