On this episode, I chat with Victor Shih about all things China. We discuss China’s massive local debt crisis, the CCP’s views on AI, what happens after Xi, and more.

Victor Shih is an expert on the Chinese political system, as well as their banking and fiscal policies, and he has amassed more biographical data on the Chinese elite than anyone else in the world. He teaches at UC San Diego, where he also directs the 21st Century China Center.

Watch on YouTube; listen on Apple Podcasts or Spotify.

Sponsors

* Scale is building the infrastructure for smarter, safer AI. In addition to their Data Foundry, they just released Scale Evaluation, a tool that diagnoses model limitations. Learn how Scale can help you push the frontier at scale.com/dwarkesh.

* WorkOS is how top AI companies ship critical enterprise features without burning months of engineering time. If you need features like SSO, audit logs, or user provisioning, head to workos.com.

To sponsor a future episode, visit dwarkesh.com/advertise.

Timestamps

(00:00:00) – Is China more decentralized than the US?

(00:03:16) – How the Politburo Standing Committee makes decisions

(00:21:07) – Xi’s right hand man in charge of AGI

(00:35:37) – DeepSeek was trained to track CCP policy

(00:39:26) – Xi versus Stalin

(00:45:35) – Local government debt crisis

(00:50:00) – Financial repression & capital controls

(00:58:12) – How corruption leads to overbuilding

(01:06:58) – How China can solve its debt problem

(01:10:46) – Probability of Taiwan invasion

(01:13:57) – Is Xi getting bad intel?

(01:18:56) – What happens after Xi

(01:25:10) – Future growth forecasts

New episode with my good friends Sholto Douglas & Trenton Bricken. Sholto focuses on scaling RL and Trenton researches mechanistic interpretability, both at Anthropic.

We talk through what’s changed in the last year of AI research; the new RL regime and how far it can scale; how to trace a model’s thoughts; and how countries, workers, and students should prepare for AGI.

See you next year for v3. Here’s last year’s episode, btw. Enjoy!

Watch on YouTube; listen on Apple Podcasts or Spotify.

----------

SPONSORS

* WorkOS ensures that AI companies like OpenAI and Anthropic don't have to spend engineering time building enterprise features like access controls or SSO. It’s not that they don't need these features; it's just that WorkOS gives them battle-tested APIs that they can use for auth, provisioning, and more. Start building today at workos.com.

* Scale is building the infrastructure for safer, smarter AI. Scale’s Data Foundry gives major AI labs access to high-quality data to fuel post-training, while their public leaderboards help assess model capabilities. They also just released Scale Evaluation, a new tool that diagnoses model limitations. If you’re an AI researcher or engineer, learn how Scale can help you push the frontier at scale.com/dwarkesh.

* Lighthouse is THE fastest immigration solution for the technology industry. They specialize in expert visas like the O-1A and EB-1A, and they’ve already helped companies like Cursor, Notion, and Replit navigate U.S. immigration. Explore which visa is right for you at lighthousehq.com/ref/Dwarkesh.

To sponsor a future episode, visit dwarkesh.com/advertise.

----------

TIMESTAMPS

(00:00:00) – How far can RL scale?

(00:16:27) – Is continual learning a key bottleneck?

(00:31:59) – Model self-awareness

(00:50:32) – Taste and slop

(01:00:51) – How soon to fully autonomous agents?

(01:15:17) – Neuralese

(01:18:55) – Inference compute will bottleneck AGI

(01:23:01) – DeepSeek algorithmic improvements

(01:37:42) – Why are LLMs ‘baby AGI’ but not AlphaZero?

(01:45:38) – Mech interp

(01:56:15) – How countries should prepare for AGI

(02:10:26) – Automating white collar work

(02:15:35) – Advice for students

Based on my essay about AI firms.

Huge thanks to Petr and his team for bringing this to life!

Watch on YouTube.

Thanks to Google for sponsoring. We used their Veo 2 model to make this entire video—it generated everything from the photorealistic humans to the claymation octopuses. If you’re a Gemini Advanced user, you can try Veo 2 now in the Gemini app. Just select Veo 2 in the dropdown, and type your video idea in the prompt bar. Get started today by going to gemini.google.com.

To sponsor a future episode, visit dwarkesh.com/advertise.