The Bayesian Conspiracy cover image

179 – The Plan (to align AI), with John Wentworth

The Bayesian Conspiracy

00:00

E. Coli Models Are Different Than Your Thermostat Models

The idea here is that not every agent will use every abstraction. And in the case of very simple agents like E. coli, they may not even be capable of reasoning over every abstraction. But it does seem like there's sort of a general data type for the kinds of things that make sense as abstract objects. If you have an agent that's powerful enough to understand that general data type, then it should be able to look at the concepts that other agents are using. We could predict more accurately what the agent is going to be doing with their model of the tree and maybe possibly even change their model ofthe tree.

Transcript
Play full episode

Remember Everything You Learn from Podcasts

Save insights instantly, chat with episodes, and build lasting knowledge - all powered by AI.
App store bannerPlay store banner