
33 segments available
Chatted with my friend Leopold Aschenbrenner about the trillion dollar cluster, unhobblings + scaling = 2027 AGI, CCP espionage at AI labs, leaving OpenAI and starting an AGI investment firm, dangers of outsourcing clusters to the Middle East, & The Project. Read the new essay series from Leopold this episode is based on here: https://situational-awareness.ai/ 𝐄𝐏𝐈𝐒𝐎𝐃𝐄 𝐋𝐈𝐍𝐊𝐒 * Transcript: https://www.dwarkeshpatel.com/p/leopold-aschenbrenner * Apple Podcasts: https://podcasts.apple.com/us/podcast/leopold-aschenbrenner-china-us-super-intelligence-race/id1516093381?i=1000657821539 * Spotify: https://open.spotify.com/episode/5NQFPblNw8ewxKolIDpiYN?si=6NaTHAugT2SxZrspW3lziw * Follow me on Twitter: https://twitter.com/dwarkesh_sp * Follow Leopold on Twitter: https://x.com/leopoldasch 𝐓𝐈𝐌𝐄𝐒𝐓𝐀𝐌𝐏𝐒 00:00:00 The trillion-dollar cluster and unhobbling 00:21:20 AI 2028: The return of history 00:41:15 Espionage & American AI superiority 01:09:09 Geopolitical implications of AI 01:32:12 State-led vs. private-led AI 02:13:12 Becoming Valedictorian of Columbia at 19 02:31:24 What happened at OpenAI 02:46:00 Intelligence explosion 03:26:47 Alignment 03:42:15 On Germany, and understanding foreign perspectives 03:57:53 Dwarkesh's immigration story and path to the podcast 04:03:16 Random questions 04:08:47 Launching an AGI hedge fund 04:20:03 Lessons from WWII 04:29:57 Coda: Frederick the Great
Leopold Aschenbrenner discusses the concept of the trillion-dollar cluster, emphasizing the industrial nature of AI development. He highlights the massive investments required for building AI infrastructure, including power plants and data centers, and the implications of this acceleration for the future of AI and global power dynamics.
"What will be at stake will not just be cool products But whether liberal democracy survives, Whether the CCP survives, what the world order for the next century is going to be The CCP is going to ..."
Aschenbrenner elaborates on the rapid growth of AI investments, particularly in data centers, and the projected costs associated with building the necessary infrastructure. He discusses the historical context of AI's evolution and the significant financial stakes involved in maintaining technological superiority.
"tell me about the trillion-dollar cluster. Unlike most things that have recently come out of Silicon Valley, AI is an industrial process. The next model doesn’t just require some code. It’s build..."
The conversation shifts to the financial aspects of AI development, with Aschenbrenner estimating the potential revenue from AI products. He explores the feasibility of achieving $100 billion in annual revenue from AI and the implications for major tech companies as they invest heavily in AI capabilities.
"most of them will be inference GPUs. US power production has barely grown for decades. Now we’re really in for a ride. When I had Zuck on the podcast, he was claiming not a plateau per se, but th..."
Aschenbrenner introduces the concept of 'unhobbling' in AI, discussing how current models are limited and the potential for future models to operate as agents rather than mere chatbots. He emphasizes the importance of integrating advanced AI systems into workflows to maximize their utility.
"how feasible is $100 billion a year from AI revenue? It’s a lot more than right now. If you believe in the trajectory of AI systems as I do, it’s not that crazy. There are like 300 million Micros..."
The discussion focuses on the timeline for achieving true AGI, with Aschenbrenner predicting that by 2027-2028, AI will reach a level of intelligence comparable to the smartest experts. He outlines the challenges and opportunities in developing AI that can perform complex tasks autonomously.
"interacting with them like coworkers. You can do Zoom calls and Slack with them. You can ask them to do a project and they go off and write a first draft, get feedback, run tests on their code, ..."
Aschenbrenner explains the significance of pre-training in AI models, arguing that it provides a foundational advantage for developing general intelligence. He contrasts this with the challenges faced in robotics and the need for models to learn independently.
"Again, if it’s 100 tokens a minute, a few million tokens is a few months of working time. There’s a lot more you can do in a few months of working time than just getting an answer right now. The ..."
The segment concludes with a discussion on the importance of self-directed learning for AI models. Aschenbrenner draws parallels between human learning processes and AI development, suggesting that future models will need to engage in deeper, more meaningful learning to unlock their full potential.
"First of all, pre-training is magical. It gave us a huge advantage for models of general intelligence because you can predict the next token. But there’s a common misconception. Predicting the n..."
In this segment, Aschenbrenner draws parallels between human learning processes and AI training. He explains how self-directed learning, akin to studying a dense textbook, can lead to deeper understanding and retention, and discusses the implications of in-context learning for AI models, highlighting the importance of practice and failure in the learning journey.
"I don’t know, between six months and three years. But it's possible. It’s also very related to the issue of the data wall. Here’s one intuition on learning by yourself. Pre-training is kind of l..."
Aschenbrenner reflects on the pivotal year of 2023 for AI, marked by the explosive popularity of ChatGPT and GPT-4. He discusses the resulting surge in capital investments and revenue for AI companies, predicting that the momentum will continue to build, leading to significant advancements and a shift in public perception regarding AGI.
"By the way, if that take sounds familiar it's because it was part of the question I asked John Schulman. It goes to illustrate the thing I said in the intro. A bunch of the things I've learned a..."
This segment explores the potential economic and political ramifications of AGI as it becomes more integrated into the workforce. Aschenbrenner speculates on the societal reactions to AI replacing jobs and the broader implications for national security, particularly in the context of the US-China superintelligence race.
"as somebody who was really following the AI stuff. What were you doing in 2023? OpenAI. When you were at OpenAI in 2023, it was a weird thing. You almost didn't want to talk about AI or AGI. It wa..."
Aschenbrenner delves into the concept of the intelligence explosion, where AGI could rapidly advance AI research and development. He discusses the potential for AI to automate cognitive jobs, leading to a rapid acceleration of technological progress and the implications for military competition and national power.
"worker. Your stuff is not going to matter. So that's done. SF, this crowd, is paying attention now. Who is going to be paying attention in 2026 and 2027? Presumably, these are years in which hund..."
In this segment, Aschenbrenner highlights the geopolitical stakes surrounding AI development, particularly the potential for espionage and competition between the US and China. He emphasizes the urgency for nations to recognize the strategic importance of superintelligence and the lengths to which countries may go to secure their AI capabilities.
"They’re superbly competent at all things. They're going to figure out robotics. We talked about that being a software problem. Well, you have a billion super smart — smarter than the smartest hum..."
Aschenbrenner discusses the implications of the race for superintelligence, considering how quickly advancements could occur once AGI is achieved. He speculates on the transformative effects this could have on various sectors, including military and technological fields, and the potential for a decisive advantage in global power dynamics.
"it's going to be an extremely intense international competition. One thing I'm uncertain about in this picture is if it’s like what you say, where it's more of an explosion. You’ve developed an A..."
In this concluding segment, Aschenbrenner reflects on the historical context of technological competition, drawing parallels to past global conflicts. He raises critical questions about whether current leaders will recognize the stakes involved in the AI race and the potential for significant geopolitical shifts as nations respond to the emerging realities of superintelligence.
"more people. The trend lines will become clear. You will see some amount of the COVID dynamic. COVID in February of 2020 honestly feels a lot like today. It feels like this utterly crazy thing is..."
Aschenbrenner reflects on the historical context of global conflicts and the stakes involved in the superintelligence race. He compares the current geopolitical climate to past wars, emphasizing the need for the U.S. to recognize the high stakes of competition with China. The discussion underscores the importance of understanding history to navigate future challenges in international relations.
"Again, a lot of this really depends on how much people are feeling it and how much people are seeing it. Our generation is so used to peace, American hegemony and nothing matters. The historical..."
In this segment, Aschenbrenner explores the terrifying possibilities that superintelligence could bring, particularly in authoritarian regimes. He discusses how advanced technologies could enable oppressive governments to maintain control and eliminate dissent. The conversation raises critical questions about the future of democracy and the potential for dictatorship in the age of superintelligence.
"The great power conflict definitely seems compelling. All kinds of different things seem much more likely when you think from a historical perspective. You zoom out beyond the liberal democracy ..."
Aschenbrenner questions the motivations of AI researchers in China who may be contributing to the CCP's agenda. He discusses the potential for Western-educated researchers to resist authoritarian control and the implications for global AI development. This segment highlights the human element in the technological race and the ethical considerations surrounding AI research.
"things I took away from growing up in Germany. A lot of this stuff feels more visceral. My mother grew up in the former East, my father in the former West. They met shortly after the Wall fell. ..."
This segment delves into the shifting power dynamics within AI labs, where employees currently hold significant influence. Aschenbrenner warns that this power may diminish as automation increases. He draws parallels to historical events, emphasizing the need for lab employees to recognize their current leverage and use it wisely in the face of impending changes.
"Will they be in charge though? In some sense, this is also the case in the US. This is like the rapidly depreciating influence of the lab employees. Right now, the AI lab employees have so much ..."
Aschenbrenner discusses the strategic importance of building AI clusters in the U.S. versus authoritarian regimes. He highlights the risks associated with outsourcing AI development to countries like the UAE, where security concerns could arise. The conversation emphasizes the need for the U.S. to maintain control over AI technology to prevent potential threats from adversarial nations.
"Interesting. Let's go back to the $100 billion revenue question. The companies are trying to build clusters that are this big. Where are they building it? Say it's the amount of energy that woul..."
In this segment, Aschenbrenner explains the significance of compute ratios in the context of AI development. He discusses how a slight advantage in computational power could lead to significant geopolitical consequences. The conversation highlights the urgency of maintaining a competitive edge in AI technology to prevent adversaries from gaining the upper hand.
"Another thing is they can just seize the compute. The issue here is people right now are thinking of this as ChatGPT, Big Tech product clusters. The clusters being planned now, three to five yea..."
Aschenbrenner outlines the potential dangers of an AI arms race, particularly in a scenario where the U.S. and China are in close competition. He emphasizes the importance of strategic planning and caution to avoid self-destructive outcomes. This segment underscores the need for careful consideration of national security in the rapidly evolving landscape of AI technology.
"You can do a lot with 33 million extremely smart scientists. That might be enough to build the crazy bio weapons. Then you're in a situation where they stole the weights and they seized the comp..."
Aschenbrenner discusses the feasibility of building AI clusters in the U.S. and the challenges involved. He presents two potential paths: utilizing natural gas or pursuing green energy projects. The conversation highlights the need for regulatory reform and a national commitment to ensure that the U.S. remains competitive in the global AI landscape.
"If you're in this really tight race, this sort of feverish struggle, that's when there's the greatest peril of self-destruction. Presumably the companies that are trying to build clusters in the ..."
Aschenbrenner outlines two potential paths for powering AGI clusters in the U.S.: utilizing natural gas or pursuing green energy megaprojects. He emphasizes the need for a deregulatory push to facilitate the development of renewable energy sources and the importance of balancing climate commitments with national security interests.
"I think natural gas production in the United States has almost doubled in a decade. You do that one more time over the next seven years, you could power multiple trillion-dollar data centers. The..."
Reflecting on World War II, Aschenbrenner draws parallels between past industrial mobilization in the U.S. and current challenges in AGI development. He discusses labor agitation during the war and the latent industrial capacity that eventually led to rapid mobilization, suggesting that similar potential exists today despite current obstacles.
"We have to do at least one. Then this stuff is possible in the United States. Before the conversation I was reading a good book about World War II industrial mobilization in the United States cal..."
Aschenbrenner examines China's industrial capabilities in the context of the AGI race. He warns against underestimating China's potential to scale up AI production and emphasizes the need for the U.S. to remain vigilant and proactive in its approach to AGI development and competition.
"It wasn’ just that. Before 1939, the American military was in total shambles. You read about it and it reads a little bit like the German military today. Military expenditures were I think less ..."
Discussing the geopolitical implications of AGI, Aschenbrenner argues that if the U.S. does not engage with Middle Eastern countries, they may turn to China for AI development. He advocates for a coalition of democracies to lead AGI development while also considering the interests of other nations, including those with authoritarian regimes.
"some point, the same way the United States and a lot of people in the US government are going to wake up, the CCP is going to wake up. Companies realize that scaling is a thing. Obviously their w..."
Aschenbrenner raises concerns about the potential for a bidding war over AGI technology among global powers. He reflects on past strategies employed by OpenAI to leverage competition between the U.S., China, and Russia, questioning the ethics and implications of such an approach in the context of national security.
"work with them, they'll just support China. There's some merit to the argument in the sense that we should be doing benefit-sharing with them. On the road to AGI, there should be two tiers of coa..."
In this segment, Aschenbrenner discusses the risks of espionage in the AI sector, highlighting the ease with which sensitive information can be stolen. He emphasizes the importance of securing AI algorithms and the potential consequences of losing proprietary technology to adversaries like China.
"Suppose you're right. We ended up in this place because, as one of our friends put it, the Middle East has billions or trillions of dollars up for persuasion like no other place in the world. Wit..."
Aschenbrenner delves into the significance of secrets in the race for AGI, arguing that while compute power is crucial, the underlying algorithms and insights are equally important. He discusses the need for robust security measures to protect intellectual property and the implications of sharing knowledge in the AI community.
"espionage, it needs to be much more intense. The thing that people aren't paying enough attention to is the secrets. The compute stuff is sexy, but people underrate the secrets. The half an order..."
Drawing historical parallels, Aschenbrenner reflects on the secrecy surrounding the development of the atomic bomb during WWII. He questions whether the current AI landscape will repeat past mistakes regarding information sharing and the potential consequences of premature disclosure of critical advancements.
"all this stuff was published until recently. Look at Chinchilla Scaling laws, MoE papers, transformers. All that stuff was published. That's why open source is good and why China can make some go..."
Aschenbrenner explores the competitive landscape of AI development, emphasizing that even a small lead can have significant implications. He discusses the potential for an intelligence explosion and how a slight advantage in AI capabilities could lead to drastic changes in global power dynamics, echoing historical patterns of technological advancement.
"The reason this would matter is if being one year ahead would be a huge advantage. In the world where you deploy AI over time they're just going to catch up anyway. I interviewed Richard Rhodes,..."
This segment delves into the risks associated with a close race in AI development between the U.S. and China. Aschenbrenner warns that a minimal lead could lead to dangerous military escalations and the emergence of new weapons of mass destruction. He stresses the need for careful alignment and strategic planning to navigate this volatile landscape.
"might be the difference between a system that's sort of human-level and a system that is vastly superhuman. It might be like five OOMs. Look at the current pace. Three years ago, on the math benc..."
Aschenbrenner highlights the often-overlooked geopolitical implications of AI advancements. He argues that the ongoing scaling of AI technology will reshape global relations and national security, similar to how COVID-19 shifted priorities worldwide. The segment emphasizes the urgency for leaders to recognize and address these emerging challenges.
"So you have to look at it from the point of view that these technologies are dangerous, from the alignment point of view. It might be really important during the intelligence explosion to have a..."