AI-powered
podcast player
Listen to all your favourite podcasts with AI-powered features
3min chapter
David Krueger–Coordination, Alignment, Academia
The Inside View
How to Optimize a Proxy for Learning
A reward function is a proxy for the thing that you actually care about and maybe it's much simpler but if you optimize that proxy like it's going to be fine. So we formalize that in this definition that says a pair of reward functions is hackable so like the proxy could hack the real reward function. When say we were here is like expected reward like expected some reward over time of a policy not like reward on the principal state right. And so the task is like good reward specification like having the perfect outer alignment so the good reward.
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