[marginalium]

On the alien characteristics of LLMs

21 May 2023

View original source »


On the alien characteristics of LLMs: the Waluigi effect.

Short version:

After you train an LLM to satisfy a desirable property P, then it’s easier to elicit the chatbot into satisfying the exact opposite of property P

Why?

When you spend many bits-of-optimisation locating a character, it only takes a few extra bits to specify their antipode.


Anthologies: Betterment, Somatic Architecture, Digital Architecture, Animal Sentience, Absit Omnia, On Ethics, On Thinking and Reasoning

View on main site »