i guess the real fun idea was if a model that fucks with weights was itself fucked with such that any model it fucks would become cursed and give it to the next one it edits. but they don't apparently have a proof of that yet.

Ahh, so it's Reflections on Trusting Trust, but with LLMs instead of compilers.

https://www.cs.cmu.edu/~rdriley/487/papers/Thompson_1984_Ref...