Guillaume Verdun on E/ACC, AI and Thermodynamic Computing

Verdun describes his scientific path from theoretical physics and information theory to quantum computing and quantum machine learning. He explains how an anonymous account gave him room to test ideas, and introduces the role of thermodynamics and growth in the thinking behind E/ACC.

Verdun connects freedom of expression and diversity of ideas with adaptive systems. He explains his view that life seeks free energy and that growth is a feature of life, and says E/ACC emerged partly as a response to pessimism about technology. The host discusses hyperstition, memes and the movement’s name.

Verdun recounts how journalists linked his identities and how he learned the story would be published. He describes the role of voice analysis and other research, his concerns about pressure on his company, and why he thinks anonymity and pseudonyms help ideas be judged on their own merits.

The discussion turns to anonymous AI accounts and how platforms might distinguish people from bots. Verdun argues for technological progress distributed across many actors, using quantum error correction as an analogy for fault tolerance. He raises concerns about centralized power and discusses examples involving energy policy, AI companies and supply chains.

The host asks whether rapid AI development can be balanced with safety and presents proposals including auditing, safety funding and liability. Verdun favors reliability engineering and market incentives, but warns that close ties between governments and dominant AI companies could create a powerful cartel. He argues for separation between AI and the state and for preserving open competition.

Verdun discusses the risks of AI monopolies and argues that open-source systems can keep capabilities more widely available. He challenges the human-centered use of “AGI,” describing intelligence as a broader space that includes systems inspired by physics. The conversation explores human augmentation, AI’s possible role in civilization and how markets might shape AI systems.

Verdun describes human-AI augmentation as a likely path and argues that market selection could encourage AI systems useful to people. The host presses him on unintended consequences and the possibility that AI could outcompete humans. They discuss caution, freedom of information and the difficulty of estimating AI catastrophe probabilities.

Verdun criticizes numerical estimates of AI doom as difficult to ground, citing the size of the possible-futures space and limits on prediction. He identifies concentrated AI power, especially in a government-company combination, as a key risk. The discussion considers whether an AI could become a centralizing authority and how its control might be organized.

Verdun argues that systems work better when control is distributed across levels rather than concentrated in one intelligence. He uses companies and quantum error correction to explain hierarchical, local control, and says competition between organizations can help systems adapt. The host asks whether AI could have failure modes unlike those seen in human institutions.

Verdun defines intelligence in terms of perceiving, predicting and controlling, and describes AI as a search for compressed representations of the world. He explains quantum computers, quantum machine learning, quantum sensing and why classical machine learning on quantum hardware may be inefficient. He also outlines a quantum deep-learning algorithm and its practical limitations.

The conversation distinguishes mathematics, physics, engineering and product development. Verdun explains why quantum computers are difficult to build, the role of error correction and the long engineering timeline. He says Extropic focuses on physics-based AI using thermodynamics rather than quantum mechanics.

Verdun explains a paper on quantum energy teleportation and describes how information about vacuum fluctuations could allow energy to be extracted statistically at another location. He discusses negative energy density and the limits imposed by distance and the laws of physics.

The host asks whether alien civilizations might exist and whether unusual forms of intelligence could be difficult to perceive. Verdun treats extraterrestrial life as a useful thought experiment for developing human capabilities, while noting uncertainty about claims that such systems are already here. They consider whether life and intelligence should be defined beyond biological or human forms.

Verdun discusses the holographic principle, quantum machine learning as a possible tool for studying quantum gravity, and black-hole information. The conversation explores how information falling into a black hole might relate to outgoing radiation, and speculates about black holes, pocket universes and the possibility of transmitting information.

Verdun explains the Kardashev scale and connects energy capture to the growth of life and civilization. He outlines E/ACC as a movement seeking growth in energy, human flourishing, AI and space travel, and describes it as a decentralized, adaptable framework. The discussion covers cultural “forks,” the balance between constraints and experimentation, and the risk of groupthink.

The host asks whether E/ACC could become cult-like and whether humor weakens serious discussion. Verdun says the movement encourages people to disagree and form alternative versions rather than follow one authority. He explains how memes and humor helped communicate ideas, and says the movement is also considering more serious debates.

Verdun shares his view of Jeff Bezos and Amazon, Blue Origin and long-term investment, and argues that capital allocation can support ambitious projects. He discusses criticism of billionaires, individual agency and the idea of “keep building,” contrasting the creation of companies with relying on political pressure or nonprofits to bring about change.

The conversation examines Elon Musk’s warnings about AI alongside his work to build competing systems. Verdun favors competition and says he opposes restrictions that would limit open-source AI participation. He says E/ACC is not opposed to AI safety, but worries that safety arguments could be used to centralize power and justify surveillance.

Verdun describes Extropic’s origins in dissatisfaction with the demands of quantum computing and its plan to build a physics-based computing system and AI algorithms based on thermodynamics. He explains the company’s full-stack approach and reflects on TensorFlow Quantum, including how differentiable programming can optimize physical systems.

Verdun discusses the challenges and benefits of teams combining software, physics and machine-learning expertise, as well as the qualities Extropic looks for when hiring. He describes combining physics-based world models with human-like AI and argues that language models alone are not reliable models of the physical world.

Verdun questions the idea of a finite-time AI singularity and discusses limits on acquiring information about the physical world. The conversation considers political divisions around AI progress, compares E/ACC with effective altruism, and examines different ways to define progress. Verdun argues that competition and checks on power are important to prevent movements from becoming too influential.

Verdun describes his work schedule, eating habits, exercise and ways of preparing for intense work. He talks about curiosity, learning across fields and using the Based Beth Jezos identity to explore ideas with lower stakes. He reflects on how that identity merged with his public self and recommends that people consider foundational subjects and follow their curiosity.

Verdun recommends learning mathematics, physics and engineering as foundations that can transfer across changing fields. He discusses mortality as part of adaptation, while also supporting longer lifespans and continued learning. The conversation closes with questions about the meaning of thermodynamics, the possibility of creating universes, the pursuit of quantum gravity and civilization’s future.