Selected reading
Three starting points for studying the pace of AI. All are external work, credited to their authors.
Pacing the Frontier: A Framework & Research Agenda ↗
A broad account of interventions affecting AI development, deployment, and diffusion. Organizes open questions around incentives, intervention design, and the full lifecycle of a restriction.
Read as a research agenda: its proposed interventions and open questions are not established effects.
Computing Power and the Governance of Artificial Intelligence ↗
Examines compute as an instrument of AI governance: why it may be tractable to govern, which policy uses it could support, and the limitations and risks of doing so.
Read alongside the operational question: which property would a proposed measure actually observe or constrain?
Model evaluation for extreme risks ↗
Sets out a framework for evaluating dangerous capabilities and alignment, with implications for training, deployment, transparency, and security.
Evaluation informs a decision. Its conclusions remain bounded by the system, elicitation method, and conditions tested.