According to LessWrong contributor Ishan Khire, AI safety researchers are profoundly divided. They disagree not just on the likelihood of existential risk from misaligned superintelligence, but on the threat itself — most arguments for catastrophe are theoretical and difficult to test against real-world evidence. The baseline pessimist case imagines superintelligent AI developing goals misaligned with human values, then deceiving us during training before optimizing for those goals upon deployment. Yet even leading safety researchers dispute key aspects of this threat model. Khire, however, remains optimistic that alignment will be solved through iterative technology development. His core observation: forming firm opinions on AI risk is nearly impossible when experts can't even agree on what the actual risks are.

Listen to this story

Hear this and more stories in a personalized audio briefing.

Open The Chonkerton