"The Waluigi Effect (mega-post)" by Cleo Nardo
LessWrong (Curated & Popular)
00:00
A Conjecture About LLMs
In 101 Dalmatians, Cruella de Vil is the complete inversion of Roger and Anita. The wallowiji eigen-simulacra are attractive states of the LLM. Heading, superpositions will typically collapse to wallowijis.
Play episode from 24:48
Transcript
