Jon Stokes Argues Utility Functions Are Post-Hoc Constructs, Making Alignment Impossible

Jon Stokes argues that utility functions are post-hoc constructs by observers rather than internal objects that can be adjusted, making AI alignment incoherent and insufficient; he calls on the AI safety community to replace the framework.

2026-10-15 ~ 2026-10-16 · 2 related posts

• Jon Stokes argues AI alignment is infeasible: utility functions are hindsight constructions — sebpaquet · 2026-10-15
• Jon Stokes: alignment is impossible because utility functions are hindsight constructions — sebkrier · 2026-10-16