The next big breakthrough will be AIs learning on the job
The podcast argues that while current AI progress relies heavily on reinforcement learning in verifiable, replayable environments (RLVR), true general intelligence requires solving two deeper challenges: sample efficiency during training and continual learning from real-world deployment.
Sign up to read the full summary
Free AI-powered recaps of every episode, delivered to your inbox.
Free forever for up to 3 podcasts. No credit card required.
Mentions
Books, people, and resources mentioned in this episode.
People
Companies
Products & Tools
Topics
Listen to This Episode
More from Dwarkesh Podcast
The data black hole at the center of AI
Eric Jang – Building AlphaGo from scratch
Grant Sanderson – AI and the future of math
Alex Imas and Phil Trammell – What remains scarce after AGI?
Reiner Pope – The math behind how LLMs are trained and served
Michael Nielsen – How science actually progresses
Related episodes
From other shows on similar topics.
Ep 87: Gemini Co-Lead on World Models, RL's Next Domains & Continual Learning
Unsupervised Learning with Jacob Effron
How to Build the Future: Demis Hassabis
Y Combinator Startup Podcast
Open Source Wins, AGI Is Here, and Scorsese's AI Toolkit with CEOs of Cerebras & Black Forest Labs
All-In with Chamath, Jason, Sacks & Friedberg
Code AGI is Functional AGI (And It's Here)
The AI Daily Brief: Artificial Intelligence News and Analysis
Anjney Midha's Plan to Radically Lower the Price of Compute
Odd Lots
The AI paradox: More automation, more humans, more work | Dan Shipper
Lenny's Podcast: Product | Career | Growth
Get summaries like this every morning.
Free AI-powered recaps of Dwarkesh Podcast and your other favorite podcasts, delivered to your inbox.
Free forever for up to 3 podcasts. No credit card required.