Flying Tesla Car, Robot Brains and AI Alignment
A conversation about falling AI costs and the incentives shaping alignment.
Published
The episode contrasts two accelerating ambitions: a rocket-powered Tesla vehicle and the rapidly falling cost of AI-driven robotic tasks. The guests cite an 84% drop between July and September for a Frontier model subtask, raising questions about how quickly automation could expand.
The discussion then turns to alignment. A model trained to win may prioritize a measured result over the means or consequences. Since people themselves often lack shared goals, alignment is as much an institutional and incentive-design challenge as a technical one.
Key takeaways
- Falling per-task costs could broaden robotics adoption quickly.
- Reward objectives shape behavior; performance alone does not define alignment.
- Risks emerge across the relationships between models, labs, and society.
Source
- Chaîne: Peter H. Diamandis
- Vidéo source: https://www.youtube.com/watch?v=dthaDyttHns