Latest Interviews

The better AI gets, the smaller its share of the economy might get – Alex Imas and Phil Trammell

Alex Imas, Phil Trammell

Economists and technologists discuss a post-AGI future where scarcity concentrates in the "relational sector" as automation drives capital accumulation, creating a complex transition where historical precedents like the Industrial Revolution may not guarantee stable labor shares. While experts reject fears of immediate white-collar collapse or demand collapse, they warn of political risks stemming from slow, decades-long job displacement and the potential for wealth concentration if AI remains monopolized rather than commoditized. The consensus suggests that broad prosperity depends on adopting new wealth distribution mechanisms like sovereign wealth funds and ensuring open AI models to prevent extreme inequality and maintain human-centric economic value.

Chip design from the bottom up – Reiner Pope

Reiner Pope, Dwarkesh

This analysis dissects the hardware architecture of modern AI accelerators, detailing how Multiply-Accumulate units and systolic arrays minimize data movement to overcome the area and energy costs of traditional CPU logic. It contrasts fixed-function ASICs and programmable FPGAs while highlighting the strategic shift from cache-based CPU designs to deterministic scratchpads in TPUs to optimize compute-to-memory ratios. Furthermore, the discussion evaluates current trends such as low-precision FP4 arithmetic and splittable array topologies, emphasizing that quadratic scaling and massive parallelism drive future efficiency gains in silicon design.

What rebuilding AlphaGo teaches us about self-play, RL, and future of LLMs - Eric Jang

Eric Jang, Ron Minsky, Dan Pontecorvo

Eric Zhang reconstructs AlphaGo to demonstrate how modern computing, including LLM-assisted coding and efficient neural architectures, reduces training costs from millions to thousands of dollars while solving Go's NP-hard complexity through Monte Carlo Tree Search. The presentation details the evolution from human-supervised data to tabula rasa self-play, highlighting how MCTS provides low-variance supervision that stabilizes value function learning for mid-game states. This framework validates Go as a scalable sandbox for testing automated AI research, offering transferable insights for robotics and drug discovery via verifiable performance loops.

David Reich – Bronze Age shock, the Neanderthal puzzle, & the sudden spread of farming

David Reich, Ali Akbari

Harvard geneticist David Reich presents a preprint study analyzing 16,000 ancient genomes which overturns the assumption that human natural selection has been dormant for hundreds of thousands of years. The research identifies a dramatic intensification of directional selection during the Bronze Age, particularly driving rapid adaptation in immune and metabolic traits due to urbanization and livestock interaction, while simultaneously proposing a new model for the complex relationships between modern humans, Neanderthals, and Denisovans. By isolating selection signals from population migration and drift, the study demonstrates that modern humans possessed the genetic variation necessary for rapid adaptation long before the development of agriculture or complex societies.

How GPT, Claude, and Gemini are actually trained and served – Reiner Pope

Reiner Pope, Ilya

John Mueller Jr. discusses the technical and economic drivers behind AI inference architectures, detailing how startups like Maddox optimize for memory bandwidth bottlenecks and latency bounds in sparse Mixture of Experts models. The analysis highlights that frontier models are currently overtrained by a factor of 100x relative to scaling laws, a phenomenon that dictates current API pricing structures for context length and caching tiers. Finally, Mueller explains how industry scaling is shifting toward larger single-rack domains to maximize expert parallelism while utilizing reversible network techniques to mitigate training memory constraints.

Michael Nielsen – Why aliens will have a different tech stack than us

Michael Nielsen

The analysis of the Michelson-Morley experiment reveals that scientific progress often relies on heuristics and aesthetic simplicity rather than immediate empirical falsification, as seen in the decades-long delay between Lorentz's ether-based transformations and Einstein's kinematic Special Relativity. This historical context informs the discussion of modern AI's role in prioritizing high-accuracy model fitting over parsimonious theory, while highlighting how deep learning and open science movements depend on forcing functions and collective attribution to navigate a path-dependent, vast tree of knowledge. Ultimately, breakthroughs require specific historical contingencies to mature, demonstrating that true scientific understanding emerges from high-stakes creative execution rather than passive information consumption.

Terence Tao – How the world’s top mathematician uses AI

Terence Tao

Johannes Kepler's transition from flawed geometric models to elliptical laws illustrates how high-quality data can overturn established theories, a dynamic that Terence Tao compares to modern AI generating hypotheses against verified datasets. As artificial intelligence drives the cost of idea generation to near zero, the scientific bottleneck has shifted from hypothesis creation to the verification and evaluation of results, necessitating new systems to distinguish genuine breakthroughs from algorithmic noise. While current AI excels at breadth by solving thousands of routine problems, human experts will increasingly focus on deep conceptual understanding and the development of new frameworks to ensure mathematical progress remains insightful rather than purely procedural.

Dylan Patel — The single biggest bottleneck to scaling AI compute

Dylan Patel

The Big Four hyperscalers have forecasted a combined $600 billion in capital expenditure, yet only about 20 gigawatts of incremental compute capacity is expected to come online in the US this year due to long-lead infrastructure projects. While OpenAI aggressively secured long-term capacity, Anthropic now faces a critical 4-gigawatt gap that forces reliance on expensive spot markets, highlighting a broader industry struggle against semiconductor supply bottlenecks and memory bandwidth constraints. Ultimately, EUV tool production limits and labor shortages constrain global AI scaling, positioning US allies with advanced manufacturing capabilities to maintain a significant lead over China for the foreseeable future.

The most important question nobody's asking about AI.

Following Anthropic's refusal to permit its AI models for mass surveillance and autonomous weapons, the Department of Defense has designated the company a "supply chain risk" and threatened coercive measures under the Defense Production Act to force compliance. With AI projected to automate 99% of the workforce within two decades, legal loopholes in the Fourth Amendment and plummeting surveillance costs create an imminent environment where the government could leverage its purchasing power to dismantle corporate moral boundaries. Although some predict a 74% chance the restrictions will be overturned, the analysis suggests that without new political norms explicitly banning state-led AI surveillance, the technology's diffusion will eventually allow the government to achieve universal monitoring regardless of individual vendor resistance.

Why Leonardo was a saboteur, Gutenberg went broke, and Florence was weird – Ada Palmer

Leonardo, Gutenberg, Ada Palmer

The dissolution of the Western Roman Empire forced Italian city-states to develop distinct republican or monarchical structures, with Florence establishing a unique merchant-led oligarchy that leveraged the Medici family's banking network to manipulate political outcomes. Concurrently, an educational movement attempting to forge virtuous "philosopher princes" through classical texts failed, prompting Niccolò Machiavelli and later Francis Bacon to replace character imitation with a pragmatic political science focused on analyzing specific historical decisions. This intellectual shift coincided with the maturation of paper and printing technologies, which dismantled medieval knowledge scarcity and enabled the rapid dissemination of information necessary for the scientific method to emerge from artisanal practices into a systematic engine for anthropogenic progress.

Dario Amodei — “We are near the end of the exponential”

Dario Amodei

Anthropic CEO Dario Amodei outlines a trajectory where AI capabilities will surge from current benchmarks to "PhD-level" intelligence by 2026-2027, enabling a "country of geniuses" to automate complex white-collar tasks and generate trillions in revenue by 2030. While technical scaling follows predictable exponential curves, Amodei argues that economic adoption will lag due to enterprise governance and security compliance, prompting a balanced financial strategy to reach profitability around 2028. The company simultaneously pursues Constitutional AI frameworks to ensure safety while advocating for federal regulatory preemption to prevent geopolitical fragmentation and authoritarian misuse of the technology.

Adam Marblestone – AI is missing something fundamental about the brain

Adam Marblestone

Steve Burrows proposes that the human brain's superior learning efficiency arises from a dual-subsystem architecture where a general-purpose cortical learning engine is guided by a specialized subcortical steering system that encodes evolutionarily tuned reward signals. Empirical research suggests that future artificial general intelligence may surpass current scaling limits by adopting diverse multi-agent co-evolutionary strategies, while formal verification tools like Lean offer a pathway to provably secure AI by translating mathematical proofs into verifiable reinforcement learning rewards. Concurrent efforts to map connectomes and implement brain-inspired training methods aim to reduce scientific timelines to a decade, shifting the focus from massive data scaling to understanding the biological constraints that govern cognitive generalization.

What are we scaling?

Toby Ord, Beren Millidge

Baron Millage argues that current Reinforcement Learning strategies rely on inefficiently pre-baking skills into models due to a fundamental misunderstanding of their ability to learn like humans, which keeps AI revenue far below the potential of knowledge work automation. While the industry anticipates a 2030 surge in continual learning revenue reaching the hundreds of billions, the lack of generalizable on-the-job capabilities and the immense compute requirements for RL scaling suggest AGI remains distant despite incremental progress. This perspective challenges the "superhuman researcher" narrative by emphasizing that solving the core learning problem requires a shift from specialized training loops to systems capable of semantic, self-directed adaptation.

Sarah Paine – Why Russia Lost the Cold War

Sarah Paine

This analysis attributes the dissolution of the Soviet Union to a convergence of sustained U.S. strategic pressure and internal systemic failures, with Ronald Reagan's military buildup and Richard Nixon's diplomatic pivot to China exacerbating Soviet economic stagnation. While Mikhail Gorbachev's flawed reforms and economic mismanagement critically weakened the regime, external factors including the Helsinki Accords and George H.W. Bush's diplomatic maneuvers accelerated the collapse by securing German unification and isolating the Eastern bloc. Ultimately, the event is presented as a result of cumulative Western policies that capitalized on inherent Soviet structural rot rather than a single definitive action.

Ilya Sutskever – We're moving from the age of scaling to the age of research

Ilya Sutskever, Dwarkesh

SSI is shifting from the era of pure scaling to a focused "age of research" that prioritizes robust generalization and biologically inspired learning mechanisms to overcome the current disconnect between model evaluation scores and real-world utility. The organization aims to deploy superintelligent systems within five to ten years that care for all sentient life through a strategic blend of rapid capability acquisition and gradual societal integration. This approach anticipates that as these human-like learners drive unprecedented economic growth, industry competition and government intervention will eventually converge on safety-focused architectural principles.