“Jensen Huang argues that the best defense against a powerful, malicious AI model is not a competing proprietary model, but a widespread deployment of cheap, open-source AI models for defense.”