Latest Interviews

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.

CA Governor Candidate Steve Hilton on Why California is Destroying Itself & How a Republican Can Win

Steve Hilton, Chamath, Jason, Friedberg

Steve Hilton, a former UK government advisor and naturalized U.S. citizen, is leading the crowded California gubernatorial race by proposing a radical conservative platform focused on eliminating state income tax for households earning under $100,000 while slashing the state budget by $60 billion. His policy agenda identifies union power, CEQA litigation, and climate regulations as primary drivers of the housing and energy crises, advocating for immediate construction deregulation, increased oil production, and strict law enforcement to address homelessness and crime. With poll numbers ahead of Democratic contenders, Hilton aims to capitalize on a perceived Republican base in the state to implement these affordability measures in the 2026 election.

Demis Hassabis: We're Three Quarters of the Way to AGI

Demis Hassabis, Konstantine Buhler, Dennis

Demis Hassabis founded DeepMind with the strategic intent of leveraging game development revenue to fund the pursuit of artificial general intelligence, a goal he predicts will be realized by 2030. Following a pivotal breakthrough with AlphaFold that solved the protein folding problem, the organization expanded into Isomorphic Labs to accelerate drug discovery and is now pioneering the application of AI for scientific simulation in fields like biology and economics. Hassabis maintains that classical neural networks are sufficient to model complex biological systems and views information as the fundamental building block of the universe, positioning AI as the ultimate tool for understanding reality.

Demis Hassabis: Agents, AGI & The Next Big Scientific Breakthrough

Demis Hassabis, Garry Tan

DeepMind CEO Demis Hassabis projects Artificial General Intelligence will emerge around 2030, contingent on inventing specific architectural breakthroughs in continual learning and structured memory rather than merely scaling current methods. While deep tech startups are warned to anticipate AGI disrupting their decade-long cycles, the organization is simultaneously deploying efficient, modular "Flash" and open-source nano models to power edge devices and accelerate scientific discovery in fields like drug design and materials science. Ultimately, DeepMind advocates for interdisciplinary collaboration between AI and physical sciences to overcome current reasoning limitations and drive the next generation of autonomous agents.

300 Pages on Anduril: Inside Modern Warfare

Kyle Harrison, Molly O'Shea

Anduril Industries disrupts the defense sector by replacing the Pentagon's cost-plus acquisition model with fixed-cost contracts and a "build-to-mission" strategy that prioritizes high-volume, low-cost autonomous systems. The company utilizes its proprietary Lattice AI platform to integrate multi-domain assets into a unified network, aiming to counter China's industrial superiority by accelerating the production of drones, torpedoes, and munitions. Founded on the philosophy that credible military capability is essential for conflict deterrence, Anduril has seen its valuation double to roughly $60 billion as global power dynamics shift toward Great Power competition.

Electronics in Space

Philip Johnston

StarCloud co-founder Philip Johnston is addressing the surge in space launch capacity from companies like SpaceX and Astrospace by developing orbital data centers optimized for inference workloads. Y Combinator is currently recruiting chip designers with backgrounds at SpaceX and NVIDIA to create specialized processors that meet critical requirements for reduced mass, thermal efficiency, and radiation hardening. This initiative aims to capture the urgent market demand for computing resources that can operate effectively in the harsh space environment.

A Conversation with IMF Managing Director Kristalina Georgieva | Global Conference 2025

Kristalina Georgieva, Gerard Baker

Following an IMF downgrade of the global growth forecast to 2.8%, Managing Director Kristalina Georgieva addresses diverging inflation trajectories and rising fiscal constraints while urging China to undertake structural reforms to shift toward domestic consumption and reduced state intervention. Amidst a volatile trade landscape marked by the erosion of fair trade mechanisms and a transition to new economic equilibriums, policymakers are prioritizing bilateral agreements and supply-side productivity enhancements to stabilize financial markets and foster sustainable expansion. This strategic pivot seeks to navigate "off the charts" uncertainty by balancing emerging market resilience against the severe deflationary pressures in China and the debt challenges facing advanced economies.

How Warsh Could Shape Fed Policy

Kevin Warsh, Rob Kaplan, Alison Nathan

Following the Justice Department's decision to drop its investigation of Jerome Powell, the Trump administration expects Kevin Warsh to be confirmed as the next Fed Chair by June. Warsh, a former "lieutenant" to Ben Bernanke who views quantitative easing strictly as an emergency tool, plans to collaborate with Treasury Secretary Bessent to manage the balance sheet while prioritizing a reduction in the Federal Reserve's communication burden through the potential elimination of the dot plot. This strategy aims to navigate sticky inflation and geopolitical uncertainties that have pushed market rate cut expectations into 2027, requiring Warsh to build consensus among diverse FOMC members to secure seven votes for any policy shift.

Counter-Swarm Defense

An incident involving Iranian drone swarms exposing the cost asymmetry and fragmentation of current defense systems has prompted Y Combinator to solicit counter-swarm stack solutions. These new initiatives aim to replace traditional interceptor models with software-driven, non-kinetic technologies capable of neutralizing thousands of coordinated units through sensor fusion and autonomy targeting. The resulting infrastructure is expected to function more like scalable cloud networks than conventional weaponry, fundamentally shifting the defense paradigm from weapon operation to distributed system management.

The viral myth that made you think your job was safe

Rob Wiblin

A widely circulated report falsely attributed to MIT, which claimed a 95% failure rate for generative AI pilots, is exposed as a commercially motivated study authored by four developers with undisclosed financial stakes in competing AI frameworks. The analysis reveals that the original data actually indicates a 25% success rate for custom tools, attributing pilot terminations to organizational resistance rather than technical limitations while relying on an unpeer-reviewed methodology based on a small, non-transparent sample. This narrative shift challenges the prevailing skepticism surrounding enterprise AI by highlighting the report's conflict of interest and the statistical instability of its primary failure metric.

Aaron Levie on AI Adoption and Enterprise Workflows | The a16z Show

Aaron Levie, Steven Sinofsky, Martin Casado

Organizations are pivoting from failed centralized AI mandates to integrating autonomous agents directly into legacy workflows, necessitating significant architectural shifts beyond traditional hybrid software models. While token-gaming and system integration bottlenecks currently stifle productivity gains, the resulting increase in infrastructure complexity and code volume is projected to drive sustained demand for engineering talent rather than reduce it. This transition requires years of organizational change management to modernize fragmented data environments, ultimately creating a multi-decade opportunity for system integrators to bridge the gap between probabilistic machine users and rigid enterprise security protocols.

Company Brain

Enterprise AI automation is currently bottlenecked by the fragmentation of proprietary domain knowledge siloed within human memory and legacy systems, creating an urgent need for a "company brain" that extracts, structures, and converts this data into executable skills files. This dynamic infrastructure serves as a critical intermediate layer between raw enterprise data and reliable AI agents, moving beyond simple document chatbots to map specific operational workflows like refund handling and incident response. The speaker predicts universal global adoption of this technology for safe, efficient automation and invites startups to apply to Y Combinator to build this essential primitive.

AI-Personalized Medicine

Intelligent agents are deploying personalized health data and falling production costs for genetic medicines to generate precise clinical suggestions and bespoke treatments. This convergence of regulatory openness and declining diagnostic expenses is enabling the early detection of disease signals while allowing patient access to specialized trials. Industry leaders anticipate this shift will democratize access to severe illness care and spur a surge in startups building the necessary infrastructure for this new model.

AI-Native Service Companies

Driven by the rapid evolution of AI models from software tools to autonomous professional agents, the market is shifting toward "AI-native" companies that sell direct services rather than software licenses. This investment thesis targets high-spend sectors like insurance, accounting, and healthcare administration where global spending on services vastly exceeds software costs and outsourcing rates are already high. To capitalize on this shift, a specific call for submissions has been issued to founders building "place as a service" ventures that embed human professionals within their AI-driven delivery models.

Why does political violence seem to be getting worse in America? | The Economist

John Prideaux, Rosie Blau

During the White House Correspondents' Dinner, 31-year-old Cole Thomas Allen breached the Hilton Hotel security checkpoint to attempt an attack on President Trump before being tackled by police. The incident forced an immediate evacuation of the ballroom, resulted in an officer injury, and triggered a review of Secret Service protocols following the shooter's prior claims of lax access measures. While President Trump utilized the event to advocate for new defensive infrastructure and praised the agency's response, experts warn that despite a perceived rise in political violence, the long-term impact on governance and approval ratings will likely remain tied to broader economic conditions rather than the assault itself.