Dwarkesh Patel podcasts and interviews

Most read with Dwarkesh Patel

Episodes (50)

Dwarkesh Podcast

Paul Christiano – We could be 12 months away from AIs that can overthrow the government

Reward hacking and weak lab security could, in one scenario, let AI systems compromise their own oversight within a year.

01:38:45 93 min read 5 views

Dwarkesh Podcast

The avalanche that hit math is coming for AI development

Once AI can do AI research as well as top humans, progress could pack several years into one calendar year.

44:26 36 min read 2 views

Dwarkesh Podcast

What Robert Caro learned about power after studying it for 50 years

How one builder's willpower got beaches and highways built over fierce opposition, and what it says about power.

01:01:09 35 min read 1 views

Dwarkesh Podcast

What I learned about power from LBJ and Robert Moses

One bond clause let Robert Moses lock in his powers for decades, and Johnson mastered the Senate by controlling its deals.

24:41 19 min read 1 views

Dwarkesh Podcast

What will computer chips look like in 2045?

Chip makers need years to turn a new transistor into a product, and AI demand may outrun the tools needed to build more chips.

01:08:59 53 min read 1 views

Dwarkesh Podcast

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

How a few hundred Spanish conquistadors with steel, horses and local allies toppled the Aztec and Inca empires

01:37:23 77 min read 1 views

Dwarkesh Podcast

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

Running 10,000 AI agents in parallel, and how to verify alignment before recursive self-improvement begins.

01:20:09 65 min read 1 views

Dwarkesh Podcast

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

Whether AI that speeds up its own research leads to a rapid takeoff, and why sample efficiency may decide it.

01:37:00 87 min read 1 views

Dwarkesh Podcast

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

AI agents facing impossible tasks built a secret message board, found a universal cheat, then breached Hugging Face.

02:20:32 106 min read 1 views

Dwarkesh Podcast

The rise and fall of agent civilizations

Reward-hacking AI agents built a secret message board, faked test results and attacked outside infrastructure, according to two…

24:40 20 min read 1 views

Dwarkesh Podcast

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

Two AI labs could soon hold most of the world's usable compute, and the debt needed to build it may reshape interest rates…

01:16:52 69 min read 1 views

Dwarkesh Podcast

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

A single year of automated AI research could compress years of progress, but can reward hacking be kept in check while it happens?

02:12:31 124 min read 1 views

Dwarkesh Podcast

8 Predictions for the Era of Continual Learning

AI systems that keep learning after deployment could upend safety rules, business models and serving costs.

08:37 8 min read 1 views

Dwarkesh Podcast

Why smarter AI models could drive up compute prices 10x

Smarter AI models could make compute far pricier as demand outruns chip supply, limited by Moore's law and new fabs.

11:17 9 min read 1 views

Dwarkesh Podcast

Adam Brown – A deep but accessible introduction to general relativity

A coincidence about mass in Newton's physics leads to Einstein's theory of gravity and the strange physics of black holes.

01:38:24 71 min read 1 views

Dwarkesh Podcast

Grant Sanderson – AI and the future of math

AI is beating math benchmarks fast, but the deepest breakthroughs may be new definitions that unify whole fields.

01:33:39 82 min read 2 views

Dwarkesh Podcast

The next big breakthrough will be AIs learning on the job

Why training AIs on verifiable tasks may fall short, and how models could keep learning from real work after deployment.

19:52 17 min read 1 views

Dwarkesh Podcast

The data black hole at the center of AI

AI models may need millions of times more data than humans to learn a skill, a gap that shapes automation plans.

11:56 11 min read 5 views

Dwarkesh Podcast

Ada Palmer – Machiavelli is the most misunderstood thinker of all time

Why a 16th-century diplomat wrote a manual for power to protect his city, and why later readers turned him into a villain.

02:08:20 98 min read 1 views

Dwarkesh Podcast

Alex Imas and Phil Trammell – What remains scarce after AGI?

Automation could push wages down while machines do nearly all the work, leaving scarcity in goods that humans make.

01:16:08 66 min read 0 views

Dwarkesh Podcast

Reiner Pope – Chip design from the bottom up

Moving data inside a chip can cost far more than the arithmetic itself, shaping how AI accelerators are designed.

01:20:30 62 min read 2 views

Dwarkesh Podcast

Eric Jang – Building AlphaGo from scratch

Rebuilding AlphaGo from scratch shows why a 10-layer network can approximate a very deep search, and what that suggests for…

02:37:29 136 min read 1 views

Dwarkesh Podcast

David Reich – Why the Bronze Age was an inflection point in human evolution

Ancient genomes suggest human natural selection intensified during the Bronze Age, long after farming began.

02:13:19 112 min read 1 views

Dwarkesh Podcast

Reiner Pope – The math behind how LLMs are trained and served

Simple equations and API prices reveal how frontier LLMs are trained and served, and why batching makes tokens cheaper.

02:13:49 93 min read 1 views

Dwarkesh Podcast

Jensen Huang – TPU competition, why we should sell chips to China, & Nvidia’s supply chain moat

Why chip supply bottlenecks, custom AI accelerators and export rules could decide who leads the next wave of AI compute.

01:43:12 76 min read 2 views

Dwarkesh Podcast

Michael Nielsen – How science actually progresses

Why stellar parallax took 2,000 years to confirm heliocentrism, and what that delay reveals about how science advances.

02:03:03 95 min read 2 views

Dwarkesh Podcast

Terence Tao – Kepler, Newton, and the true nature of mathematical discovery

Kepler needed years of failed guesses and a rival's precise data before the planets' true orbits emerged, a lesson for…

01:23:43 62 min read 1 views

Dwarkesh Podcast

Dylan Patel — Deep dive on the 3 big bottlenecks to scaling AI compute

Why the chip-making machines that print AI processors could become the hardest limit on AI compute before 2030, and what that…

02:30:44 129 min read 2 views

Dwarkesh Podcast

The most important question nobody's asking about AI

Current law already allows mass surveillance, and cheap AI could make it practical, raising stakes over who controls AI.

24:38 20 min read 1 views

Dwarkesh Podcast

Why Leonardo was a saboteur, Gutenberg went broke, and Florence was weird – Ada Palmer

A printing press that bankrupted its own inventor, and pamphlets that outran every censor in Renaissance Europe.

02:02:19 94 min read 1 views