What story are we telling?
A solution to the AI Alignment problem.
A 2026 study found that changing the story changes the model: misaligned behavior dropped from 45% to 9% on its test.
#AIAlignment#aisafety
Paul Christiano: There's a 50% chance of losing control to AI if we race to build it as fast as possible. It might happen within a year.
Some AI CEO's & investors: Let's race to build it as fast as possible!
(Paul is a brilliant AI researcher who invented RLHF and lead the science team at the US government's AI safety institute.)
@elonmusk@elonmusk I think two people are a necessary addition to xAI, @ESYudkowsky for alignment & @realGeorgeHotz for scaling & improvements .. so how about you extend an invitation to them both
@lexfridman@lexfridman ask Elon what the gameplan is to stop AI from destroying the world, what's the latest with Neuralink, and how far is he from getting government approval for FSD. Thanks!