ELK And The Problem Of Truthful AI
Astral Codex Ten Podcast
00:00
The a I and the E L K Report
Some of the possible action sequences are simple enough that humans can fully understand what's going on. Scott writes, we train the a i by running millions of simulations where it plays against simulated thieves. At first it flails randomly, but as time goes on, it moves towards strategies that make it win more often and learn how to deploy its dudads most effectively. It even starts extruding new traps and dudads. We did n't design things. We have no idea what they even do.
Transcript
Play full episode
