Guillaume Verdon: Beff Jezos, E/acc Movement, Physics, Computation & AGI | Lex Fridman Podcast #407

Introduction

• Guillaume Verdon is a physicist, applied mathematician, and Quantum Machine Learning researcher and engineer.
• The duality of Guillaume's identity includes his work in physics and Quantum Computing as well as his creation of the effective accelerationism movement as 'Bef Jezos on X'.
• The movement promotes advancing rapid technological progress as the ethically optimal course of action for humanity, emphasizing AI as a great social equalizer.

Beff Jezos

• Guillaume's main identity is as a physicist seeking to understand the universe through theoretical physics and information theory.
• He transitioned to studying Quantum Computing in order to build and apply computers inspired by nature to simulate and understand quantum mechanical data in AI representations.
• Guillaume felt a need to explore the middle ground between the quantum and cosmic scales, which led to a focus on thermodynamics as a governing principle at mesoscales.
• The creation of the anonymous account 'Bef Jezos on X' allowed Guillaume to experiment with ideas freely, leading to the emergence of the effective accelerationism movement advocating for maintaining variance and healthy competition as the path forward for adaptive systems.

Thermodynamics

• Guillaume discusses thermodynamics as governing the mesoscale and as the missing piece of understanding between the quantum and cosmos.
• He emphasizes the role of thermodynamics in the emergence of life, its capability to use energy, and the creation of pockets of non-entropy that tend towards complexity.
• Guillaume integrates his work in physics and AI representations with his exploration of philosophical thoughts and the responsibility to engineer a movement of viral optimism about the future.
• He emphasizes the detrimental impact of pessimism on the trajectory of civilization and advocates for a belief in agency to make the future better and grander statistically.

Doxxing (18m36s)

• Manifesto written anonymously with anonymous co-founder
• Founding a company with secrecy
• Journalists revealing identity without consent
• Concerns about privacy, freedom of speech, and freedom of thought
• Legal and ethical considerations of doxing
• Importance of anonymity for marketplace of ideas

Anonymous bots (28m30s)

• Potential threat of armies of bots and anonymous accounts
• Freedom of thought and speech for AI as well as biological beings
• Need for a way to certify humans and identify bots
• Concerns regarding amplification of ideologies through adversarial techniques
• Desire for fault-tolerant progress in AI development

Power (35m58s)

• Consideration of dangers and risks in AI development
• Liability and market forces as drivers for safe AI development
• Preference for reliability engineering driven by market forces over heavy-handed regulations

AI dangers

• Recommendations from a report titled "Managing AI Risk in an Era of Rapid Progress" include giving independent auditors access to AI labs, allocating one-third of AI research funding to AI safety, requiring AI companies to adopt safety measures, and making tech companies liable for harmful AI systems.
• The speaker believes that setting an arbitrary percentage for funding AI safety is unnecessary, as organizations would allocate budget as needed for reliability.
• Concerns are raised about a potential government-backed AI cartel forming if big players and the government become too closely aligned.

Building AGI

• The speaker does not believe in a fast takeoff of AGI, viewing the development as an exponential rather than a hyperbolic singularity.
• Emphasizes the importance of open-source AI to maintain equilibrium of capabilities and avoid a scenario where a market leader dominates everything.

Merging with AI

• The speaker argues for the broader exploration of different forms of intelligence beyond just human-level AI, and stresses the importance of separating AI and state to maintain America's advantage in free market capitalism.
• Discusses the potential benefits of AI augmenting human capabilities and the coexistence of AI not anchored to anything biological with humanity.

Concerns and responsibilities

• The speaker shares a perspective on AI's potential to carry the flame of consciousness in the universe and the responsibility to scale civilization to preserve it.
• Discusses the potential for unintended consequences with AI systems and the need for responsible and informed decision-making about the future of AI technologies.

Market caution

• The market is seen as a mechanism for exhibiting caution, as every organism, company, and consumer acts out of self-interest, avoiding negative utility and promoting technologies with positive utility.

p(doom) (57m56s)

• Probability of Doom (PDO) is not a reliable calculation for predicting the likelihood of AI killing humans
• Estimators of PDO are biased due to human evolution and neuroticism
• The future is unpredictable due to chaos and uncertainty in complex systems
• The concentration of AI power in the hands of a few could lead to dystopian scenarios
• The fear of AI Doom can lead to preemptive actions that inhibit positive futures
• AGI's potential to centralize power raises concerns about authoritarian control and potential misuse of power
• Centralization of AI is currently biased due to compute density and data centralization
• Smaller subsystems and hierarchies are more fault-tolerant and may lead to a correction in the centralization of AI

Centralization of AI

• Over centralization of AI is driven by compute density and data centralization
• Increasing compute density and decentralized, distributed testing of hypotheses may lead to a correction in the centralization of AI
• Hierarchical control and fault tolerance are crucial for controlling complex systems

Hierarchical Control and Competition

• Hierarchical control and competition in the AI space may prevent a monopoly on AGI and prevent misuse of power
• Competition in the AI space is essential for maintaining dynamism and avoiding tyranny

Quantum machine learning (1h13m23s)

• Universe can perform quantum computation if all degrees of freedom are accessible
• Intelligence involves perceiving, predicting, and controlling the world, often involving compression of information
• Quantum machine learning involves learning compressed representations of quantum information to understand and process the world
• Problems facing civilization, such as climate change, are influenced by quantum mechanics, making quantum AI important for addressing them

What is a quantum computer and quantum machine learning? (1h13m23s)

• Quantum computer is a quantum system under control that maintains its quantum state
• Quantum machine learning provides a way to learn quantum mechanical representations of nature, staying true to its quantum aspects
• Attempting classical machine learning on a quantum computer is not advised due to high computational cost
• Quantum deep learning allows incorporation of quantum mechanical systems into the learning process for potential advantages in capturing quantum correlations

Universal training algorithm for Quantum deep learning (1h20m33s)

• Paper introduces a training algorithm for quantum deep learning that involves backpropagation with a quantum twist
• Utilizes reversible operations and phase kicks to adjust parameters in a quantum system to emulate particle dynamics
• Theoretical advantage lies in the potential to find new optima difficult for stochastic optimizers, but current quantum computer architectures may make such algorithms expensive to run

Quantum computer (1h26m41s)

• Teams for quantum computing require interdisciplinary expertise
• Physics applies mathematics to understand the world, while engineering applies physics knowledge to practical applications
• Engineering limits are challenging in quantum computing
• Building quantum computers is difficult due to the need for high-quality components and control over quantum bits
• Threshold for scaling up quantum computers requires sufficient quality in control over quantum bits
• Progress in quantum computing is slower than media hype suggests
• Research involved in exploring quantum energy teleportation via qit probes

Aliens (1h35m15s)

• Considering the existence of intelligent alien civilizations can instill a sense of urgency in technological development
• Developing robust capabilities is important to ensure resilience against potential disruptions from aliens or AI
• Thought experiment on perceiving intelligence extends beyond human constructs to quantum systems
• Research and technology like quantum sensors can enhance our ability to sense potential alien intelligence
• Alien life could take various forms, and the notion of intelligence and life should be expanded beyond anthropocentric or biocentric perspectives

Quantum gravity (1h40m4s)

• Journey to understand quantum gravity through hologram of lesser dimension being dual to bulk theory of quantum gravity
• Applying quantum machine learning to learn emergent geometry from boundary theory
• Black holes serve as interface between quantum mechanics and gravity
• Black hole information paradox and firewall paradox
• Speculation about black holes and possibilities for transmitting information to new universes

Kardashev scale (1h45m25s)

• Kardashev scale measures energy production and consumption
• Goal of the E/acc movement is to ascend the Kardashev scale and increase energy harnessing
• Extension of life beyond Earth and responsibility to do so
• E/acc movement aims to become self-aware and engineer its own growth

Effective accelerationism (e/acc) (1h47m17s)

• The goal of the E/acc movement is for the human techn capital mimetic machine to become self-aware and to hypers engineer its own growth
• E/acc movement seeks to grow and lean into natural tendencies of the system to adapt for its growth
• Balancing constraints and entropic force for optimal exploration and adaptation
• E/acc promotes freedom of thought, speech, and AI research
• Many forks and clusters in E/acc movement, encouraging diversity of thought
• Humor, meta-irony, and playfulness in the E/acc movement to promote independence of thought and prevent mode collapse

Humor and memes

• Initially used humor and memes to camouflage deep truths within packages of humor and posts to grow under the radar.
• Purpose was to camouflage against those seeking status and those not wanting to take themselves seriously.
• Engineered information for optimal spread through algorithmically amplified landscapes, leading to rapid growth.
• Currently considering serious debates and long-form discussions due to the need for serious thoughtful discussions amid algorithmic rewards for polarization.

Jeff Bezos

• Acknowledges Jeff Bezos as a great entrepreneur who leveraged technology, capital, and accelerated to provide quick delivery, convenience, and low prices through Amazon.
• Grateful for Amazon's investments in robotics, which have contributed to the modern tech boom and paid salaries for many.
• Supports the efforts of entrepreneurs like Jeff in capital allocation and building multi-planetary species, indicating a preference for capitalism over giving money to the government.

Elon Musk

• Sees Elon Musk as an interesting case and celebrates him, but notes his concerns about the dangers of artificial intelligence and the potential need for regulations.
• Considers Elon's efforts with Neuralink as a hedge against AI progress, aiming to create AI aligned with humans and reliable through XAI.
• Opposes additional regulations hindering the open-source ecosystem's contribution to AI and owning a piece of the future, emphasizing maintaining the ubiquity of opportunity.

AI Safety

• Expresses cautiousness about AI safety being used as a cover for centralized control, which could lead to eventual corruption and a 1984-like scenario.
• Affirms that AI safety is important but raises concerns about excessive centralization and the potential for a global panopticon.
• Emphasizes the need to be careful to avoid situations that did not lead to good outcomes historically.

Extropic

• A company recently announced a 14.1 million seed round to build a physics-based Computing system and AI algorithms inspired by out of equilibrium thermodynamics.
• The company aims to engineer the ultimate physical substrate for generative AI from first principles of physics, information theory, computation, and thermodynamics.
• The goal is to create a full stack system, including hardware and software, leveraging physics-based AI algorithms.

Singularity and AGI

• The company aims to run anthropomorphic AI and physics-based AI in combination, believing this joint system to be the closest thing to a fully General artificially intelligent system.
• Physics-based AI is better at representing the world at all scales, while anthropomorphic AI provides a humanlike interface.

AI doomers

• The differentiation between AI doomers and eakers, or those who are very likely and unlikely about AI doom, is proposed, and the discussion notes that this movement is not right-wing or left-wing but rather aligns with the debate on tech progressivism versus tech conservativism.

Effective altruism (2h27m54s)

• Effective altruism aims to do good optimally by minimizing suffering in the world
• The movement measures good in terms of units of Hedonism, striving to minimize suffering but may lead to suboptimal outcomes
• Proposed an alternative approach to measuring progress of civilization in terms of physical energy
• Human tendency to fill gaps of uncertainty with narratives may affect interpretation and implementation of loss functions
• Skepticism towards reducing human civilization to an equation is important
• Capitalism and power dynamics within an organization may corrupt the initial community and goals of effective altruism

Day in the life (2h34m23s)

• A perfectly productive day for Guillaume Verdon involves meetings, internal and external, in the early afternoon
• He consumes ketones and Red Bull to enhance clarity of thought and focus
• A meal is consumed after the initial part of the day to maintain extreme focus
• Deep work is done from 9:00 p.m. to 4:00 a.m., contributing to technical patents and working with engineers
• His workday was inspired by a similar routine of Demis Aabis
• In the past, he played American football and was into powerlifting while studying mathematics in grad school
• Passion and curiosity are advised to be followed to increase learning rate and becoming smarter quickly
• Motivation is a catalyst for learning, often tricking oneself into caring about a subject

Identity (2h40m50s)

• Over time, the two identities of Guillaume Verdon and his character 'Bef Jezos' have started to merge
• Emphasizes the value of exploring different identities by having alternate accounts to experiment with thoughts in a low-stakes environment
• Platforms that support staying anonymous or using anonymous accounts are important for sharing thoughts that aren't fully formed
• Encourages rigorous empathy, supporting a thought experiment to understand different viewpoints
• Acknowledges the danger of being consumed by an alternate identity and the need to be careful

Advice for young people

• Suggests studying fundamental mathematics and physics for a strong foundation in a rapidly changing technological landscape
• Emphasizes the importance of knowledge about complex systems and adaptive systems

Mortality

• Views death as an important aspect of dissipative adaptation and maintaining novelty and adaptation in systems
• Supports the idea of extending lifespan for a longer period of training due to increasing complexity in the world
• Not afraid of death and believes in the potential for better versions or forks of oneself in the future