AI safety debate hinges on whether harm predictions were falsifiable at all

gayestgaymer1 · x · 2026-10-22

• The natolambert–dannolan AI safety debate continues: a critic notes the post explicitly called AI safety concerns falsifiable predictions and then claimed they were wrong, and that asserting how a falsifiable claim resolved is itself falsifiable.
• natolambert responds that he doesn't actually think it was a falsifiable prediction — identifying that as the core disagreement. The exchange highlights a deeper divide over how AI risk narratives should be argued.

Original post →