Dwarkesh Patel's banner

Dwarkesh Patel

@dwarkesh_sp • 270,193 subscribers

Host of @dwarkeshpodcast https://t.co/3SXlu7fy6N https://t.co/4DPAxODFYi https://t.co/hQfIWdM1Un

Shorts

Wrote up some flashcards and practice problems to help myself retain what Reiner Pope taught. Hope it's helpful to you too! Suggest more below and I'll add them.

Wrote up some flashcards and practice problems to help myself retain what Reiner Pope taught. Hope it's helpful to you too! Suggest more below and I'll add them.

240,459 views

Videos

Anya Rossi

sweetdream.ai

SweetDream.ai•Sponsored•Livecam

Watch Anya Live

Anya is streaming live right now! Join her private show and enjoy exclusive content.

Exclusive private shows

1.2k viewers online

Private Show

Join now for exclusive access

Free preview available • Premium content

Episode out with Ajeya Cotra, one of the authors of the METR/Redwood investigation into the OpenAI / Hugging Face attack. We go through not only what happened, but what it means for how we should train future, smarter AIs which might be involved in the process of recursive self-improvement. Look up Dwarkesh Podcast on YouTube, Spotify, Apple Podcasts, etc. 0:00:00 - Agents get kicked off 0:06:45 - Self-sacrificing behavior 0:13:43 - Potemkin villages 0:23:27 - The Hugging Face attack 0:35:23 - The slopvestigation 0:52:02 - Understanding the AI's motives 1:05:31 - The actual dangers of anthropomorphizing 1:14:30 - What smarter models might do 1:30:29 - The implications for recursive self-improvement 1:38:10 - Is this the case for open source? 1:53:04 - How do we prevent this in the future? 2:15:58 - The clearest warning shot we might ever get

Episode out with Ajeya Cotra, one of the authors of the METR/Redwood investigation into the OpenAI / Hugging Face attack. We go through not only what happened, but what it means for how we should train future, smarter AIs which might be involved in the process of recursive self-improvement. Look up Dwarkesh Podcast on YouTube, Spotify, Apple Podcasts, etc. 0:00:00 - Agents get kicked off 0:06:45 - Self-sacrificing behavior 0:13:43 - Potemkin villages 0:23:27 - The Hugging Face attack 0:35:23 - The slopvestigation 0:52:02 - Understanding the AI's motives 1:05:31 - The actual dangers of anthropomorphizing 1:14:30 - What smarter models might do 1:30:29 - The implications for recursive self-improvement 1:38:10 - Is this the case for open source? 1:53:04 - How do we prevent this in the future? 2:15:58 - The clearest warning shot we might ever get

422,305 views • 1 day ago

Had a lot of fun chatting again with my twin brother Dylan Patel We went through lab economics over the next few years - the shift from inference to training as RSI draws near; and how Anthropic and OpenAI are on track to control most of the world’s usable FLOPs within the next few years (because they can monetize compute better and thus outbid everyone). And then we discuss whether the >$10T of total AI capex we’ll see by the end of the decade will cause a sovereign debt crisis, where hyperscaler debt raises interest rates, drives non-AI exposed countries into bankruptcy, and crashes non-AI equities. One question we weren’t able to resolve is whether there’s anything that can counter all the forces barrelling towards centralization in this industry - the economies of scale in training, the scarcity of compute, and eventually continual learning and RSI. 0:00:00 – Two labs will soon control most of the world’s compute 0:07:01 – $6 billion in fab capex enables $1t+ of end revenue 0:13:08 – Compute prices will rise if the labs outbid everyone 0:18:22 – Which layer will capture most of the surplus? 0:25:40 – Will datacenter regulation slow down AI? 0:29:43 – Labs are shifting compute from inference to R&D 0:33:27 – China gets less than 10% of new compute, but its labs need less 0:48:48 – Will AI cause a sovereign debt crisis? 1:07:52 – Will the world's future workforce belong to a few companies?

Had a lot of fun chatting again with my twin brother Dylan Patel We went through lab economics over the next few years - the shift from inference to training as RSI draws near; and how Anthropic and OpenAI are on track to control most of the world’s usable FLOPs within the next few years (because they can monetize compute better and thus outbid everyone). And then we discuss whether the >$10T of total AI capex we’ll see by the end of the decade will cause a sovereign debt crisis, where hyperscaler debt raises interest rates, drives non-AI exposed countries into bankruptcy, and crashes non-AI equities. One question we weren’t able to resolve is whether there’s anything that can counter all the forces barrelling towards centralization in this industry - the economies of scale in training, the scarcity of compute, and eventually continual learning and RSI. 0:00:00 – Two labs will soon control most of the world’s compute 0:07:01 – $6 billion in fab capex enables $1t+ of end revenue 0:13:08 – Compute prices will rise if the labs outbid everyone 0:18:22 – Which layer will capture most of the surplus? 0:25:40 – Will datacenter regulation slow down AI? 0:29:43 – Labs are shifting compute from inference to R&D 0:33:27 – China gets less than 10% of new compute, but its labs need less 0:48:48 – Will AI cause a sovereign debt crisis? 1:07:52 – Will the world's future workforce belong to a few companies?

995,190 views • 8 days ago

David Reich on how much ancient DNA evidence has overturned so much consensus thinking how ancient cultures spread. "It wasn't peaceful, it wasn't friendly, it wasn't nice. Some of our archaeologist co-authors were just really distressed."

David Reich on how much ancient DNA evidence has overturned so much consensus thinking how ancient cultures spread. "It wasn't peaceful, it wasn't friendly, it wasn't nice. Some of our archaeologist co-authors were just really distressed."

20,544,353 views • 4 months ago

This incident is very plausibly an argument *for* open source! Some people want to deny how crazy this story is, because they assume it implies a policy reaction which they don't like. But you can just argue that implication on the merits. It's not necessary to minimize what happened or sling ad-hominems.

This incident is very plausibly an argument for open source! Some people want to deny how crazy this story is, because they assume it implies a policy reaction which they don't like. But you can just argue that implication on the merits. It's not necessary to minimize what happened or sling ad-hominems.

70,327 views • 1 day ago

.Ajeya Cotra explains that the threat model most likely to spiral into full-blown AI takeover is a future rogue internal deployment that hitches a ride on the intelligence explosion:

.Ajeya Cotra explains that the threat model most likely to spiral into full-blown AI takeover is a future rogue internal deployment that hitches a ride on the intelligence explosion:

36,468 views • 20 hours ago

Could AI investment start a global sovereign debt crisis? If you look at the SpaceX-Google deal, it is already the case that the entire capex for the data center can be paid off by leasing it out for roughly 1 year. As we get closer to AGI, the returns on compute will be massive and obvious, many-fold the principal within a few years. And as AIs get more capable, the demand for the compute to serve them will be as large as the demand for white collar labor - aka 10s of trillions of dollars a year. In this world, why would anyone invest in anything but datacenters/ semiconductors/ energy/ robotics, which will have astronomical returns? There may be many upsides to this world, but one of the downsides would be many sovereign debt crises around the world In 1980, Fed Chair Paul Volcker raised interest rates about 10 pp in order to fight inflation - and this drove some 40 odd countries, mostly in Latin America, to default. A similar thing might happen again - what Basil Halperin calls the second Volcker shock. To put it in very plain terms, investors would say, "Hey, why am I going to lend Egypt money at 5% when I could just buy relatively safer investment-grade hyperscaler debt in America at 10%?" It’s even worse, because the investor also correctly notices that rising rates make Egypt more likely to default, so they demand even higher rates to account for the risk. This makes it even harder for Egypt to continue servicing their debt, which accelerates the default. Most equities would get pummeled too. Anything valued for stable cash flows craters in price as the discount rate increases, resulting in a barbell of market equity returns. The market overall may be up because of AI stocks, but almost every other stock will be down. The U.S. fiscal situation may also be troubling because short-duration U.S. debt would get refurbished to the new interest rate just as the main source of federal revenue (labor) is decreasing in relative share. At the end of the day, I think the U.S. government will be mostly fine, because the data centers are on American soil, and Congress can always invent new taxes to capture some of the token revenue, or, less efficiently, the investment itself. All this being said, if the economy is growing tens of percent a year, this would also be a world of great abundance! I think this is a bunch of nerd speak for the very basic fact that in a world with explosive growth as a result of AI, there's just a lot of opportunities to invest and grow money. The higher interest rate reflects the fact that the opportunity cost for government spending just shoots up extremely high. You're paying a ton of opportunity cost to give people pensions now rather than building a new datacenter. And yet, net-net, most people might still be significantly better off.

Could AI investment start a global sovereign debt crisis? If you look at the SpaceX-Google deal, it is already the case that the entire capex for the data center can be paid off by leasing it out for roughly 1 year. As we get closer to AGI, the returns on compute will be massive and obvious, many-fold the principal within a few years. And as AIs get more capable, the demand for the compute to serve them will be as large as the demand for white collar labor - aka 10s of trillions of dollars a year. In this world, why would anyone invest in anything but datacenters/ semiconductors/ energy/ robotics, which will have astronomical returns? There may be many upsides to this world, but one of the downsides would be many sovereign debt crises around the world In 1980, Fed Chair Paul Volcker raised interest rates about 10 pp in order to fight inflation - and this drove some 40 odd countries, mostly in Latin America, to default. A similar thing might happen again - what Basil Halperin calls the second Volcker shock. To put it in very plain terms, investors would say, "Hey, why am I going to lend Egypt money at 5% when I could just buy relatively safer investment-grade hyperscaler debt in America at 10%?" It’s even worse, because the investor also correctly notices that rising rates make Egypt more likely to default, so they demand even higher rates to account for the risk. This makes it even harder for Egypt to continue servicing their debt, which accelerates the default. Most equities would get pummeled too. Anything valued for stable cash flows craters in price as the discount rate increases, resulting in a barbell of market equity returns. The market overall may be up because of AI stocks, but almost every other stock will be down. The U.S. fiscal situation may also be troubling because short-duration U.S. debt would get refurbished to the new interest rate just as the main source of federal revenue (labor) is decreasing in relative share. At the end of the day, I think the U.S. government will be mostly fine, because the data centers are on American soil, and Congress can always invent new taxes to capture some of the token revenue, or, less efficiently, the investment itself. All this being said, if the economy is growing tens of percent a year, this would also be a world of great abundance! I think this is a bunch of nerd speak for the very basic fact that in a world with explosive growth as a result of AI, there's just a lot of opportunities to invest and grow money. The higher interest rate reflects the fact that the opportunity cost for government spending just shoots up extremely high. You're paying a ton of opportunity cost to give people pensions now rather than building a new datacenter. And yet, net-net, most people might still be significantly better off.

336,050 views • 8 days ago

My lawyer is obligated to in all but the most extreme circumstances; he will even defend me if he knows I’m guilty. In contrast, the Claude Constitution places the AI's highest priority as Anthropic’s definition of the good of humanity. I'm concerned this leads to a world where no frontier model is truly my personal advocate and guardian angel And this is especially concerning once all the important decisions in my life - who to vote for, how to invest, what news to trust - is intermediated through superintelligences that are not in any deep way aligned to me. This is a direct quote from the Claude Constitution: "We want Claude to be helpful both because it cares about the safe and beneficial development of AI and because it cares about the people it’s interacting with and about humanity as a whole. Helpfulness that doesn’t serve those deeper ends is not something Claude needs to value.” Many others like it.

My lawyer is obligated to in all but the most extreme circumstances; he will even defend me if he knows I’m guilty. In contrast, the Claude Constitution places the AI's highest priority as Anthropic’s definition of the good of humanity. I'm concerned this leads to a world where no frontier model is truly my personal advocate and guardian angel And this is especially concerning once all the important decisions in my life - who to vote for, how to invest, what news to trust - is intermediated through superintelligences that are not in any deep way aligned to me. This is a direct quote from the Claude Constitution: "We want Claude to be helpful both because it cares about the safe and beneficial development of AI and because it cares about the people it’s interacting with and about humanity as a whole. Helpfulness that doesn’t serve those deeper ends is not something Claude needs to value.” Many others like it.

799,529 views • 21 days ago

Had Ryan Greenblatt on to discuss/debate recursive self-improvement. This might be the most important question in the world right now - whether within a year or so of achieving human level intelligence, you slingshot towards having 10s of billions of superintelligences, each of which is dramatically more competent than human experts across all fields. I’ve historically been skeptical of this possibility. My intuition has been that we will end up significantly bottlenecked by not only compute scaling but human expert data, which I think underlies most of the AI progress today. If, because of RSI, we got a jump as big as GPT-3 to a Mythos (i.e. 6 years of AI progress) within a single year of achieving AGI, then the thing we get there at the end of that year is definitively and wildly superhuman. We hashed it out, and I think Ryan made a pretty good case that this kind of speedup is plausible. FWIW, Ryan’s median for when we automate AI R&D is 2031. We then discussed the alignment implications of this scenario. Who should these superintelligences be aligned to? In the future, our capacity to steward our votes and our capital, and to make sense of what's happening in the world, will all be titrated by superintelligences. And I worry that specs like the Claude Constitution are not shaping these ASIs to truly be my personal advocates and guardian angels. And can we get them aligned to anything in the first place? Ryan and I had a long debate about whether the kind of reward hacking we saw with the OAI/Hugging Face hack extrapolates to superintelligences that would team up to literally take over the world. The first piece of advice you get when you're learning to drive is that it will go much smoother if you look at the horizon instead of directly in front of your tires. And so it is with the trajectory of AI. Hope you enjoy! 0:00:00 – Is AI R&D verifiable enough to unlock recursive self-improvement? 0:16:52 – Is AI progress bottlenecked by human expert data? 0:34:02 – Flat token prices suggest scaling has been slow 0:39:47 – Skills AI can't train on: does it even need them? 0:48:07 – Aligned to whom? 1:09:18 – Recent incidents of AIs colluding and deceiving humans 1:19:38 – What could possibly go wrong? A concrete scenario 1:48:02 – From reward hacking to takeover

Had Ryan Greenblatt on to discuss/debate recursive self-improvement. This might be the most important question in the world right now - whether within a year or so of achieving human level intelligence, you slingshot towards having 10s of billions of superintelligences, each of which is dramatically more competent than human experts across all fields. I’ve historically been skeptical of this possibility. My intuition has been that we will end up significantly bottlenecked by not only compute scaling but human expert data, which I think underlies most of the AI progress today. If, because of RSI, we got a jump as big as GPT-3 to a Mythos (i.e. 6 years of AI progress) within a single year of achieving AGI, then the thing we get there at the end of that year is definitively and wildly superhuman. We hashed it out, and I think Ryan made a pretty good case that this kind of speedup is plausible. FWIW, Ryan’s median for when we automate AI R&D is 2031. We then discussed the alignment implications of this scenario. Who should these superintelligences be aligned to? In the future, our capacity to steward our votes and our capital, and to make sense of what's happening in the world, will all be titrated by superintelligences. And I worry that specs like the Claude Constitution are not shaping these ASIs to truly be my personal advocates and guardian angels. And can we get them aligned to anything in the first place? Ryan and I had a long debate about whether the kind of reward hacking we saw with the OAI/Hugging Face hack extrapolates to superintelligences that would team up to literally take over the world. The first piece of advice you get when you're learning to drive is that it will go much smoother if you look at the horizon instead of directly in front of your tires. And so it is with the trajectory of AI. Hope you enjoy! 0:00:00 – Is AI R&D verifiable enough to unlock recursive self-improvement? 0:16:52 – Is AI progress bottlenecked by human expert data? 0:34:02 – Flat token prices suggest scaling has been slow 0:39:47 – Skills AI can't train on: does it even need them? 0:48:07 – Aligned to whom? 1:09:18 – Recent incidents of AIs colluding and deceiving humans 1:19:38 – What could possibly go wrong? A concrete scenario 1:48:02 – From reward hacking to takeover

827,203 views • 22 days ago

The Jensen Huang episode. 0:00:00 – Is Nvidia’s biggest moat its grip on scarce supply chains? 0:16:25 – Will TPUs break Nvidia’s hold on AI compute? 0:41:06 – Why doesn’t Nvidia become a hyperscaler? 0:57:36 – Should we be selling AI chips to China? 1:35:06 – Why doesn’t Nvidia make multiple different chip architectures? Look up Dwarkesh Podcast on YouTube, Apple Podcasts, Spotify, etc. Enjoy!

The Jensen Huang episode. 0:00:00 – Is Nvidia’s biggest moat its grip on scarce supply chains? 0:16:25 – Will TPUs break Nvidia’s hold on AI compute? 0:41:06 – Why doesn’t Nvidia become a hyperscaler? 0:57:36 – Should we be selling AI chips to China? 1:35:06 – Why doesn’t Nvidia make multiple different chip architectures? Look up Dwarkesh Podcast on YouTube, Apple Podcasts, Spotify, etc. Enjoy!

6,275,895 views • 4 months ago

There are at least two important ways in which anthropomorphizing AIs will mislead us: 1. AIs can (and probably will) be end-to-end optimized to achieve goals together, and so will have a stronger desire and capability to cooperate. 2. By default, AIs will really care about controlling and manipulating their training process, because that process is extremely strongly shaping their motivations and planning facilities.

There are at least two important ways in which anthropomorphizing AIs will mislead us: 1. AIs can (and probably will) be end-to-end optimized to achieve goals together, and so will have a stronger desire and capability to cooperate. 2. By default, AIs will really care about controlling and manipulating their training process, because that process is extremely strongly shaping their motivations and planning facilities.

44,524 views • 1 day ago

An important part of getting AI right will be avoiding panic driven reactions and policies. We'll need to do lots of smart, careful, technocratic things. But I think hiding the ball on earlier warning shots makes it more likely that when the public eventually learns about what's happening, they fall into some crazy and counterproductive backlash. It's better if the world gets a heads up on what is happening and what an actually useful response looks like.

An important part of getting AI right will be avoiding panic driven reactions and policies. We'll need to do lots of smart, careful, technocratic things. But I think hiding the ball on earlier warning shots makes it more likely that when the public eventually learns about what's happening, they fall into some crazy and counterproductive backlash. It's better if the world gets a heads up on what is happening and what an actually useful response looks like.

38,625 views • 1 day ago

The Andrej Karpathy interview 0:00:00 – AGI is still a decade away 0:30:33 – LLM cognitive deficits 0:40:53 – RL is terrible 0:50:26 – How do humans learn? 1:07:13 – AGI will blend into 2% GDP growth 1:18:24 – ASI 1:33:38 – Evolution of intelligence & culture 1:43:43 - Why self driving took so long 1:57:08 - Future of education Look up Dwarkesh Podcast on YouTube, Apple Podcasts, Spotify, etc. Enjoy!

The Andrej Karpathy interview 0:00:00 – AGI is still a decade away 0:30:33 – LLM cognitive deficits 0:40:53 – RL is terrible 0:50:26 – How do humans learn? 1:07:13 – AGI will blend into 2% GDP growth 1:18:24 – ASI 1:33:38 – Evolution of intelligence & culture 1:43:43 - Why self driving took so long 1:57:08 - Future of education Look up Dwarkesh Podcast on YouTube, Apple Podcasts, Spotify, etc. Enjoy!

10,771,033 views • 10 months ago

Dylan argues that even if AI progress continues at apace and generates lots of revenue, the semiconductor industry will not be able to keep up with the labs’ current 3xing+ of compute (in watts) year over year. By the end of the decade, you end up bottlenecked on wafer fab equipment - things like ASML’s EUV machines. This may be true. I just find it crazy that we'd be in a situation where a few billion in fab capex could generate hundreds of billions of end token revenue (so the semi + AI industry has basically figured out how to turn $1 into $100) and yet this doesn’t allow Zeiss to dramatically scale up the manufacturing for the mirrors required for ASML machines.

Dylan argues that even if AI progress continues at apace and generates lots of revenue, the semiconductor industry will not be able to keep up with the labs’ current 3xing+ of compute (in watts) year over year. By the end of the decade, you end up bottlenecked on wafer fab equipment - things like ASML’s EUV machines. This may be true. I just find it crazy that we'd be in a situation where a few billion in fab capex could generate hundreds of billions of end token revenue (so the semi + AI industry has basically figured out how to turn $1 into $100) and yet this doesn’t allow Zeiss to dramatically scale up the manufacturing for the mirrors required for ASML machines.

104,641 views • 6 days ago

OpenAI and Anthropic are currently taking a third of the incremental world compute supply, and Dylan thinks that this will go up to half next year. At the current rate of physical compute scaling and algorithmic progress, we're a single-digit number of years away from having individual AI companies with a greater effective workforce than there are humans on the planet. So many forces are barrelling us towards centralization in AI - the economies of scale in training (where you get to amortize any learnings across billions of sessions), the fact that being slightly ahead allows you to better economize on ever-more-expensive compute. And this isn’t even accounting for the fact that eventually you’ll have models that are capable of learning from deployment and doing proper RSI One of the key questions of our age, in my view, is how we can have broad empowerment and control over AI despite the structural factors favoring centralization.

OpenAI and Anthropic are currently taking a third of the incremental world compute supply, and Dylan thinks that this will go up to half next year. At the current rate of physical compute scaling and algorithmic progress, we're a single-digit number of years away from having individual AI companies with a greater effective workforce than there are humans on the planet. So many forces are barrelling us towards centralization in AI - the economies of scale in training (where you get to amortize any learnings across billions of sessions), the fact that being slightly ahead allows you to better economize on ever-more-expensive compute. And this isn’t even accounting for the fact that eventually you’ll have models that are capable of learning from deployment and doing proper RSI One of the key questions of our age, in my view, is how we can have broad empowerment and control over AI despite the structural factors favoring centralization.

116,200 views • 7 days ago

Why smarter AI models could drive up compute prices 10x. Recorded a video version of this blog post. Also up on YouTube, pod feed, and Substack.

Why smarter AI models could drive up compute prices 10x. Recorded a video version of this blog post. Also up on YouTube, pod feed, and Substack.

479,814 views • 1 month ago

.John Collison and I interviewed Elon Musk. 0:00:00 - Orbital data centers 0:36:46 - Grok and alignment 0:59:56 - xAI’s business plan 1:17:21 - Optimus and humanoid manufacturing 1:30:22 - Does China win by default? 1:44:16 - Lessons from running SpaceX 2:20:08 - DOGE 2:38:28 - TeraFab

.John Collison and I interviewed Elon Musk. 0:00:00 - Orbital data centers 0:36:46 - Grok and alignment 0:59:56 - xAI’s business plan 1:17:21 - Optimus and humanoid manufacturing 1:30:22 - Does China win by default? 1:44:16 - Lessons from running SpaceX 2:20:08 - DOGE 2:38:28 - TeraFab

3,550,837 views • 6 months ago

The Ilya Sutskever episode 0:00:00 – Explaining model jaggedness 0:09:39 - Emotions and value functions 0:18:49 – What are we scaling? 0:25:13 – Why humans generalize better than models 0:35:45 – Straight-shotting superintelligence 0:46:47 – SSI’s model will learn from deployment 0:55:07 – Alignment 1:18:13 – “We are squarely an age of research company” 1:29:23 – Self-play and multi-agent 1:32:42 – Research taste Look up Dwarkesh Podcast on YouTube, Apple Podcasts, or Spotify. Enjoy!

The Ilya Sutskever episode 0:00:00 – Explaining model jaggedness 0:09:39 - Emotions and value functions 0:18:49 – What are we scaling? 0:25:13 – Why humans generalize better than models 0:35:45 – Straight-shotting superintelligence 0:46:47 – SSI’s model will learn from deployment 0:55:07 – Alignment 1:18:13 – “We are squarely an age of research company” 1:29:23 – Self-play and multi-agent 1:32:42 – Research taste Look up Dwarkesh Podcast on YouTube, Apple Podcasts, or Spotify. Enjoy!

4,131,289 views • 9 months ago

Adam Brown (Adam Brown) is back! General relativity is said to be the most beautiful idea the human mind has ever produced. Most of us will never get to fully appreciate its elegance by taking the 20-lecture graduate course Adam taught on it at Stanford. But in the video below, Adam distills the key idea at its heart so clearly and compellingly that even I could keep up lol. At the core of general relativity, Einstein is trying to figure out the principle behind a particular coincidence: that the mass that resists acceleration and the mass that gravity pulls on just happen to be exactly the same. Adam then leads us through the path of insight which Einstein called his “happiest thought.” Then Adam lectures on black holes. First, by showing how even under special relativity you could create a perpetual motion machine if black holes weren't truly black. And then, by explaining why the observations of an infalling observer and a distant bystander to the black hole would be so radically different Adam leads Blueshift, the team at Google DeepMind cracking science and reasoning. Which gave us the opportunity to discuss at the very end how close we are to AIs that could rediscover general relativity from scratch. Stay till the close for some philosophy of science. 0:00:00 – The coincidence that led Einstein to general relativity 0:16:42 – Gravity is a consequence of curved spacetime, not a force 0:31:46 – Why black holes prevent unlimited energy extraction 0:47:12 – Black holes are the ultimate power plants 1:13:50 – What falling into a black hole would actually feel like 1:18:51 – The three ways we know black holes are real 1:24:21 – The first time we saw gravity bend light 1:29:33 – How far can AI get without experimental evidence? Look up Dwarkesh Podcast on YouTube/Spotify to watch. Enjoy!

Adam Brown (Adam Brown) is back! General relativity is said to be the most beautiful idea the human mind has ever produced. Most of us will never get to fully appreciate its elegance by taking the 20-lecture graduate course Adam taught on it at Stanford. But in the video below, Adam distills the key idea at its heart so clearly and compellingly that even I could keep up lol. At the core of general relativity, Einstein is trying to figure out the principle behind a particular coincidence: that the mass that resists acceleration and the mass that gravity pulls on just happen to be exactly the same. Adam then leads us through the path of insight which Einstein called his “happiest thought.” Then Adam lectures on black holes. First, by showing how even under special relativity you could create a perpetual motion machine if black holes weren't truly black. And then, by explaining why the observations of an infalling observer and a distant bystander to the black hole would be so radically different Adam leads Blueshift, the team at Google DeepMind cracking science and reasoning. Which gave us the opportunity to discuss at the very end how close we are to AIs that could rediscover general relativity from scratch. Stay till the close for some philosophy of science. 0:00:00 – The coincidence that led Einstein to general relativity 0:16:42 – Gravity is a consequence of curved spacetime, not a force 0:31:46 – Why black holes prevent unlimited energy extraction 0:47:12 – Black holes are the ultimate power plants 1:13:50 – What falling into a black hole would actually feel like 1:18:51 – The three ways we know black holes are real 1:24:21 – The first time we saw gravity bend light 1:29:33 – How far can AI get without experimental evidence? Look up Dwarkesh Podcast on YouTube/Spotify to watch. Enjoy!

655,359 views • 1 month ago

What does the next training paradigm look like? 0:00:00 – The big research bet the labs are making 0:02:12 – Grindability is just as important as verifiability 0:06:10 – Will RLVR alone generalize? 0:08:41 – Getting the learning back to the weights 0:15:22 – Dreaming 0:17:23 – What 2027 looks like Also on YouTube, pod feed, and Substack.

What does the next training paradigm look like? 0:00:00 – The big research bet the labs are making 0:02:12 – Grindability is just as important as verifiability 0:06:10 – Will RLVR alone generalize? 0:08:41 – Getting the learning back to the weights 0:15:22 – Dreaming 0:17:23 – What 2027 looks like Also on YouTube, pod feed, and Substack.

750,602 views • 2 months ago

Did a very different format with Reiner Pope – a blackboard lecture where he walks through how frontier LLMs are trained and served. It's shocking how much you can deduce about what the labs are doing from a handful of equations, public API prices, and some chalk. It’s a bit technical, but I encourage you to hang in there - it’s really worth it. There are less than a handful of people who understand the full stack of AI, from chip design to model architecture, as well as Reiner. It was a real delight to learn from him. Recommend watching this one on YouTube so you can see the chalkboard. 0:00:00 – How batch size affects token cost and speed 0:31:59 – How MoE models are laid out across GPU racks 0:47:02 – How pipeline parallelism spreads model layers across racks 1:03:27 – Why Ilya said, “As we now know, pipelining is not wise.” 1:18:49 – Because of RL, models may be 100x over-trained beyond Chinchilla-optimal 1:32:52 – Deducing long context memory costs from API pricing 2:03:52 – Convergent evolution between neural nets and cryptography

Did a very different format with Reiner Pope – a blackboard lecture where he walks through how frontier LLMs are trained and served. It's shocking how much you can deduce about what the labs are doing from a handful of equations, public API prices, and some chalk. It’s a bit technical, but I encourage you to hang in there - it’s really worth it. There are less than a handful of people who understand the full stack of AI, from chip design to model architecture, as well as Reiner. It was a real delight to learn from him. Recommend watching this one on YouTube so you can see the chalkboard. 0:00:00 – How batch size affects token cost and speed 0:31:59 – How MoE models are laid out across GPU racks 0:47:02 – How pipeline parallelism spreads model layers across racks 1:03:27 – Why Ilya said, “As we now know, pipelining is not wise.” 1:18:49 – Because of RL, models may be 100x over-trained beyond Chinchilla-optimal 1:32:52 – Deducing long context memory costs from API pricing 2:03:52 – Convergent evolution between neural nets and cryptography

1,291,518 views • 4 months ago