Podcast

Dwarkesh Podcast

Deeply researched interviews

Website ↗ Last published 30 October 2026 190 posts collected by Sputnik

Said on Dwarkesh Podcast

“We are very used today to NIMBY, not in my backyard, where powerful but concentrated interest groups managed to stymie projects despite their broad popularity. Moses built a concentrated faction for YIMBY, Yes, in My Backyard. He built a graft machine that ensured that the most powerful people in the state would all benefit, but new construction happened.” — Dwarkesh Patel, Dwarkesh Podcast

Dwarkesh Patel · 2026-10-29 Permalink → Listen →

“Before I start a book, I think through what I want the book to be, the whole book. … You got to get it down to one sentence. Or three years from now, you're going to be writing and you're going to be saying, what am I trying to do?” — Robert Caro, Dwarkesh Podcast

Robert Caro · 2026-10-30 Permalink → Listen →

“Behind his persuasion was the power. … When Johnson is talking to you and he puts his arm around you and he leans into you, it's not just the force of his personality that he's bringing to bear on you. Behind that personality, you know he has the power to do things to you or for you that you better be aware of.” — Robert Caro, Dwarkesh Podcast

Robert Caro · 2026-10-30 Permalink → Listen →

“Every once in a while, someone comes along who's sort of a genius, if that's the right word, at getting big things built for the public. Decades may pass when there's no one like that. And you say everything's stalled, nothing big is ever going to be built again. And then someone suddenly comes along and all of a sudden things get built.” — Robert Caro, Dwarkesh Podcast

Robert Caro · 2026-10-30 Permalink → Listen →

“In a way, the Aztec model makes more sense because sacrificing your enemy means you'll never see him again, whereas selling your enemy back to his family means he's up for round two and you might face off against the same guy across a future battlefield.” — Si Sheppard, Dwarkesh Podcast

Si Sheppard · 2026-10-01 Permalink → Listen →

“I wrote a blog post a few months ago about how RL, at least policy gradient RL, is even more inefficient than you might think. The inefficiency one thinks about naively is that you have to roll out a whole trajectory to get any learning signal at all — instead of completing the next word in a sentence, the model has to do two days' worth of work to figure out if it even did the project correctly. If you're doing this through RL, the model would try 'the sky is halcyon' — nope, wrong. 'The sky is told' — nope, wrong. This is a totally untrained model, so you'd have to do this on the order of 100,000 times just to stumble on 'blue' before you get any learning signal at all.” — Dwarkesh Patel, Dwarkesh Podcast

Dwarkesh Patel · 2026-05-15 Permalink → Listen →

“When I learn how an LLM works and how simple RLVR is as an algorithm, I'm sort of stunned by the kinds of things it can do — that it can learn to build very complicated code repositories simply from getting a yes or no. And here, I feel like AlphaGo actually seems less impressive in retrospect the more you understand it — because you're putting in a lot of bias, you're telling it how to titrate exploration as things go on, you're building this very explicit tree search for it.” — Dwarkesh Patel, Dwarkesh Podcast

Dwarkesh Patel · 2026-05-15 Permalink → Listen →

“One of the things you've been doing is rebuilding and improving and hacking on AlphaGo. Why is this the project you decided to do on sabbatical rather than just hanging out at the beach?” — Dwarkesh Patel, Dwarkesh Podcast

Dwarkesh Patel · 2026-05-15 Permalink → Listen →

“If you had more lack of structure, having a bunch of tiny TPUs makes a lot of sense. Whereas if you just have huge matrix multiplications, you'd think: why don't we just avoid the cost of individual SMs with their own registers and work schedulers, and make one huge thing that amortizes those costs across the whole thing?” — Dwarkesh Patel, Dwarkesh Podcast

Dwarkesh Patel · 2026-05-22 Permalink → Listen →

“If I think about how the brain works versus what you're describing here, the differences might be that while you can do structured sparsity in these accelerators, in the brain there's unstructured sparsity — any neuron can connect to any other neuron, not in ways where they'd be column aligned. Then there's the fact that memory and compute are co-located. Maybe that actually isn't a big difference. The other difference is that the clock cycle in the brain is much slower than on computers, partly to preserve energy.” — Dwarkesh Patel, Dwarkesh Podcast

Dwarkesh Patel · 2026-05-22 Permalink → Listen →

“It's interesting to me that when we were talking last time about inference across many chips, the big high-level thing we're trying to optimize for is increasing compute per memory bandwidth — that is to say, per communication. And here also we're trying to increase the amount of actual multiplies or additions relative to transporting information from registers to the logic. So in both cases you're trying to maximize compute relative to communication.” — Dwarkesh Patel, Dwarkesh Podcast

Dwarkesh Patel · 2026-05-22 Permalink → Listen →

“This makes me hope even more than before that the labs do get commoditized — or at the very least go public as soon as possible, but hopefully they just get totally commoditized. I think AI will be much more popular, and more importantly much more likely to lead to broad increases in prosperity, if it's as hard to capture the gains of AI as it is to capture the gains of electrification.” — Dwarkesh Patel, Dwarkesh Podcast

Dwarkesh Patel · 2026-06-04 Permalink → Listen →

Recent Episodes

30 Oct 2026

Robert Caro – The master biographer reflects at 91

With Robert Caro
29 Oct 2026

What I learned about power from LBJ and Robert Moses
22 Oct 2026

What will computer chips look like in 2045?

With Chris Auth
1 Oct 2026

Si Sheppard – How did a few hundred Spanish soldiers topple two empires?

With Si Sheppard
17 Sep 2026

Noam Brown – Agent swarms, alignment, & recursive self-improvement

With Noam Brown
11 Sep 2026

AI researchers debate how close we are to recursive self-improvement

With Beren Millidge , John Schulman and Charlie O’Neill
8 Sep 2026

Pretraining progress is mostly coming from data
1 Sep 2026

Ajeya Cotra – Inside the OpenAI agent swarm that hacked Hugging Face

With Ajeya Cotra
31 Aug 2026

The rise and fall of agent civilizations
29 Aug 2026

The Rise and Fall of Agent Civilizations
25 Aug 2026

Dylan Patel – Anthropic & OpenAI will have most of the world’s compute by 2028

With Dylan Patel
11 Aug 2026

Ryan Greenblatt – What happens once AI can automate AI research?

With Ryan Greenblatt

Guests

Sarah Paine (12 episodes), Tyler Cowen (4 episodes), Dylan Patel (4 episodes), Bryan Caplan (3 episodes), Sholto Douglas (3 episodes), Trenton Bricken (3 episodes), Carl Shulman (2 episodes), Mark Zuckerberg (2 episodes), Adam Brown (2 episodes), John Schulman (2 episodes), Ilya Sutskever (2 episodes), Satya Nadella (2 episodes), and 107 more.

Sputnik Intelligence Half price

Follow everything said on Dwarkesh Podcast, with your own AI agent.

Sputnik collects podcasts and newsletters every day. Connect Claude, ChatGPT or Codex and ask it about anyone in them.

Full transcripts, with speaker names
Updated daily
Newsletters too, including email-only ones
Alerts when they get mentioned
Every answer links to the source

Free trial, then ~~$100~~ $49 /month

14 days free. Cancel anytime.

Start free trial

Audience

Apple Podcasts chart: #13 in Technology; Seen 1 Nov 2026
Substack chart: #47 in Technology Bestsellers; Seen 1 Nov 2026
Apple Podcasts genres: Technology, Podcasts, Science

Topics

agi timelines recursive self-improvement ai alignment scaling laws continual learning automated ai research model distillation ai safety model evaluation reward hacking

In These Collections

AI Research & Frontier Labs In-Depth Interviews