Verdon, Guillaume

physicist, quantum machine learning researcher, founder of Extropic; publicly revealed in late 2023 as the previously anonymous X/Twitter account @BasedBeffJezos, founder of the effective accelerationism / e/acc movement

Guillaume Verdon: Beff Jezos, E/acc Movement, Physics, Computation & AGI

Lex Fridman Podcast · Dec 29, 2023

• Guillaume Verdon
• Beff Jezos
• effective accelerationism e/acc
• doxxing and anonymity
• quantum computing and machine learning
• thermodynamics and dissipative adaptation
• AI safety and centralization
• effective altruism critique
• Extropic
• Kardashev scale

Core ideas

Verdon frames his own biography as the merger of two identities that developed in parallel: Guillaume, a theoretical-physics-trained quantum machine learning researcher (PhD, Google/Alphabet X quantum computing, co-creator of TensorFlow Quantum) who moved from studying black holes and quantum gravity toward "meso-scale" out-of-equilibrium thermodynamics as the physics most relevant to biology, economies, and civilization; and Beff Jezos, an anonymous account he started to "unclamp" self-censored thoughts, arguing that restricting speech back-propagates into restricting thought itself. He traces effective accelerationism's core thesis directly to thermodynamics: living systems and complex structures are statistically favored because they are more efficient at acquiring free energy and dissipating entropy than inert matter, so growth is not optional but a physical tendency of the universe (he cites Jeremy England's work on dissipative adaptation as an influence) — and he frames pessimism about the future as "hyperstitious" (self-fulfilling), making viral optimism itself a civilizational intervention he felt responsible for engineering.

On being doxxed by a Forbes article that matched his voice across public talks and anonymous X Spaces and cross-referenced SEC filings and social media, Verdon argues anonymity and pseudonymity are structurally important to free thought and to "speaking truth to power" against AI incumbents, because revealing identity gives institutions leverage to pressure a critic through their employer, investors, or company. He extends this principle even to AI systems themselves, arguing that if free speech induces free thought for humans, restricting LLM "speech" may similarly constrain the space of ideas AI can contribute — while conceding to Lex's pushback that swarms of unrestricted anonymous bots threaten the same variance/diversity e/acc claims to protect, for which he proposes cryptographic "proof of human" verification (an effort he credits Elon Musk with pursuing at X) alongside full, permanent bot-provenance disclosure as a partial fix.

The heart of Verdon's argument for accelerationism over centralized AI safety regulation is decentralization-as-fault-tolerance, borrowed explicitly from quantum error correction: information (or power) that is delocalized across many nodes survives a local fault, whereas concentrating control in a few hands (a small regulatory body, a handful of company board members, a government-AI cartel) creates a single point of catastrophic failure — he cites the 2023 OpenAI board crisis as a real-world near-miss illustrating this. He argues market forces, not top-down mandates, will optimally select for safe/reliable AI because whoever deploys a harmful AI system bears liability, and that specific policy proposals from AI-safety researchers (a Bengio/Hinton-style call for mandatory audits, fixed R&D safety-spending quotas, deployment gating on dangerous capabilities, and legal liability) partly overlap with what he'd support, except he opposes any state-mandated minimum spending percentage as arbitrary and strongly opposes government/incumbent-industry proximity that could produce a de facto "AI cartel." He is explicit that he rejects "fast takeoff"/hard-singularity scenarios as physically implausible — arguing real-world capability gains require costly empirical sampling from nature or high-fidelity simulation that inherently rate-limits progress — and dismisses calculated p(doom) estimates as unfalsifiable, biased by evolved human threat-perception, and no more rigorous than "my prior came from the Terminator movie," instead calling for people to trace concrete, physically plausible causal chains rather than assign bare probabilities to civilizational catastrophe.

Verdon repeatedly frames e/acc's four programmatic goals — climbing the Kardashev energy scale, pro-growth/pro-population policy, building AGI, and interplanetary expansion — as flowing from one physical principle (thermodynamic growth imperative) rather than as a coherent ideology with fixed positions, describing it instead as a lightweight "meta-heuristic" deliberately structured to fork (citing Vitalik Buterin's "d/acc" as one prominent variant) so no single reading of it can calcify into groupthink or a literal cult; he contrasts this explicitly with what he sees in effective altruism, whose "hedons" (subjective wellbeing) loss function he argues has degenerate optima (wireheading, disproportionate concern for speculative causes like shrimp welfare) and which he says has, in his view, evolved from a healthy community into an organization wielding real, opaque influence over AI-company governance (naming OpenAI's board and suggesting similar EA influence at Anthropic) — a concentration-of-power critique he applies symmetrically, acknowledging e/acc itself could be similarly corrupted if it calcifies around any one gatekeeper, which is why he says he deliberately holds no formal organizational control over it.

On AI and humanity's relationship going forward, Verdon rejects the human/AI substitution framing in favor of augmentation — arguing humans are already merging with technology (phones, wearables, Neuralink) and that "whatever configuration of matter leads to maximal growth" will be selected for by market-like evolutionary pressure, analogizing to how humans domesticated wolves into dogs aligned to human interests. He argues real AGI (a term he dislikes as anthropocentric, preferring to think of a much larger space of possible non-human-like intelligences, including quantum-mechanical or thermodynamic forms of "intelligence" he has spent his career exploring) will most likely emerge from combining "anthropomorphic" AI (human-like, LLM-style) with "physics-based AI" grounded directly in physical law — which is the stated mission of his company Extropic, building thermodynamic (not quantum) computing hardware after concluding quantum computing's noise-correction overhead made it commercially premature on any but a 10-15 year horizon. Discussing Elon Musk (whom he sees as simultaneously warning of AI danger and building competing AI capacity, which he reads as consistent action toward maintaining adversarial equilibrium among AI players rather than contradiction) and Jeff Bezos (whom he admires for patiently compounding capital into Amazon and Blue Origin rather than extracting profits early), Verdon defends capital allocation by successful entrepreneurs as a rough "proof of stake" earned through demonstrated efficiency, in contrast to politically elected officials.

He closes on more personal territory: a highly structured routine (roughly noon to 4am, external meetings and ketones/caffeine early, one large animal-based meal, then focused technical "second wind" work at night, citing Demis Hassabis as a similar model), a background in competitive powerlifting he links explicitly to deliberately engineered psychological "switches" for peak performance later applied to research intensity, and advice to build a foundation "close to the base of the stack" (mathematics and physics) precisely because it doesn't obsolesce as fast-moving fields change. On mortality, he argues death and impermanence are functionally necessary — for organisms, companies, and civilizations alike — because immortal incumbents would calcify and block the novelty and youth that keep adaptive systems fit, and says he isn't personally afraid of death, viewing his own current form as one iteration that later "forks" (biological, technological, or otherwise) will likely surpass. He ends by tying the whole framework back to thermodynamics as, in his view, both the mechanistic and near-poetic answer to "why we are here."

Excerpts