ELK And The Problem Of Truthful AI
Astral Codex Ten Podcast
00:00
The E L K Contest - What Would a Human Expect to Hear?
A r c team discussed two classes of translation system. The direct translator, good, looks at what the security a i is thinking and faithfully translates it to its programmes. The human simulator, bad, focuses on what it would expect a human to think in that situation and tells its programme as that. A l k contest was come up with a strategy that insures your reporter a i ends up the direct translator and not as the human simulator.
Transcript
Play full episode
