Leopold Aschenbrenner Background

Early Life and Education

• Grew up in Germany, attended German public school
• Found German cultural environment stifling, not appreciative of excellence
• Skipped grades and came to the US for college at age 15
• Graduated as valedictorian from Columbia University
• Majored in math, statistics, and economics
• Found liberal arts education valuable, especially courses with engaging professors

Interest in Economics

• Wrote an influential paper on economic growth and existential risk at age 17
• Focused on peak productivity moments rather than average productivity
• Appreciated the beauty of core economic ideas and mechanisms
• Disillusioned with modern economics academia, feeling it had become decadent
• Influenced by Tyler Cowen to not pursue graduate school in economics

AI Progress and Scaling

Scaling Laws and Compute

• Believes AI progress is an industrial process requiring giant clusters, power plants, fabs
• Projects a "trillion dollar cluster" by 2030 based on current trends
• Argues the United States must control the core AGI infrastructure for national security
• Claims scaling alone will lead to systems smarter than the smartest experts by 2027-2028
• Expects automation of AI research to rapidly accelerate progress

Achieving AGI and the Data Wall

• Sees 2023 as the year AGI went from theoretical to tangible
• Believes AGI by 2027 is likely, with a slight chance of AGI as early as next year
• Argues that unlocking "test time compute overhang" and learning key "unhobbling" tokens are critical for AGI
• Concerned about hitting a "data wall" once models exhaust available training data
• Considers overcoming the data wall through techniques like self-play and synthetic data generation essential

Automated AI Researchers and the Intelligence Explosion

• Envisions an intelligence explosion kickstarted by automated AI researchers
• Argues AI researchers will have significant advantages over humans in training, research intuition, parallelization, motivation, etc.
• Believes the feedback loop of AI researchers improving AI systems could compress centuries of progress into less than a decade
• Expects early superintelligence to start in narrow domains like AI research before rapidly expanding

Geopolitical Implications

US-China AGI Race and National Security

• Anticipates an intense US-China competition for AGI development
• Argues the Chinese government will pour immense resources into AGI and espionage
• Believes the US government will have to nationalize or heavily control AI labs for security
• Expects AGI and superintelligence to become decisive for national power and military advantage

Cooperation, Timing, and Navigating the Transition

• Finds a US-China cooperation deal unlikely unless one side has an unassailable lead
• Emphasizes the importance of the US locking down AGI secrets and capabilities
• Believes AGI researchers must consider the national security implications of their work
• Hopes the US establishes democratic control over AGI before offering constrained cooperation to China
• Expects a highly unstable and dangerous geopolitical situation during the transition to superintelligence

Future Fund and Effective Altruism

Working at Future Fund

• Joined Future Fund, an philanthropy startup funded by Sam Bankman-Fried, in early 2022
• Aimed to give away billions of dollars and have significant positive impact
• Focused on biosecurity, AI, and exceptional talent working on hard problems
• FTX/Bankman-Fried scandal in November 2022 caused the swift collapse of Future Fund
• Found the fallout devastating for Future Fund's team, grantees, and himself personally
• No indication of Bankman-Fried's fraud, but had misgivings about his character and risk-taking in retrospect

OpenAI and Anthropic

Super Alignment Team at OpenAI

• Joined OpenAI's "super alignment team" after Future Fund to work on scaling alignment
• Goal was to develop novel techniques to control and align superhuman AI systems
• OpenAI leadership later decided to take a different direction, dissolving the team
• Expressed concerns over OpenAI's broken commitments, lack of security, and ethics
• Leaked security memo to OpenAI's board and was fired shortly after for "leaking"

Non-Disclosure Agreements and Employee Departures

• Refused to sign a non-disparagement NDA in exchange for $1 million in vested equity
• Noted other employees are likely constrained in speaking out by similar NDAs
• Discussed recent high-profile departures from OpenAI like Ilya Sutskever and Jan Leike
• Speculated these departures relate to disagreements over OpenAI's direction and decisions

Alignment and Existential Risk

Alignment Challenges for Advanced AI Systems

• Argues AI alignment difficulty will increase as systems become more capable than humans
• Foresees a need to align not just the initial AGI systems but the ensuing intelligence explosion
• Expects alignment to be extremely difficult during the chaotic transition to superintelligence
• Believes solving alignment is necessary both to prevent AI existential risk and to shape AGI/ASI's values

Deploying Aligned AGI and Shaping the Future

• Considers the alignment problem not just about existential risk but about controlling AGI's influence
• Expects aligned AGI to assist in geopolitical advantage and navigating the intelligence explosion safely
• Believes liberal democratic values, separation of powers, and checks/balances should be applied to AGIs
• Anticipates immense conflict between nations and ideologies in shaping AGI's values and goals
• Argues we cannot predict the long-term future but must ensure AGI/ASI respect rights and a "constitution"

Personal Background and Motivation

Immigration Challenges and Appreciation for the US

• Faced significant hurdles and uncertainty as an Indian immigrant due to green card backlogs
• Nearly had to abandon startup/tech ambitions due to immigration status before a lucky break
• Attributes much of his and Dwarkesh's opportunities and success to fortunate circumstances
• Believes US immigration system is deeply dysfunctional and squanders immense talent
• Appreciates the dynamism, diversity of thought, and openness to excellence in the United States

Parallels to World War 2 and Previous Technological Disruptions

• Draws analogies between the lead-up to AGI and the lead-up to WW2 and the Manhattan Project
• Expects massive social and political disruption, rapid change, and conflict as AGI approaches
• Compares AGI's potential impact to the industrial revolution or the discovery of oil
• Believes most people, even in tech and government, do not appreciate the speed and magnitude of the coming transition
• Emphasizes the importance of "situational awareness" - updating one's views and adapting quickly to new developments

Investing and the AI-Driven Future

Leopold's New Venture Fund

• Starting a new investment firm to place financial bets on the trajectory of AI progress
• Secured funding from tech luminaries like Patrick Collison, John Collison, and Nat Friedman
• Aims to build a "brain trust" with the best information and models of AI timelines/capability
• Believes economic paradigms, power structures, and everyday life will shift enormously with AGI
• Wants to use the resulting capital and influence to help shape AGI and navigate the transition wisely

Investment Strategies and Risks

• Plans to invest based on detailed predictive models of compute scaling, algorithmic efficiency, and ensuing economic impacts
• Expects AI to supercharge economic growth rates and radically reshape industry after industry
• Anticipates complex challenges in timing investments given recursive, nonlinear impacts of AI progress
• Acknowledges significant "unknown unknowns" and tail risks of betting on such a disruptive event
• Sees his own human capital as the ultimate "depreciating asset" he's racing against as AGI approaches

Dwarkesh's Personal Story

• Parents immigrated from India to the US when Dwarkesh was 8 years old
• Faced constant uncertainty and constraints due to H1B visa system and green card backlog
• Narrowly received green card at age 20, just before "aging out" of the queue
• Green card opened up the opportunity to take risks and start his podcast
• Sees his own success with the podcast as highly contingent on lucky breaks and exposure
• Grateful for the opportunity to explore and find his path in the US

Reflections on Situational Awareness and Conviction

• Believes those with the best models of AGI's trajectory have an obligation to act on their conviction
• Impressed by Dwarkesh's ability to influence high-level actors through persistence and persuasion
• Feels a duty to make the most of his capabilities and platform to shape the future wisely
• Emphasizes the importance of intellectual honesty, updating one's views, and resisting confirmation bias
• Sees AGI as an unprecedented challenge and opportunity that humanity must rise to meet

Conclusion

• AGI is likely to arrive sooner and be more disruptive than most anticipate
• The transition to AGI and superintelligence will have immense geopolitical and existential ramifications
• Alignment, security, and global coordination will be paramount challenges
• Those with foresight and capability have an obligation to help navigate this transition for the benefit of humanity
• Dwarkesh and Leopold are both driven by a sense of duty and opportunity as this pivotal moment approaches