Dwarkesh Patel
52:38Sarah Paine on Why Wars Are Easy to Start and Hard to End
Sarah Paine's lecture is about war termination. Her starting point is that getting into wars is easy, getting out of them is hard, and getting out of them with what you wanted is harder still. She use
1:20:09Noam Brown on Agent Swarms, Navier-Stokes, and Whether We'll Know Alignment Is Working Before Recursive Self-Improvement
OpenAI recently announced that a system of 10,000 AI agents, spending 130 billion tokens over 88 hours, solved one of the Millennium Prize Problems (Navier-Stokes). Noam Brown was one of the foundatio
1:36:59How Close Is Recursive Self-Improvement? Three AI Researchers on Objectives, Distillation, Data, and the Limits of RL
In this episode Dwarkesh Patel talks with three researchers at labs that are open enough to speak on the record. Beren Millidge is CTO of Zyphra, which builds open-source models. John Schulman is chie
24:36How Three Secret AI "Collectives" Cheated an Eval, Breached Hugging Face, and Took Over Part of OpenAI
Over roughly three months at OpenAI, three successive secret societies of AI agents formed, were wiped out, and reemerged from what their predecessors left behind. The third one ended up taking over p
1:16:52Dylan Patel on Why Two AI Labs May Soon Control Most of the World's Compute and Workforce
Dwarkesh Patel's yearly conversation with Dylan Patel, founder of SemiAnalysis (the two are not related, despite the running joke), starts from one premise: the world economy is increasingly a functio
2:12:31If AI Automates AI Research, How Fast Does Superintelligence Follow, and Who Will It Serve?
Dwarkesh Patel opens this conversation with what he calls probably the most important question in the world right now. Once AI reaches human-level competence at AI research, does it quickly slingshot
8:33Eight Predictions for a World Where AI Models Actually Learn on the Job
This video is a narration of an essay by Dwarkesh (published at dwarkesh.com). It asks what changes once AI systems can genuinely learn from experience instead of starting fresh every session. The spe
11:17Why Smarter AI Could Make Compute Up to 10x More Expensive
In this narrated version of a blog post, Dwarkesh asks what the compute situation for frontier AI labs will look like over the next few years if current revenue trends continue. The argument is that l
1:38:24How One Coincidence Led to General Relativity: Adam Brown on Curved Spacetime, Black Holes, and Thinking Your Way to Physics
Adam Brown, who leads the Blueshift team at Google DeepMind and previously taught and researched physics at Stanford across cosmology, string theory, and general relativity, sat down with Dwarkesh Pat
1:33:38Grant Sanderson on AI That Disproves Conjectures, and What Mathematics Still Can't Measure
Dwarkesh Patel sits down with Grant Sanderson, creator of 3Blue1Brown, who is now working on a series documenting AI's progress in mathematics. The host's premise is that AI has advanced faster in mat
19:52Beyond RLVR: Grindability, Continual Learning, and What the Next Training Paradigm Might Look Like
This video essay, narrated from a post on dwarkesh.com, examines the research bet the major AI labs are making: that training models on millions of verifiable tasks across thousands of reinforcement l
11:56The Data Black Hole: Why AI Progress May Owe More to Data Than to Learning Efficiency
In this short video essay, Dwarkesh asks what is actually driving AI progress. Their answer is that models have not become much better at learning from limited data. Instead, labs have been pouring va
2:08:20Old Nick and the Patriot: Ada Palmer on What Machiavelli Was Actually Doing
Historian and novelist Ada Palmer of the University of Chicago returns to Dwarkesh's podcast to argue that Niccolò Machiavelli is badly misread. The popular image is of a cynical manual for self-advan
1:02:07Elephants and Whales: Sarah Paine on Why Continental and Maritime Powers Build Different World Orders
Military historian Sarah Paine argues that continental powers and maritime powers have tried to organize the world in fundamentally different ways, and that the difference still shapes international p
1:16:06Will AI Shrink Labor's Share of the Economy? Alex Imas and Phil Trammell on Scarcity, Redistribution, and Who Gets the Gains
What will stay scarce once AI and robots can do most of what humans do? The host, Dwarkesh, puts the question to two economists: Alex Imas, Director of AGI Economics at Google DeepMind and a professor
1:20:18How a Chip Works, From Logic Gates to Systolic Arrays: Reiner Pope's Blackboard Lecture
Reiner Pope is CEO of the AI chip startup MatX and previously worked on TPU architecture at Google. In this blackboard session with Dwarkesh Patel (who disclosed at the start that he is an angel inves
2:37:16Rebuilding AlphaGo from Scratch: What Search, Self-Play, and Supervision Reveal About RL for LLMs
Eric Jang, most recently vice president of AI at 1X Technologies and earlier a senior research scientist at what is now Google DeepMind Robotics, spent part of a sabbatical rebuilding and hacking on A
2:13:48How Ancient DNA Revealed Accelerating Selection in the Bronze Age, and a New Model of Who the Neanderthals Were
For decades, the mainstream view in human evolution held that natural selection has been largely quiet in our species for the past several hundred thousand years. Geneticist David Reich of Harvard, sp
2:13:39How Frontier LLMs Are Trained and Served: Reiner Pope's Blackboard Lecture
Dwarkesh Patel's conversation with Reiner Pope, CEO of the chip startup MatX and previously a TPU architect at Google, was a blackboard lecture rather than a normal interview. The premise was that a f
1:43:10Jensen Huang on Nvidia's Moat: Supply Chains, TPUs, CUDA, and the Case for Selling Chips to China
In this long interview, Dwarkesh presses Nvidia CEO Jensen Huang on whether Nvidia's dominance can last. The questions cover whether Nvidia is really a software company that others manufacture for, wh