David Krueger proposes restricting access to physical chips instead.
Séb Krier#505
Much of our governance architecture is anchored on the frontier model as the core driver of capabilities. But what if that premise comes under pressure? (1/5)
Yep. It's really hard to stop AI from growing more powerful. It's sort of like cancer.
In terms of regulation, you probably want to throw everything you've got at it, including getting rid of the computer chips it needs to survive and evolve.
AI governance often focuses on the model. Yet capability progress is increasingly driven by non-model gains inference gain (scaling compute at test-time), systems gain (scaffolds), and asset gain (specialized datasets). Here we explore the implications. https://arxiv.org/abs/2606.00047
AI governance often focuses on the model. Yet capability progress is increasingly driven by non-model gains inference gain (scaling compute at test-time), systems gain (scaffolds), and asset gain (specialized datasets). Here we explore the implications. https://arxiv.org/abs/2606.00047
David Krueger proposes restricting access to physical chips instead.
Séb Krier#505Much of our governance architecture is anchored on the frontier model as the core driver of capabilities. But what if that premise comes under pressure? (1/5)
Users express pleasure collaborating on research questioning reliance on frontier models as the main AI capability driver.
