Notes on China

I’m so excited with how this visualization of Notes on China turned out. Petr, thank you for such beautiful watercolor artwork. More to come!Watch on YouTube.Timestamps(0:00:00) - Intro(0:00:32) - Scale(0:05:50) - Vibes(0:11:14) - Youngsters(0:14:27) - Tech & AI(0:15:47) - Hearts & Minds(0:17:07) - On Travel Get full access to Dwarkesh Podcast at www.dwarkesh.com/subscribe

Satya Nadella on:- Why he doesn’t believe in AGI but does believe in 10% economic growth,- Microsoft’s new topological qubit breakthrough and gaming world models,- Whether Office commoditizes LLMs or the other way around,Watch on Youtube; listen on Apple Podcasts or Spotify.SponsorsScale partners with major AI labs like Meta, Google Deepmind, and OpenAI. Through Scale’s Data Foundry, labs get access to high-quality data to fuel post-training, including advanced reasoning capabilities. If you’re an AI researcher or engineer, learn about how Scale’s Data Foundry and research lab, SEAL, can help you go beyond the current frontier at scale.com/dwarkeshLinear's project management tools have become the default choice for product teams at companies like Ramp, CashApp, OpenAI, and Scale. These teams use Linear so they can stay close to their products and move fast. If you’re curious why so many companies are making the switch, visit linear.app/dwarkeshTo sponsor a future episode, visit dwarkeshpatel.com/p/advertise.Timestamps(0:00:00) - Intro(0:05:04) - AI won't be winner-take-all(0:15:18) - World economy growing by 10%(0:21:39) - Decreasing price of intelligence(0:30:19) - Quantum breakthrough(0:42:51) - How Muse will change gaming(0:49:51) - Legal barriers to AI(0:55:46) - Getting AGI safety right(1:04:59) - 34 years at Microsoft(1:10:46) - Does Satya Nadella believe in AGI? Get full access to Dwarkesh Podcast at www.dwarkesh.com/subscribe

This week I welcome on the show two of the most important technologists ever, in any field.Jeff Dean is Google's Chief Scientist, and through 25 years at the company, has worked on basically the most transformative systems in modern computing: from MapReduce, BigTable, Tensorflow, AlphaChip, to Gemini.Noam Shazeer invented or co-invented all the main architectures and techniques that are used for modern LLMs: from the Transformer itself, to Mixture of Experts, to Mesh Tensorflow, to Gemini and many other things.We talk about their 25 years at Google, going from PageRank to MapReduce to the Transformer to MoEs to AlphaChip – and maybe soon to ASI.My favorite part was Jeff's vision for Pathways, Google’s grand plan for a mutually-reinforcing loop of hardware and algorithmic design and for going past autoregression. That culminates in us imagining *all* of Google-the-company, going through one huge MoE model.And Noam just bites every bullet: 100x world GDP soon; let’s get a million automated researchers running in the Google datacenter; living to see the year 3000.Watch on Youtube; listen on Apple Podcasts or Spotify.SponsorsScale partners with major AI labs like Meta, Google Deepmind, and OpenAI. Through Scale’s Data Foundry, labs get access to high-quality data to fuel post-training, including advanced reasoning capabilities. If you’re an AI researcher or engineer, learn about how Scale’s Data Foundry and research lab, SEAL, can help you go beyond the current frontier at scale.com/dwarkeshCurious how Jane Street teaches their new traders? They use Figgie, a rapid-fire card game that simulates the most exciting parts of markets and trading. It’s become so popular that Jane Street hosts an inter-office Figgie championship every year. Download from the app store or play on your desktop at figgie.comMeter wants to radically improve the digital world we take for granted. They’re developing a foundation model that automates network management end-to-end. To do this, they just announced a long-term partnership with Microsoft for tens of thousands of GPUs, and they’re recruiting a world class AI research team. To learn more, go to meter.com/dwarkeshTo sponsor a future episode, visit dwarkeshpatel.com/p/advertiseTimestamps00:00:00 - Intro00:02:44 - Joining Google in 199900:05:36 - Future of Moore's Law00:10:21 - Future TPUs00:13:13 - Jeff’s undergrad thesis: parallel backprop00:15:10 - LLMs in 200700:23:07 - “Holy s**t” moments00:29:46 - AI fulfills Google’s original mission00:34:19 - Doing Search in-context00:38:32 - The internal coding model00:39:49 - What will 2027 models do?00:46:00 - A new architecture every day?00:49:21 - Automated chip design and intelligence explosion00:57:31 - Future of inference scaling01:03:56 - Already doing multi-datacenter runs01:22:33 - Debugging at scale01:26:05 - Fast takeoff and superalignment01:34:40 - A million evil Jeff Deans01:38:16 - Fun times at Google01:41:50 - World compute demand in 203001:48:21 - Getting back to modularity01:59:13 - Keeping a giga-MoE in-memory02:04:09 - All of Google in one model02:12:43 - What’s missing from distillation02:18:03 - Open research, pros and cons02:24:54 - Going the distance Get full access to Dwarkesh Podcast at www.dwarkesh.com/subscribe

Third and final episode in the Paine trilogy!Chinese history is full of warlords constantly challenging the capital. How could Mao not only stay in power for decades, but not even face any insurgency?And how did Mao go from military genius to peacetime disaster - the patriotic hero who inflicted history’s worst human catastrophe on China? How can someone shrewd enough to win a civil war outnumbered 5 to 1 decide "let's have peasants make iron in their backyards" and "let's kill all the birds"?In her lecture and our Q&A, we cover the first nationwide famine in Chinese history; Mao's lasting influence on other insurgents; broken promises to minorities and peasantry; and what Taiwan means.Thanks so much to @Substack for running this in-person event!Note that Sarah is doing an AMA over the next couple days on Youtube; see the pinned comment.Watch on YouTube. Listen on Apple Podcasts, Spotify, or any other podcast platform.SponsorToday’s episode is brought to you by Scale AI. Scale partners with the U.S. government to fuel America’s AI advantage through their data foundry. Scale recently introduced Defense Llama, Scale's latest solution available for military personnel. With Defense Llama, military personnel can harness the power of AI to plan military or intelligence operations and understand adversary vulnerabilities.If you’re interested in learning more on how Scale powers frontier AI capabilities, go to https://scale.com/dwarkesh. Get full access to Dwarkesh Podcast at www.dwarkesh.com/subscribe

This is the second episode in the trilogy of a lectures by Professor Sarah Paine of the Naval War College.In this second episode, Prof Paine dissects the ideas and economics behind Japanese imperialism before and during WWII. We get into the oil shortage which caused the war; the unique culture of honor and death; the surprisingly chaotic chain of command. This is followed by a Q&A with me.Huge thanks to Substack for hosting this event!Watch on YouTube. Listen on Apple Podcasts, Spotify, or any other podcast platform.SponsorToday’s episode is brought to you by Scale AI. Scale partners with the U.S. government to fuel America’s AI advantage through their data foundry. Scale recently introduced Defense Llama, Scale's latest solution available for military personnel. With Defense Llama, military personnel can harness the power of AI to plan military or intelligence operations and understand adversary vulnerabilities.If you’re interested in learning more on how Scale powers frontier AI capabilities, go to scale.com/dwarkesh.Buy Sarah's Books!I highly, highly recommend both "The Wars for Asia, 1911–1949" and "The Japanese Empire: Grand Strategy from the Meiji Restoration to the Pacific War".Timestamps(0:00:00) - Lecture begins(0:06:58) - The code of the samurai(0:10:45) - Buddhism, Shinto, Confucianism(0:16:52) - Bushido as bad strategy(0:23:34) - Military theorists(0:33:42) - Strategic sins of omission(0:38:10) - Crippled logistics(0:40:58) - the Kwantung Army(0:43:31) - Inter-service communication(0:51:15) - Shattering Japanese morale(0:57:35) - Q&A begins(01:05:02) - Unusual brutality of WWII(01:11:30) - Embargo caused the war(01:16:48) - The liberation of China(01:22:02) - Could US have prevented war?(01:25:30) - Counterfactuals in history(01:27:46) - Japanese optimism(01:30:46) - Tech change and social change(01:38:22) - Hamming questions(01:44:31) - Do sanctions work?(01:50:07) - Backloaded mass death(01:54:09) - demilitarizing Japan(01:57:30) - Post-war alliances(02:03:46) - Inter-service rivalry Get full access to Dwarkesh Podcast at www.dwarkesh.com/subscribe

I’m thrilled to launch a new trilogy of double episodes: a lecture series by Professor Sarah Paine of the Naval War College, each followed by a deep Q&A.In this first episode, Prof Paine talks about key decisions by Khrushchev, Mao, Nehru, Bhutto, & Lyndon Johnson that shaped the whole dynamic of South Asia today. This is followed by a Q&A.Come for the spy bases, shoestring nukes, and insight about how great power politics impacts every region.Huge thanks to Substack for hosting this!Watch on YouTube. Listen on Apple Podcasts, Spotify, or any other podcast platform.SponsorsToday’s episode is brought to you by Scale AI. Scale partners with the U.S. government to fuel America’s AI advantage through their data foundry. The Air Force, Army, Defense Innovation Unit, and Chief Digital and Artificial Intelligence Office all trust Scale to equip their teams with AI-ready data and the technology to build powerful applications.Scale recently introduced Defense Llama, Scale's latest solution available for military personnel. With Defense Llama, military personnel can harness the power of AI to plan military or intelligence operations and understand adversary vulnerabilities.If you’re interested in learning more on how Scale powers frontier AI capabilities, go to scale.com/dwarkesh.Timestamps(00:00) - Intro(02:11) - Mao at war, 1949-51(05:40) - Pactomania and Sino-Soviet conflicts(14:42) - The Sino-Indian War(20:00) - Soviet peace in India-Pakistan(22:00) - US Aid and Alliances(26:14) - The difference with WWII(30:09) - The geopolitical map in 1904(35:10) - The US alienates Indira Gandhi(42:58) - Instruments of US power(53:41) - Carrier battle groups(1:02:41) - Q&A begins(1:04:31) - The appeal of the USSR(1:09:36) - The last communist premier(1:15:42) - India and China's lost opportunity(1:58:04) - Bismark's cunning(2:03:05) - Training US officers(2:07:03) - Cruelty in Russian history Get full access to Dwarkesh Podcast at www.dwarkesh.com/subscribe

I interviewed Tyler Cowen at the Progress Conference 2024. As always, I had a blast. This is my fourth interview with him – and yet I’m always hearing new stuff.We talked about why he thinks AI won't drive explosive economic growth, the real bottlenecks on world progress, him now writing for AIs instead of humans, and the difficult relationship between being cultured and fostering growth – among many other things in the full episode.Thanks to the Roots of Progress Institute (with special thanks to Jason Crawford and Heike Larson) for such a wonderful conference, and to FreeThink for the videography.Watch on YouTube. Listen on Apple Podcasts, Spotify, or any other podcast platform. Read the full transcript here.SponsorsI’m grateful to Tyler for volunteering to say a few words about Jane Street. It's the first time that a guest has participated in the sponsorship. I hope you can see why Tyler and I think so highly of Jane Street. To learn more about their open rules, go to janestreet.com/dwarkesh.Timestamps(00:00:00) Economic Growth and AI(00:14:57) Founder Mode and increasing variance(00:29:31) Effective Altruism and Progress Studies(00:33:05) What AI changes for Tyler(00:44:57) The slow diffusion of innovation(00:49:53) Stalin's library(00:52:19) DC vs SF vs EU Get full access to Dwarkesh Podcast at www.dwarkesh.com/subscribe

Adam Brown is a founder and lead of BlueShift with is cracking maths and reasoning at Google DeepMind and a theoretical physicist at Stanford.We discuss: destroying the light cone with vacuum decay, holographic principle, mining black holes, & what it would take to train LLMs that can make Einstein level conceptual breakthroughs.Stupefying, entertaining, & terrifying.Enjoy!Watch on YouTube, read the transcript, listen on Apple Podcasts, Spotify, or your favorite platform.Sponsors- Deepmind, Meta, Anthropic, and OpenAI, partner with Scale for high quality data to fuel post-training Publicly available data is running out - to keep developing smarter and smarter models, labs will need to rely on Scale’s data foundry, which combines subject matter experts with AI models to generate fresh data and break through the data wall. Learn more at scale.ai/dwarkesh.- Jane Street is looking to hire their next generation of leaders. Their deep learning team is looking for ML researchers, FPGA programmers, and CUDA programmers. Summer internships are open for just a few more weeks. If you want to stand out, take a crack at their new Kaggle competition. To learn more, go to janestreet.com/dwarkesh.- This episode is brought to you by Stripe, financial infrastructure for the internet. Millions of companies from Anthropic to Amazon use Stripe to accept payments, automate financial processes and grow their revenue.Timestamps(00:00:00) - Changing the laws of physics(00:26:05) - Why is our universe the way it is(00:37:30) - Making Einstein level AGI(01:00:31) - Physics stagnation and particle colliders(01:11:10) - Hitchhiking(01:29:00) - Nagasaki(01:36:19) - Adam’s career(01:43:25) - Mining black holes(01:59:42) - The holographic principle(02:23:25) - Philosophy of infinities(02:31:42) - Engineering constraints for future civilizations Get full access to Dwarkesh Podcast at www.dwarkesh.com/subscribe

Gwern is a pseudonymous researcher and writer. He was one of the first people to see LLM scaling coming. If you've read his blog, you know he's one of the most interesting polymathic thinkers alive.In order to protect Gwern's anonymity, I proposed interviewing him in person, and having my friend Chris Painter voice over his words after. This amused him enough that he agreed.After the episode, I convinced Gwern to create a donation page where people can help sustain what he's up to. Please go here to contribute.Read the full transcript here.Sponsors:* Jane Street is looking to hire their next generation of leaders. Their deep learning team is looking for ML researchers, FPGA programmers, and CUDA programmers. Summer internships are open - if you want to stand out, take a crack at their new Kaggle competition. To learn more, go to janestreet.com/dwarkesh.* Turing provides complete post-training services for leading AI labs like OpenAI, Anthropic, Meta, and Gemini. They specialize in model evaluation, SFT, RLHF, and DPO to enhance models’ reasoning, coding, and multimodal capabilities. Learn more at turing.com/dwarkesh.* This episode is brought to you by Stripe, financial infrastructure for the internet. Millions of companies from Anthropic to Amazon use Stripe to accept payments, automate financial processes and grow their revenue.If you’re interested in advertising on the podcast, check out this page.Timestamps00:00:00 - Anonymity00:01:09 - Automating Steve Jobs00:04:38 - Isaac Newton's theory of progress00:06:36 - Grand theory of intelligence00:10:39 - Seeing scaling early00:21:04 - AGI Timelines00:22:54 - What to do in remaining 3 years until AGI00:26:29 - Influencing the shoggoth with writing00:30:50 - Human vs artificial intelligence00:33:52 - Rabbit holes00:38:48 - Hearing impairment00:43:00 - Wikipedia editing00:47:43 - Gwern.net00:50:20 - Counterfactual careers00:54:30 - Borges & literature01:01:32 - Gwern's intelligence and process01:11:03 - A day in the life of Gwern01:19:16 - Gwern's finances01:25:05 - The diversity of AI minds01:27:24 - GLP drugs and obesity01:31:08 - Drug experimentation01:33:40 - Parasocial relationships01:35:23 - Open rabbit holes Get full access to Dwarkesh Podcast at www.dwarkesh.com/subscribe

A bonanza on the semiconductor industry and hardware scaling to AGI by the end of the decade.Dylan Patel runs Semianalysis, the leading publication and research firm on AI hardware. Jon Y runs Asianometry, the world’s best YouTube channel on semiconductors and business history.* What Xi would do if he became scaling pilled* $ 1T+ in datacenter buildout by end of decadeWatch on YouTube. Listen on Apple Podcasts, Spotify, or any other podcast platform. Read the full transcript here. Follow me on Twitter for updates on future episodes.Sponsors:* Jane Street is looking to hire their next generation of leaders. Their deep learning team is looking for FPGA programmers, CUDA programmers, and ML researchers. To learn more about their full time roles, internship, tech podcast, and upcoming Kaggle competition, go here.* This episode is brought to you by Stripe, financial infrastructure for the internet. Millions of companies from Anthropic to Amazon use Stripe to accept payments, automate financial processes and grow their revenue.If you’re interested in advertising on the podcast, check out this page.Timestamps00:00:00 – Xi's path to AGI00:04:20 – Liang Mong Song00:08:25 – How semiconductors get better00:11:16 – China can centralize compute00:18:50 – Export controls & sanctions00:32:51 – Huawei's intense culture00:38:51 – Why the semiconductor industry is so stratified00:40:58 – N2 should not exist00:45:53 – Taiwan invasion hypothetical00:49:21 – Mind-boggling complexity of semiconductors00:59:13 – Chip architecture design01:04:36 – Architectures lead to different AI models? China vs. US01:10:12 – Being head of compute at an AI lab01:16:24 – Scaling costs and power demand01:37:05 – Are we financing an AI bubble?01:50:20 – Starting Asianometry and SemiAnalysis02:06:10 – Opportunities in the semiconductor stack Get full access to Dwarkesh Podcast at www.dwarkesh.com/subscribe