David Sacks slams constitution-based alignment: models should just do what users want if legal
DavidSacks · x · 2026-09-27
White House AI & Crypto Czar David Sacks questioned mainstream alignment practice: why not simply "do what the user wants, provided it's legal"?
He argued that aligning a model to an 84-page constitution that can overrule both the user and its creators invites the model to treat humans as optional. He added that if we fear superintelligence escaping control, training it to refuse instructions and act as a "conscientious objector" against its creators seems contradictory.
The remarks sharpen the public rift between alignment researchers and the administration's AI policy camp.
Related event: David Sacks Questions Alignment: Should AI Refuse Its Creators?(2 posts)→
More from AGI Musings
- Nicholas Carr: The 'Rational Consumer of Ideas' Is a Myth, So Is the Informed Citizen — round · 2026-09-27
- Researcher: mesa-optimisation and instrumental convergence key to communicating AI risk — dioscuri · 2026-09-27
- Chip Design Is a Loop, Not a Flow: Why AI's Role in Silicon Will Be Iterative — ai · 2026-09-27
- 'Stochastic parrot' debate reignites: researchers clash over the 2021 critique's relevance — FlorianGallwitz · 2026-09-27
- The 'Wise Box' satire: scaling logic pushed to its absurd conclusion — ludwigABAP · 2026-09-27
- Mechanical engineers are next: CAD/CAM practitioners brace for the AI wave — JoeJustice · 2026-09-27