BETA RELEASE

Summary

An analysis of legal and policy considerations for defining 'frontier models' in AI regulation, examining technical inputs, capabilities, risk, and U.S. administrative law doctrines.

Key quotes

Laws that fail to carefully define ambiguous technical terms often fail in their intended purposes, sometimes with disastrous results.
Training compute thresholds function as a useful proxy for model capabilities because capabilities tend to increase as computational resources used to train the model increase.

The paper explores various methodologies for defining high-capability AI models, comparing approaches used in the US Executive Order 14110, California’s SB 1047, and the EU AI Act. It specifically analyzes the tension between rigid statutory definitions and flexible regulatory definitions in the context of rapid technological change.