About
I’m an independent researcher on semiconductors and AI policy, figuring out my next big bet. Right now that means working on a number of projects at the compute-AI intersection — from China’s semiconductor industry to how society adapts to increasingly capable AI systems to trying to get AI to help with AI policy (aka do my job). I also still regularly advise governments and AI companies on policy and technical decisions. I’m based in Washington, DC.
Until recently, I was leading the Compute Team at RAND’s Center on AI, Security, and Technology within RAND Global and Emerging Risks. I was also a Professor of Policy Analysis at RAND’s School of Public Policy, teaching classes on AI policy.
More broadly, my research focuses on the role of compute for advanced AI systems and how compute can be leveraged as an instrument for AI governance, with an emphasis on policy development and security implications.
My publications cover the impacts and governance of advanced AI systems and empirical trends in machine learning, such as training compute, data, and AI hardware. I’m also researching the role that compute/cloud providers play in AI governance and how society can adapt to increasingly advanced AI systems.
Previously, I was a Senior Research Fellow at the Centre for the Governance of AI (GovAI), with whom I remain affiliated as an Adjunct Fellow. I’m also a member of the OECD.AI Expert Group on AI Compute and Climate, a Pro Forecaster at INFER Pub, and a co-founder of Epoch, where I was also a researcher. I have a background in computer engineering and studied at ETH Zürich and RWTH Aachen.
You can find more about my professional background on LinkedIn, my publications on Google Scholar, musings on my blog, and my tweets here. You can contact me using the contact details in the footer.
Research
You can find most of my publications on my research page and Google Scholar.
Some selected recent publications:
-
China’s AI Models Are Closing the Gap—but America’s Real Advantage Lies Elsewhere (RAND website & X thread): China will likely match U.S. AI model capabilities this year, triggering inevitable concerns about America’s technological edge. However, this snapshot comparison misses the bigger picture.
-
Understanding the AI Diffusion Framework (X thread): I explain and analyze the AI Diffusion Framework—what it does, how it works, its rationale, why it was needed, why China can’t easily fill the void, and some thoughts on model weight controls.
-
Training Compute Thresholds: Features and Functions in AI Regulation (X thread): Regulators in the US and EU are using training compute thresholds to identify general-purpose AI models that may pose risks. But why are they using them if training compute is only a crude proxy of risk?
-
Societal Adaptation to Advanced AI (X thread): Increasingly advanced AI systems will diffuse into society. How do we manage the accompanying risks?
-
Governing Through the Cloud: The Intermediary Role of Compute Providers in AI Regulation (X thread): We propose compute providers as an important node for AI safety, both in providing secure infrastructure and acting in an intermediary role for AI regulation, leveraging their unique relationships with AI developers and deployers.
-
Computing Power and the Governance of AI (X thread and blog post summary): Recent AI progress has largely been driven by increases in the amount of computing power used to train new models. Governing compute could be an effective way to achieve AI policy goals, but could also introduce new societal risks.
-
Increased Compute Efficiency and the Diffusion of AI Capabilities (X thread and blog post summary): Falling development costs allow more and more groups to reproduce existing AI capabilities. But falling costs also benefit large compute investors, helping them maintain their leads by pushing forward the frontier.
Previously
I’m usually pretty good at keeping myself busy. In the past, I’ve done research in neuroscience, explored wireless and embedded systems, helped people use evidence and reason to achieve their altruistic goals, explored Amazon from the inside, became a whale watching guide (and subsequently saved the whales), deployed ML to resource-constrained devices, fixed windshields on Kauai, measured radiated spurious emissions of 5G, investigated charities with legacies.now, cuddled with a bunch of dogs, and, lastly, failed a start-up on particulate sensors with friends of mine.
That’s not the correct order, and some of those things happened simultaneously—I’ll leave this to you.
“The real problem of humanity is the following: we have paleolithic emotions, medieval institutions, and god-like technology"