
4 - Risks from Learned Optimization with Evan Hubinger
AXRP - the AI X-risk Research Podcast
Is There a Way to Decompose a Module?
i'm doing some reasoning about what types of mess optimisers we might see, and it seems like there's going to be paramters that are dedicated to the gulls or the masa objective. And one thing i'm wondering is, if you think about these like classically understood part s of an optimizer, so er something doing searchright? So like y maybe has perception, belief, skulls and planning. But do you think there's necessarily going to be different perometers, am of your machine learned system, um, for these different parts? Or do you think that potentially there might be parts, like spread across r perometers, om divided to different parts in which case you
00:00
Transcript
Play full episode
Remember Everything You Learn from Podcasts
Save insights instantly, chat with episodes, and build lasting knowledge - all powered by AI.