AI-powered
podcast player
Listen to all your favourite podcasts with AI-powered features
5min chapter
Jeremie Harris: Realistic Alignment and AI Policy
The Gradient: Perspectives on AI
The Pros and Cons of the Cheese Vector
The idea of a task vector is very simple. It's just like, hey, see if you can find a vector for this subtracted out from the activations that are going to do it. And there are a lot of these strategies have as well scaling properties that I think are quite desirable in constitutional AI. That scales implicitly. If you can identify things like cheese vectors or task vectors, that has, you know, good scaling properties as well. We ought to prepare ourselves for a world in which we may have a bag of tricks that could be just sufficient enough to get us to a point where AIs can help with alignment research.
00:00
Transcript
Episode notes
Get the Snipd
podcast app
Unlock the knowledge in podcasts with the podcast player of the future.
AI-powered
Discover
highlights
Listen to the best highlights from the podcasts you love and dive into the full episode
Save any
moment
Hear something you like? Tap your headphones to save it with AI-generated key takeaways
Share
& Export
Send highlights to Twitter, WhatsApp or export them to Notion, Readwise & more
AI-powered
podcast player
Listen to all your favourite podcasts with AI-powered features
Discover
highlights
Listen to the best highlights from the podcasts you love and dive into the full episode