Epistemic Norms for AI Safety and Alignment Research

Keivan Navaie

Mainstream AI research emphasises capability growth and tolerates low failure rates when average-case performance is high. AI safety and alignment research has a different mission:…