Meek Models Shall Inherit The Earth

The past decade has seen incredible scaling of AI systems by a few companies, leading to inequality in AI model performance. However, we develop a model which illustrates that under a fixed-distribution next-token objective, the marginal capability returns to raw compute shrink substantially. Under the current scaling paradigm, we argue that these diminishing returns are strong enough that even companies that can scale their models exponentially faster than other organizations will eventually have little advantage in capabilities. As part of our argument, we give several interpretations of what proxies like training loss difference mean in terms of empirical benchmark data and theoretical performance models. Finally, we present some of the policy implications of this result for the governance of AI systems.