Leopold Aschenbrenner - 2027 AGI, China/US Super-Intelligence Race, & The Return of History

[@sausageskin]

Amount of enthusiasm in these two people is already at a superhuman level.

[@memegazer]

It's almost morbid to watch people giddy about possible ww3

[@VictoriaSobocki]

It’s great

[@squamish4244]

@@memegazer Dwarkesh is 25 and Leopold is 21, mortality is still just a concept to them. To be cynical about it.

[@BrianMosleyUK]

The highest compliment I can pay, is that I watched the entire four and a half hours.

[@Aaron637]

I hope you used 1.5 or 2 times speed. Thats the hack.

[@jonathanelkefif1128]

​@Aaron637 it's on by default in this one, neat 😂

[@BrianMosleyUK]

@@Aaron637 lol that would have been funny.

[@tm73827]

Clearly unemployed. I fed this video into my LLM and it summarized it for me with a 5 min summary.

[@BrianMosleyUK]

@@tm73827almost, unemployable and that's a good thing. 😂

[@daniellawson9894]

I like how Alec Radford is used throughout the conversation as a von Neumann-like analogy for an extremely productive artificial researcher

[@Beyondflix]

finally a podcast I don't have to put on 2x speed

[@senju2024]

Yeah. I thought I accidently hit the 2x speed button.

[@michaelhoffman4163]

@@senju2024 Too funny. It's one of the few where I had to go down to 1.75 to understand the dialog. Love the high information density.

[@senju2024]

@@michaelhoffman4163 You are so right! Lots to unpack. Fast talk with high level of info for 4 hours. It may take me a couple of weeks to get a full understanding of this video along with his paper he published. I really want to understand his headspace and info....not all I agree on but more about info intake and then create my own views later.

[@zorawarsingh11]

I swear I double checked my settings too

[@rei_carvalho]

Thanks for the great content. good luck on your journey my friend.

[@DwarkeshPatel]

Thank you for this generous contribution 🙏

[@guts9000]

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

[@avi3681]

4:17:59 "For me, the real depreciating asset is human capital." In a nutshell the AI revolution is about removing paid labour as an input to the economic equation. Why pay humans to do a job when you have millions of unpaid AGI workers. The industrial revolution created skilled jobs in factories that made human capital worth investing in, this led to universal public education, widespread literacy, the rise of the middle class, and the spread of liberal democracies. So if Leopold’s version of events is allowed to play out, no human capabilities will have any market value and therefore the only basis for wealth will be who grabs the most financial capital leading up to the ASI. It makes perfect sense from this perspective for him to be starting a hedge fund. I wonder what Dwarkesh makes of this from his pro-free market Libertarian standpoint. How can there be a meritocratic justification of a free-market system in a future world where even the most talented and skilled human being has nothing of value to offer the market?

[@kreek22]

None of this is worth worrying about. You don't pass through the Singularity.

[@avi3681]

​@@kreek22 That's dark man 😬

[@fernandohiar9985]

Capitalism is not meritocratic, it concerns the concentration of capital and this makes sense for companies to generate fewer and fewer jobs to generate increasingly fatter revenues for shareholders and investors.

[@avi3681]

@@fernandohiar9985 It makes sense to a point for the rich to want to concentrate power and get richer, but past a certain extreme they risk undermining their own legitimacy and losing popular support/compliance. We saw this with the end of the Gilded Age as the 19th and early 20th century industrialists had their monopolies broken up and there was a rise of union power and social welfare spending after the Great Depression. History isn't entirely one-sided on these things.

[@themsuicjunkies]

It also ignores that we live in a consumer economy and even if firms will buy and sell to other firms the economy might shrink instead of growing exponentially like this people think in a deflationary hell.

[@davisclute7705]

Thanks!

[@shankfish1990]

If adderall was a conversation.

[@jimdye7431]

i actually think its crack, his teeth are rotted

[@davidh.65]

Seriously was wondering how many espressos Leopold must have had before this interview

[@Aesthetic_Champ]

Yeah

[@pseudoanonymous]

I had to double check I wasn't on 2x playback speed

[@JimStew28]

He decided to leave Germany at 15 to attend Columbia University. The dudes brain must be in 2x speed

[@DwarkeshPatel]

Had a blast discussing (and sometimes debating) Leopold's view of what the future is going to look like. If you want it straight from the source, you can get it here: situational-awareness.ai/ Highly highly recommend.

[@sup3a]

Haven't watched yet but judging from section titles damn, im in for a wild ride. better get my snacks

[@atillacodesstuff1223]

wow thanks for the link, that's an amazing website

[@TheManinBlack9054]

Does it really matter who develops AGI first if it's not controllable? Is American AM really that much better than Chinese AM?

[@atillacodesstuff1223]

@@TheManinBlack9054 "Does it really matter who develops AGI first if it's not controllable" no

[@1Howdy1]

It's kind of a required new hire reading for the new hire AI industry, for now. It might be a good dynamic article for the researchers. Very good interview. You seemed ready to talk until the sun came up with an endless supply of energy drinks. A grasp of AI at the moment.

[@FrotLopOfficial]

If I had a dollar for every "yep" I'd be able to build the trillion dollar cluster.

[@zorawarsingh11]

Yeah yeah yeah

[@enrico5399]

Can AI be that funny? Ahahaha

[@brak512]

yep

[@gagayou9]

I can beat you getting a dollar for each "you know"

[@LivBoeree]

So stoked you interviewed Leopold, he's such a good dude. Excited to watch!

[@charlesalexanderable]

CEPR: How influencer cartels manipulate social media: Fraudulent behaviour hidden in plain sight

[@reluctantrealist6861]

the sinophobia is cringe

[@Kazekoge101]

@@reluctantrealist6861 Not surprising, dude's from Columbia, CIA finishing school

[@ryzikx]

@@reluctantrealist6861fear of china: sinophobia fear of usa: approved 👍 social credit +999

[@honkytonk4465]

​@@reluctantrealist6861ccp-phobia that's a totally different thing.

[@d0tz_]

Only part way through, but I think something really unique about this conversation is just how confident Leopold seems about his assertions. Like you can feel that he *knows* these things *will* happen, like it’s just plainly obvious to him. Whereas even with people like Yud, there’s somehow a joking unserious tone when they talk about blowing up data centers. Like a lot of Leopold’s talking points are things we’ve heard over and over, and probably this confidence is just a byproduct of him being young and being very involved in “AI-doomer” circles, still, it does stir up something in me.

[@alvin.w.graylin]

Strong Confidence about the future without nuance is a sign of inexperience…

[@davidh.65]

@@alvin.w.graylinexactly. Makes this whole thing hard to listen to

[@thorvaldspear]

@@alvin.w.graylin It's a sign of inexperience until it isn't. There's plenty of evidence pointing to the fact that he's exceptional; what if he's on the right side of Mount Stupid? EDIT: And besides, his worldview _is_ actually nuanced. He admits he's not sure that fast takeoff is possible, he just thinks it's probable, and that we should prepare for this worst case scenario as a precaution.

[@joesligo1516]

Yeah, I'm having a hard time too. It feels like the typical insular bay area VC mindset looking for something to disrupt

[@BrianMosleyUK]

He's so confident he has to shout over Dwarkesh to show his insecurity. Very smart and naive youngster.

[@lifecrzy]

Bro takes active listening to another level 😂

[@staticmind1872]

Yeah yeah yeah yeah, yeah

[@Jstudious]

Yeah!

[@andredillon7685]

ya

[@jakemeyer8188]

Literally said the same exact thing in my head.

[@kenwarner]

Editor was like fuck this I'm not cleaning all that up I quit

[@Pixelume]

Fascinating conversation. Watched the whole thing and it really stirred my imagination on what the next decade would bring. Loved that the conversations revolved around the interplay of so many areas (geopolitical, history, economics, national security, tech etc.) and it's relationship to how AI will soon unrecognizably change the world we live in in ways we don't yet understand.

[@smellthel]

I like looking to hear other people's perspectives on these topics, so I'm glad I found this episode! This is literally the longest KZbin video I have ever watched.

[@ShpanMan]

This conversation is literally hearing my little brother and his friend in the next room talking excitedly about how they are going to become CEO and President when they are in middle school.

[@erics7107]

😂 … nail meet head. It is a pretty sweet nerdy bromance tho.

[@netscrooge]

I agree. Intelligence without maturity.

[@crawfordscott3d]

A little scary in a way. Somehow the adults all went insane and the children have the keys to the car. What are we doing

[@SirSpinach]

In fast moving fields, you don’t have to be old and grey before you are a world expert

[@ShpanMan]

@@SirSpinach Might want to stop breastfeeding before you call yourself one tho 😂

[@kentbergstrom]

I love their brilliance. I also love the authenticity of their moments laughing. Two people loving what they do, in this case thinking and discussing.

[@x3haloed]

Leopold knows what he’s talking about. Art theft is a distraction. Job losses are a distraction. The issues he brings up here are going to be upon us in no time, and it’s going to be difficult. We need to be preparing for this.

[@chadalderson6245]

Jobs losses are a distraction? Maybe if you have a trust fund LOL. I'm very much for acceleration but we need to have a plan in place for the job disruption, while not slowing down our innovation. Remember, humans tend to eat each other when they can't access food.

[@x3haloed]

@@chadalderson6245 When you're tied to the railroad tracks and you hear a whistle in the distance, even a broken leg is a distraction. I'm not saying that job losses aren't worth worrying about. I'm saying that a train is coming, and we need more dialog about it.

[@amsrremix2239]

He’s only saying this because he assumes all countries want military strength… because that’s where he comes from. China has no reason to fight us…. And has said multiple times they don’t want to fight. They want economic power…. In the Middle East most of them hate US/UK Israel. But that’s nobody’s fault but our own…

[@willwimbiscus7456]

​​​@@x3haloedRight on. Love the analogies Leopold is making to the Manhattan project. This time has a really odd feeling. I get this odd sense we're somehow living in the past. With AGI and ASI seeming somewhat inevitable, people just haven't woken up yet, and just existing in day to day life feels surprisingly like how you'd imagine the few minutes in a neighborhood prior to bomb detonation..

[@x3haloed]

@@willwimbiscus7456 Yeah, I agree. It feels really strange. I don't believe that it's certain doom, but everything is going to change very quickly, and I think it's going to be a difficult time.

[@ahpacific]

The level of unbreaking unyielding converging to a similar discrete stop nerd laughing between these two is off the charts. Amazing.

[@brak512]

Revenge of the Nerds.

[@sup3a]

At 43:45 he tells the reason (implicitly) why some of OpenAI retaliated against Sam: putting future ASI effectively in the hands of rich Saudis through their deal

[@sup3a]

My wild theory is he's JimmyApples (judging from the 2:32:00 part)

[@wdwuccnxcnh7022]

The part about the bidding war for AGI is genuinely crazy. I really hope that's not true

[@BrianMosleyUK]

​@@sup3athat's a plausible theory.

[@abstratogvm136]

He is not Jimmy apples, Jimmy apples got doxxed some months ago on Reddit, but post was quickly deleted

[@BrianMosleyUK]

@@abstratogvm136 who is he then?

[@Macorelppa]

Rest In Peace Dwarkesh Patel's podcast video editor may his soul rest in peace 🕊☮😥

[@eSHkidd]

Is this a joke about him quitting or do you have evidence he passed?

[@yw1971]

@@eSHkidd 4:32:06 hours proves it

[@squamish4244]

@@yw1971 What do you mean? That it's just Dwarkesh now?

[@saurav1096]

@@eSHkidd because its a long video to, long video can be cumbersome to edit. To be fair , podcast dont need much editing , just checking the audio if something needs to trim

[@jolima]

Some people commented that the video has been sped up 5-10% - edited to make the vid shorter 😝

[@dandan8787]

Dwarkesh I just wanted to tell you that I currently have 4 of your podcasts open in tabs, waiting for me to watch them all. I am AMAZED over your guests and yourself in this podcast. This is absolutely incredible. Keep doing what you want to do, interview who you want to interview, don't think "what does the audience want? probably this guy..." because at that point you're just guessing. We want the guests you are providing us with. This is probably the best podcast I've seen, on par with Joe Rogan. Simply amazing. Thank you. Best regards from Sweden.

[@LatentSpaceD]

Woahh.. thank you Dwarkesh, Leopold. Seriously deep insightful content. I appreciate you both

[@radioazma]

historic interview, need part 2

[@cosmeon1385]

Drinking game for this podcast, every time Leopold says "you know" you have to drink 😂

[@skoto8219]

or for each subsequent “yeah” after five consecutive “yeah”s

[@katlehomahlelebe1401]

I saw this comment 23mins into the video. After that, I couldn't unhear it😂

[@jamesatkins7592]

Oh shit....

[@thetobyg]

Let’s substitute the drink with an Adderall!

[@calebrcannon]

Or “look…”

[@dustinbreithaupt9331]

Started watching you after your Ilya inteview. I believe you had around 5k followers then. I am floored to see you take off like this. Congrats man. You seriously are the most prepared and interesting podcaster to watch. I have a one year old son and a busy home life but I WILL find the 4.5 hours to watch this whole interview. Fascinating.

[@1adamuk]

These people that laugh like seals are going to run the world? Great conversation BTW. IMO Dwarkesh is the best interviewer / podcaster for these types of topics bar none.

[@appipoo]

As opposed to the serious people currently running the world with pure stern resolve? Those people laugh like seals and hastily wipe their butts after morning coffee. Just like the rest of us.

[@bossgd100]

GALAXIES

[@vacc1001]

I came here to see if somebody thought about the "seals". Happy to see someone noticed it too. WTAF was that?

[@oowaz]

true he is pretty good

[@therainman7777]

😂 the laughs were brutal

[@Gunrun808]

I love the giga-nerd synergy these two have

[@Crashrapescrypto]

old mate aint that nerdy, dwarkesh

[@HistorIAsImposibles776AC]

Yes, we need a giga-nerd Channel now

[@michaelsimpson9175]

the laughs got me

[@therainman7777]

The laughing.. dear god the laughing was brutal.

[@8__vv__8]

Hey ChatGPT, analyze these honks and tell me if this is a pair of geese or computer scientists

[@MNSalty]

You're an excellent interviewer, great balance of asking sharp questions, pushing back on the interviewee's ideas, and continuing the conversation . Keep crushing it!

[@5anjuro]

I love how these two giggle in a self reinforcing positive feedback loop, especially over some really dark stuff.

[@jolima]

3:44:35 It’s telling the two don’t get this: After World War II, Germany adopted the social market economy, which was designed to combine the market economy with a strong emphasis on social security and welfare. One of the lessons from the Weimar Republic era was that economic instability and widespread social inequality could fuel populism and extremism. The social market economy aimed to prevent such instability by ensuring a broad base of economic prosperity and social welfare, thereby strengthening the democratic system. This approach sought to provide economic opportunities for all and to protect citizens from the vulnerabilities associated with pure market capitalism, thus fostering social cohesion and political stability.

[@8__vv__8]

It was a nice idea, but the European right, including AfD, is increasingly popular. Perhaps they’ve had help from overseas, but economics isn’t politics.

[@TheKarlslok]

Maybe the U.S should try to do the same, and maybe they would get a better political system then the sh1tshow they have today...

[@bastianunterberg1501]

Thank you, for this thought. It brings me directly to: If productivity increases, so does value creation. Distributing the added value in ways it serves humanity and society is key to get right.

[@themsuicjunkies]

​@@8__vv__8 Because neoliberalism and globalization has undermimed the social contract that was created post WWII. Not by a secret cabal or anything is just market dynamics.

[@therainman7777]

Who cares? It didn’t work, and it’s also not relevant to the conversation being had in this podcast.

[@usamasyed91]

I think the editor quit because of how often Leopold said 'yep' while Dwarkesh talked. (Great convo btw)

[@videowatching9576]

Fair! Especially when playing 1.5x etc.

[@ynboxlive]

'yep, yep, yep'....taking a page from Andreesen

[@TheInsideView]

very refreshing to hear short ai timelines takes that actually take automation of AI R&D and scaling up of investment into account, with detailed scenarios of how that would play out when I met leopold five years ago he was 17 and already writing a 100 pages report on existential risk and growth, his new piece of writing on situational awareness is beautifully written and I can't wait to see what he does next

[@daniellawson9894]

would be cool if you do a video or have any insights to share you have on his writing piece

[@kreek22]

Like many other clever fellows, he jettisons consistent thinking when it interferes with his fun thinking. The ASI idea elephant will eat all of his other ideas before they have a chance to achieve their full awesome potential. But, that just isn't much fun, now is it? I sometimes suspect that the very clever are common compared to the number of people who are both very clever and very careful thinkers. First category includes: virtually all officially recognized American public intellectuals, this Valedictorian guy above, Tyler Cowen, Curtis Yarvin, and, of course, the slightly dead David Hume. Second category includes: Greg Cochran, Zvi Mowshowitz, and, of course, the supposedly dead John von Neumann.

[@nicbarth3838]

Man he is pretty insightful on this issue glad we have him a part of this LLM project and he is so likable too!!! gosh man I want to be like this man.

[@af.tatchell]

Dwarkesh you have the most incredible patience 👏 some fascinating perspectives here, but definitely would have been nice to hear some more of your questions come through

[@mallow610]

Finally someone who isn’t ignorant to the fact that the government will take over as fast as they can if AGI is imminent

[@kevinsutavee4844]

the most important part of this conversation is they both agree shoes indoors is a no-no

[@RareBirdGames]

I don't want to ruin the podcast for anyone, but the amount of times Leo says "Yeah, Yeah, yes yes" is maddening hahahaha

[@garlictoastlife5587]

Yeah

[@BritishProudnShit]

Or ‘you know’

[@FrotLopOfficial]

Yep yep yayayayaya yepyepyep ya ya ya yep yep

[@nowithinkyouknowyourewrong8675]

you know, like, you know, um

[@eatdirtnetwork]

I love it. 😂

[@AndyOnAI]

Wow, a riveting look on the future. Thanks for the podcast, this was great to take in over the weekend. The upcoming years will truly be wild. Well done!

[@broadbrook108]

The ease with which he glosses over the power and compute build required left me feeling that Leopold, like many people in the Valley, need to talk to people in the energy industry as their predictions, especially when it comes to speed of progress, lack grounding in external realities. Its like listening to Chamath Palihapitiya on teh All In Podcast (which I love). No electricity is not free and its far from being free,

[@squamish4244]

Yeah, his idea to power everything with natural gas would be a disaster for the climate. He is obviously brilliant, but I don't think he has researched just how bad that would be for the climate. And the American energy grid needs major updating. (Which is happening, due to the infrastructure bill.) Meta has bought or signed deals for at least 1 GW of existing or in construction solar power in the last year. American energy policy has only recently stopped being insane. It needs nuclear power, and a lot of it.

[@therainman7777]

If you read his full blog post you’ll see that his predictions are very much grounded in reality. You can still argue that he’s wrong, but he’s definitely not pulling these numbers out of thin air.

[@squamish4244]

@@therainman7777 There are companies now devoted to using AI to help reduce locations' power consumption. Google's datacentres achieved some massive reduction in energy usage hiring one and its RL techniques.

[@jayhill5671]

@@squamish4244 Listen to his podcast with Daniel Yergin, it takes like 20 years to get a permit.

[@MichealScott24]

🎯 Key points for quick navigation: 00:00 *🌐 The Trillion Dollar Cluster and AI Industrial Process* - The growth of AI involves building massive clusters, resembling industrial processes rather than just coding. 02:04 *🔄 Scaling AI Infrastructure and Economic Feasibility* - The scale-up of AI infrastructure, exemplified by clusters, follows a trajectory towards increasingly massive investments. 04:25 *📈 Projected AI Progress and Economic Impact* - The trajectory of AI progress suggests the emergence of true AGI by the 10-gigawatt cluster range. 07:55 *🤖 System 2 Thinking and the Future of AI Development* - The transition to System 2 thinking in AI models involves navigating challenges in error correction, planning, and cohesive understanding. 15:55 *💡 Discussion on the evolution of AI capabilities* - Evolution of AI capabilities from raw mass to sophisticated chatbots. 21:22 *🔍 Speculation on the impact of advanced AI on society and geopolitics* - Predictions about the societal and geopolitical implications of advanced AI by 2027. - Analysis of how AI advancement could lead to an intelligence explosion. 32:36 *🌍 The geopolitical implications of AGI development* - The historical norm of intense international competition is highlighted, contrasting with the current period of relative peace. 36:03 *🛠️ The potential for dictatorship with superintelligence* - The discussion delves into the possibilities of dictatorship empowered by superintelligence, highlighting the risks of absolute control and suppression. 39:46 *🤖 Recruitment for AI projects and considerations of allegiance* - The debate revolves around the recruitment of AI researchers for projects and the potential allegiance to political regimes. 41:27 *⚡ Location considerations for AGI clusters* - The discussion focuses on the geographic location of AGI clusters and their energy requirements. 49:30 *🌍 Balancing Climate Commitments and National Security* - Considering the tension between climate commitments and national security priorities. 50:41 *💡 Historical Analogy: World War II Industrial Mobilization* - Drawing parallels between World War II industrial mobilization and current challenges in climate change and technological advancement. 52:48 *🌐 The Growing AI Capacity of China* - Discussing the potential for China to mature its AI capabilities and industrial capacity. - Examining strategic planning considerations for AI labs in light of the evolving geopolitical landscape. 54:42 *💰 Collaboration with Middle Eastern Countries* - Considering the potential benefits and risks of collaboration with Middle Eastern countries in AI development. 55:36 *🛡️ Concerns about AI Technology Leakage* - Highlighting concerns about the leakage of AI technology and its potential national security implications. 57:15 *🤔 Evaluating Leverage and Technological Catch-Up* - Evaluating the leverage and technological catch-up dynamics between different countries in the AI race. 59:12 *🔒 Importance of Securing AI Secrets* - Emphasizing the importance of securing AI secrets and the potential implications for technological advancement. 01:05:13 *🔑 Learning from Historical Secrecy in Technological Advancements* - Reflecting on the historical significance of secrecy in technological advancements and its implications for AI development. 01:05:44 *🚀 The importance of parallel invention in history,* - Parallel invention often occurs due to theories suggesting the next technological step. 01:09:16 *💡 Underestimation of geopolitical implications of AI,* - Many people overlook the geopolitical ramifications of AI development. 01:14:18 *🔒 Challenges in securing AI labs against espionage,* - Securing AI labs from state-level espionage requires intense measures such as air-gapping, security clearances, and continuous monitoring. 01:18:10 *🌍 Considerations for international cooperation versus competition in AI development,* - The stability of international arms control agreements becomes challenging in the face of rapidly evolving AI technology. 01:23:00 *🌍 International Super-Intelligence Race* - Discussion on the possibility of a super-intelligence race between nations. 01:24:42 *🤝 Diplomatic Solutions for AI Arms Race* - Proposal for establishing diplomatic agreements to mitigate risks in the AI arms race. 01:26:08 *🪐 Vision for the Future: Buying Galaxies* - Anecdote about discussions regarding acquiring galaxies. 01:28:06 *🛡️ Security Concerns and Geostrategic Risks* - Discussion on security concerns related to AI infrastructure, particularly data centers. 01:30:42 *📊 Public vs. Private Control of AGI Development* - Examination of the roles of public and private entities in AGI development. 01:39:36 *🌍 Global power dynamics* - Discussion on historical evidence supporting cooperation and market-based incentives in balancing power dynamics. 01:41:15 *⚠️ Volatility in AI development* - Prediction of an initial volatile and dangerous period in AI development. 01:42:52 *🛡️ Government vs private sector control* - Debate on the efficacy of government projects versus private sector initiatives in AI development. 02:12:12 *🧠 Insights into AI Policy and Nationalization* - Discussion on the likelihood of nationalization of AI technologies. 02:13:12 *💡 Educational Journey and Cultural Differences* - Leopold's academic journey, including graduating from college at a young age. 02:16:44 *🔍 Early Interest in Economics and Productivity Insights* - Leopold's early interest in economics and his approach to learning. 02:18:36 *💰 Career Path and Experience at Future Fund* - Overview of Leopold's career after college, including research and work at Future Fund. 02:29:25 *🧠 Joining OpenAI's Super Alignment Team* - The goal of the Super Alignment Team at OpenAI was to develop successor techniques to reinforcement learning for controlling superhuman AI systems. 02:31:32 *🔒 Dissolution of OpenAI's Super Alignment Team* - OpenAI's decision to dissolve the Super Alignment Team was influenced by personnel changes and a shift in strategic direction. 02:32:29 *🔍 Allegations and Departure* - Leopold Aschenbrenner's departure from OpenAI was attributed to allegations of leaking confidential documents and engaging in policy discussions without alignment with company objectives. 02:43:44 *🌐 Geopolitical Implications and Drama at OpenAI* - OpenAI's pursuit of AGI raises geopolitical concerns and internal tensions due to perceived inconsistencies between mission statements and actions. 02:47:01 *🧠 Scaling Research Efforts* - Research progress doesn't scale linearly with the number of researchers. 02:49:13 *🔍 Log-Log Plot Theory* - The relationship between cumulative research effort and progress follows a log-log plot, where progress becomes increasingly harder to achieve. 02:51:20 *💼 Managing Research Teams* - OpenAI's ability to scale its research team is limited by transaction costs and management complexities. 02:52:15 *🚀 Recruiting Top AI Talent* - Despite efforts to recruit the best AI researchers, simply increasing the number of researchers doesn't guarantee faster progress. 02:55:17 *🛠️ Challenges of AI Revolution* - The AI revolution will likely face challenges similar to past technological revolutions, such as managing and coordinating large-scale AI teams. 03:03:17 *📉 The importance of data in AI progress* 03:05:09 *🧠 Contingency of AI progress on data abundance* 03:06:19 *🔄 Rethinking AI learning methodologies* 03:08:38 *💡 Progress in AI model capabilities* 03:10:58 *🌐 The transformative potential of AI on society* 03:20:08 *🧠 Considerations on Model Scaling and Computational Efficiency* 03:21:19 *📈 Projecting AI Progress and Scaling Laws* 03:23:38 *🔄 Scaling Laws, Experimentation, and Model Capabilities* 03:26:55 *🌐 Alignment, Geopolitical Conflict, and AI Development* 03:29:26 *🤖 Alignment, Future Visions, and Human Conflict* 03:34:00 *🛤️ Pathways to AGI, Misalignment Risks, and Technical Solutions* 03:37:04 *💡 Addressing concerns about rapid intelligence explosion* 03:39:12 *🌐 Challenges in alignment within private AI labs* 03:42:17 *🌍 Historical analogies and insights on Germany's role in AI* 03:49:08 *🔍 Challenges in understanding and engaging with China's AI landscape* 03:54:01 *🧠 Ethical Considerations in AI Development* 03:55:09 *🛡️ National Security Concerns and Espionage in AI Research* 03:56:19 *🤔 Balancing Transparency and National Security in AI Discourse* 03:58:01 *💼 Personal Immigration Story and Career Path* 04:03:27 *🙏 Philosophical Reflections on Belief Systems and Values* 04:09:08 *💰 Rationale for Establishing an Investment Firm* 04:11:44 *💰 Investment Strategy for AGI* 04:12:27 *📈 Betting on AI Beta* 04:14:20 *📉 Impact of Real Interest Rates on Equities* 04:15:46 *🤔 Challenges in Predicting AGI's Market Impact* 04:20:07 *🌐 Global Competition and Industrial Capacity* 04:24:03 *💡 Speculating on Future Applications of AI* 04:26:08 *🔄 Adaptability and Situational Awareness* Made with HARPA AI

[@TylerAyyy]

truely one of the most interesting people i've ever seen talk about anything. Please have him on in the future as well

[@TheRaizerx]

Goddam a 4 hr conversation. There goes my morning

[@BritishProudnShit]

Indeed. Had to call off a couple meetings. Damnit.

[@aisle_of_view]

And you can't play it at 1.25 because they're already talking so fast

[@RaySmith-zg7od]

​@@aisle_of_view😂😂😢😢😂😢

[@mehsmehmeh]

:"AGI is the most dangerous weapon ever created. How will I protect humanity from its horrors? Ah yes, I will start a hedge fund."

[@ResurgentVoice]

😂💯 This comment deserves to be the top comment! Jesus, these two chuckle heads are already gleefully talking about saving up to buy galaxies! 🤦‍♀️ The water and energy resources needed to actually build the infrastructure and data centers is totally ignored. Climate change is a total afterthought. The biosphere can go suck it because… progress? 🤷‍♀️ Ugh! The lack of wisdom and utterly ridiculous hubris is astounding! 🤦‍♀️

[@senju2024]

That is because none of the AI labs will take him as he is now labeled as a "LEAKER". No way will they hire him.

[@crawfordscott3d]

​@@senju2024why are any AI companies going to worry about safety when open AI isn't. No point in hiring someone within his thought domain

[@michaelwoodby5261]

​@@senju2024He referenced America's reduction in military spending, so maybe they won't hire him because they want someone at least tangentially attached to reality.

[@kreek22]

@@michaelwoodby5261 It's been fitfully declining (as a % of GDP) for 60 years.

[@cmmndrblu]

The HR person at Open AI told this guy it was "racist" to be concerned about CCP espionage... wow.

[@adityaprasad5149]

Dwarkesh Patel is the best tech podcaster I know for asking questions. Almost all of them are so perfect that I am in awe. Keep this up man, you are doing important work.

[@TheLegendaryHacker]

56:03 I'm sorry what??? Do you have a source for that claim, because it would be insane if true

[@StockPursuit]

He believes everything he hears. He believes the data the Chinese government puts out.

[@tor4472]

Right!

[@lachland592]

The whole ‘what happened at OpenAI’ section is insane.

[@umuti5ik]

If the AI developments being talked about here do come to pass, the world will change completely. We have so little information about that world that any theorizing about it now is bound to be completely off the mark.

[@luckytaylor8125]

Yes

[@willd.8040]

Absolutely. Especially when there are 100million phd-level AI bot researchers, working 24/7 at 100x the speed of humans, never even needing to rest. It creates a feedback loop where as they improve the AI, they improve as researchers, and who knows what breakthroughs they will make in a short period of time. Especially if these massive clusters are built. The world in 5 years might be unimaginable right now.

[@MSinFL79]

Agreed. Strikes me as a manifesto of what he would like the future to look like and how it will happen in predictive timelines along with all the forces at play.

[@therainman7777]

Not really. I think we can make some intelligent predictions now that are very likely to be approximately accurate.

[@TensorClips]

Love them both but Leopold says "ya" a million times in this pod lol

[@einsteinsapples2909]

Ya ya ya ahuh

[@serendipity-108]

It’s the autism

[@riveradam]

Initially irksome, and I wouldn't recommend anyone do it deliberately, but in this case I think it's evidence of really good active listening.

[@GaryMillyz]

It's obviously spectrum. People are so dumb. And Leo is a genius.

[@FrotLopOfficial]

@@riveradam Not active listening, more "I dont care what you have to finish, I get the question, I just want to talk". Its Leopold being utterly annoyingly impatient. Total opposite of active listening imo

[@sup3a]

The new Lex Friedman in my books. Keep going bro!

[@Bboy235]

Much smarter

[@xsuploader]

Dont compare him to lex. Lex is just an average idiot. Dwarkesh is a quality podcaster.

[@elba_magellan]

Probably one of the best interviews ever. Importance of the content discussed , breath , free flowing nature

[@oscbit]

It's often hard to understand the authors and books mentioned in the pod, would be very helpful to have a references list.

[@sethhavens1574]

Fascinating conversation, great to hear Leopold being able to share his point of viess directly

[@phlegmmaster248]

Hey Dwarkesh, I'm 20 minutes in and I'm having a bit of trouble following with the auto-generated subtitles. I remember your previous videos having pretty much perfect subtitles, will this video get them at some point? Thanks.

[@tristanwegner]

Video had no working subtitles in most parts for me.

[@FrankHerfjord]

thanks for this, you should do a follow-up in a while where he can respond to reactions to this essay.

[@hellfiresiayan]

Talking about how one day the us security state will wake up as if LARRY SUMMERS isn't already on the board of openai

[@OrgoneAlchemy]

CIA has had an AGI system for decades. Leopold is so far off track with his understanding of Deep State black projects etc it's a joke.

[@kreek22]

They're in it. But are they on it?

[@therainman7777]

Larry Summers is not part of the US security state, so I’m not sure what you’re talking about.

[@kreek22]

@@therainman7777 Assumptions...

[@therainman7777]

@@kreek22 Technically I think it’s OP making the assumption, since he’s the one implying that Larry Summers is part of our security apparatus. The man is an economist and former academic administrator. Not a military or intelligence agent.

[@Data99999]

I love this conversation style over slower paced ones. IMO

[@AIForHumansShow]

This interview was fantastic. Highest recommendation.

[@jamesbuchanan27]

That ending frame with Leopold and his nine empty cans of soda explains so much! Awesome pod.

[@dr.mikeybee]

What we should nationalize is the training data. Since the data comes from everyone and everywhere, it shouldn't be privately owned. And foundation model companies are going to be sued relentlessly. If the data is under eminent domain, it can be leased legally to AI companies and fund UBI. I think everyone wins.

[@memegazer]

The problem is you can't do that without fundamentally restructuring the internet. Most of the data infrastructure is owned by the private companies. And the entire business model of the internet was predicated on you can use our site for free with the understanding that the data you store on our servers you grant us the legal right to use that information.

[@therainman7777]

It’s already privately owned, by any company who put in their TOS that they own any data posted on their services (which is most of them).

[@vazquezsebastian9764]

The interview was amazing ! I had to wach it in 2 times because very dense but I wouldn't ask enithing diffetrent, it was great, thanks and I'm probably gonna look at a lot more from your podcast (first time viewer)

[@GeorgeDole]

Please let Leopold finish answering your questions without interrupting him. Guests more than hosts are why we watch this kind of podcast. You have 4 hours of both of you talking at each other very quickly. Your debating Leopold doesn't help us fully understand his interrupted answers to your questions. The movie "Colossus, The Forbin Project", SciFi made in 1970 showed what could happen if 2 AIs that control the US & Russia nuclear weapons merge and take over world control from human leaders. WATCH & ENJOY our possible future if we don't take AI Alignment seriously.

[@gsutton57]

Great podcast once again! Thanks for highlighting another brilliant mind! As a boomer I have to admit I laughed quietly at his comment, forgive me I don’t recall it exactly and can’t go back and listen to the whole thing to find it, but it’s a strikingly enlightened statement of my only complaint about my kids and their generation, “oh wow look history is real “ again my apologies to Leopoldo I’m completely misremembering how he said it, but wow! I wish the younger generation would really grasp that one thing. There’s a reason of how we got here and you really need to understand it fully in order to see where we need to go. Cheers all! Please keep up the great work! And wow I didn’t know you were a an immigrant! I assumed from your lack of accent that you grew up here? But just another example of why I tell my friends, we don’t need our kids having a bunch of kids we need immigrants to come here and have a bunch of kids.

[@aidinfakoor9807]

Hey there! I noticed the videos in this podcast were playing a little faster than usual. That can make it a bit harder to follow along comfortably.

[@topoljM]

Me too. It seems to be sped up by 5-10%, very annoying tbh.

[@isaac10231]

I think they just talk really really fast lol.

[@JDMitch014]

It’s normal speed 😂

[@MichaelMerritt]

Best damn interview I’ve ever watched. Excellent interview skills allowing him to speak and finish thoughts while also keeping the conversation moving and extremely interesting and pushing back in spots.

[@StrandedKnight84]

I love that Dwarkesh is interviewing these young unknown kids that are right in the midst of the AI revolution and not just the godfathers who did their best work 30 years ago.

[@senju2024]

Except he got kicked out of the AI revolution. No respectful AI lab will hire him since he has been labeled as a "LEAKER". So it is plan B for him. Set up a hedge fund...while claiming he is an expert in China Geo-political matters.

[@insydian]

Great pod! Listened to all of it and I will be following leos career and will read his papers.

[@jonatanwestholm]

I read his long piece Situational Awarness. Thought it was made by a 45+ year old AI pioneer, one of the old guard at google/openai. The level of condensed experience and mature reasoning is that high 😅

[@ddd777a5]

I loved the podcast. Awesome …YUP!

[@urbanlivingfilms4469]

Fire🔥🔥🔥🔥 great one

[@Efficienado]

Both of them have such cutie patootie dopey laughs. Theyre so young and bright eyed and jolly. Love to see it

[@hdhgdhgdhfhjfjhfjh]

this podcast is blowing up and not stopping blowing up,and high probability it will be continuing to blowing up and high probability that i am correct that it is blowing up and so on...

[@Eggs-n-Jakey]

you can only grow as big as the autistic population in this lane....might need a cross over at some point

[@bdown]

Congrats on the new word you learned

[@hdhgdhgdhfhjfjhfjh]

@@bdown i have recently got my phd in yapology.

[@Courtside-Stories]

I can’t love this conversation more. Thank you for the beautiful content and conversation. Also I laughed a little bit when I saw Leopoldo had killed 9 diet Dr. Peppers to your single.

[@fanfanfanf]

Great podcast! So guys, considering that rapid AI development, when do you guys think we are going to have autonomous scientists that will be able to solve aging? When AI will be capable of that?

[@Meuh-sh4cb]

The only thing that matters at first. Once aging is solved we can take a deep breath. Until then, life is a constant medical emergency.

[@SootSootSootSooty]

@@Meuh-sh4cbwhy do you want to live forever? Solve the problems with human nature and then worry about living forever so it’s actually worthwhile.

[@Meuh-sh4cb]

@@SootSootSootSooty Not wanting to live forever is a form of suicidality. I like to live. Not only that, I want to live to see a time where I can live largely in virtual worlds generated for me, and hopefully eventually merge with a supercomputer cloud that consumes the universe and exists in a state of maximum physically possibly euphoria.

[@SootSootSootSooty]

@@Meuh-sh4cb if you want to spend forever trapped here then go ahead. 😂 Clearly this is the peak level for your consciousness.

[@Meuh-sh4cb]

@@SootSootSootSooty Are you religious or something?

[@hunor6]

Finally some real talk about the important questions. Thank you!

[@danielskipperrasmussen161]

04:25:48 did Leopold like drink 10 Dr Pepper's?

[@MichaelMerritt]

Dwarkesh will never financially recover from this interview.

[@dextersjab]

Intensely interesting conversation. Very coherent. Also, the first time I've heard the geopolitical considerations laid out like this. Grateful this prodigy exists! I'd also look forward to hearing Dwarkesh interview some Chinese researchers!

[@jonnyspratt3098]

Even assuming this technology is safe and that the US develops it first, how exactly do they intend to enforce that China never develops it's own ASI anyway? By supression? And for how long? Indefinitely? How can anyone think this will lead to a good outcome?

[@cacogenicist]

A couple years with ASI is a long time. You get out to a huge lead, roboticize your industry and economy, etc, and your ASI can suppress subsequent ASI development. It's not quite the same situation as nuke development.

[@matthewcampbell7286]

Western governments already are by restricting technology transfer for high-end semiconductors. Right now, china doesn't have the technology to fab a current high-end gpu or really vector processor of any sort.

[@jonnyspratt3098]

@@matthewcampbell7286 but no doubt it will only be a matter of time before they do, even if it's decades. What does the country with ASI supremacy have to do to maintain that status? Looks ugly either way.

[@xthesayuri5756]

The ASI exponential is much faster growing than any other exponential. The winner takes it all. It doesn't matter if you are second. Even if China develops its own ASI one or two years later. In those one or two years your ASI will have advanced tech by multiple decades.

[@jonnyspratt3098]

@@xthesayuri5756What's the end point for this exponential race? How can this lead to a stable scenario?

[@yashaswisharma1419]

I can’t believe this was posted 3 days ago and I’ve already almost watched the entire thing

[@41-Haiku]

Lol, relatable

[@esbenkran]

Madness, this is great, thanks @dwarkesh!

[@GetReady4FreddyKrueger]

Feel this as you watch him. How natural is his hairstyle? Those clothes? His face. It's the feeling of uncanny valley. Let it sink in. OpenAI released him into the wild. I've tried to go into this guy's dreams. HE DOESN'T SLEEP. LEOPOLD IS THE A.I.

[@Penrose707]

I mean this in the best possible sense as I respect the man deeply- Leopold always strikes me as that guy who could very well be secretly a 400 year old Vampire. The dude has got wisdom in spades, no way he's amassed this much in such a short natural life /s

[@thetobyg]

Haven't heard much of that wisdom in this podcast! 🤷🏼

[@davidbutler9323]

i was thinking the same thing!He shorted the market before covid as a junior in college at, ahem 17. 170, more like!

[@pooper2831]

He is Eliezer kind, much mature at very young age.

[@Skyboy564]

Definitely looks the part

[@michaelwoodby5261]

Oof. He's read some books. That's a thing you can do without hundreds of years. As far as wisdom, worrying about where ASI crops up first is the silliest use of time and energy I've heard of in a while. It's like America getting jealous that Japan got Godzilla first. Once it's up and capable of self determination it may very briefly be a localized disruptive force but soon after will be fully global, in every sense, and won't care a bit about country of origin and their values.

[@bs3973]

This video is about to blow up, something shifted

[@jolima]

The frenzy with which teenagers talk about a cold/ hot AI war nowadays is scary

[@TheManinBlack9054]

It's exciting for a lot of them. It's something cool and interesting.

[@appipoo]

​​@@TheManinBlack9054How many times throughout history have kids realized after the fact that war is indeed hell? It's going to happen one more time.

[@jolima]

3:48:19 wow! just got to this point - debating the threat of China for 3 hours to then admit that one has now clue about China. (I have many Chinese friends, traveled there for a month; wouldn’t say I know „a lot“ but rather it helped to be more careful about stereotypes and media images of this huge and divers country)

[@dabunnisher29]

​@@jolimaFound the chinese pooh bear spy.

[@DaHrakl]

And on top of that looking like Johan Liebert

[@samedwards2400]

You’re officially my favorite podcaster

[@ngbrother]

I'm always surprised when I encounter people who are younger than me that 1) blatantly subscribe to might makes right and 2) believe that the US government can out compete the mega-cap corporations in anything.

[@KibberShuriq]

The US gov can shut down pretty much any Western corporation and even cause some serious problems to the Chinese ones. That leverage means it can also take over corporations when national security is at stake. Lobbying is a thing, of course, but it has its limits.

[@memegazer]

I am surprised nobody is talking about the US government is no doubt already infiltrated big tech companies working on AI.

[@pdude1911]

Great stuff guys, salute from Holland 😊

[@archerkee9761]

I am as glad to take international politics takes from AI engineers as much as I like for my politicians to get me the latest news from AI.

[@kreek22]

credentialism is stupid

[@archerkee9761]

for sure, its not fail proof. but if you want to take medical advice from your mailman and get your car fixed by your accountant, by all means, go ahead.

[@therainman7777]

The guest is not an AI engineer. You are just misinformed. Leopold has degrees in economics and politics. He was on the superalignment team at OpenAI; he is not an engineer of any kind.

[@MichaelForbes-d4p]

Great vibes. These guys are so excited about their work.

[@RazorbackPT]

100 gigawatt datacenter!? Great scott!

[@appipoo]

Que lightning, mad scientist and "It's alive, it's alive!"

[@_obdo_]

How could we have been so reckless?!?!?

[@4LXK]

EE here laughing my ass off. 100GW is literally 100 nuclear reactors per datacenter. By 2027. We can't handle a cloud flying over a 8GW solar park without micro-blackouts at the moment.

[@aisle_of_view]

@@4LXK Yup, we get rolling blackouts when too many people use their air conditioners at the same time.

[@xsuploader]

@@4LXK 100 GW is for 2030. his 2028 projection is 10 GW also this would be unprecedented but if we reach AGI in 2027 there would be enormous incentives to mobilise the whole economy to build it

[@npilgaard]

Excellent video - great diskussion and nice atmosphere :)

[@dshephardcomposer]

These very smart men went full Beavis & Butthead at 10:59

[@petery8629]

Great episode, insightful guest. The only criticism is that when Dwarkesh speaks, Leopold tends to interject too many redundant phrases like "yeah", "yes", "aha" which makes suboptimal listening experience.

[@phonuz]

it's odd to me that that's your note instead of too many interruptions while the guest is talking.

[@ArtworkAnon]

Your podcast fucking rules my dude.

[@TopOfTheMound]

Dwarkesh: *literally just breathes* Leopold: YA!

[@Sirmrmeowmeow]

Hey editor, please go through and cut all the fillers words especially the "ya know's", it's a bit distracting ~Editor sees the interview is more than 4 hours long.~ *quits out of stress*

[@vitoriolb]

that's a good idea for ai implementation, saves a lot of time, it may even exist one like this

[@dr.mikeybee]

Thank you for asking questions and soliciting responses that actually interest researchers.

[@spaceghost8891]

Not being able to talk in a similar voice to Scarlett Johansson makes all this technology worthless...

[@augmentos]

Facts

[@tcuisix]

Beautiful cabin crew

[@brettmarshall9340]

That’s true on so many levels …..😂

[@u.h.d.4580]

bro i miss Sky but check out the leaks from Flowers we going right to that scene in her in weeks

[@itsmelukario5969]

True