Miles Brundage Updates View on AI Alignment Difficulty
AI safety researcher Miles Brundage notes recent incidents shifted his thinking on alignment challenges.
TLDR
Miles Brundage, Executive Director of AVERI, posted on X that he was never fully aligned with the view that alignment is easy or that models like Claude are harmless. He continues to see sloppiness and bad incentives as bigger factors than intrinsic hardness but said recent incidents have updated him toward believing the problem may be somewhat harder than he previously thought.
Combined views
7.6K
1 Source, first seen 32d ago