x
Beyond Adversarial Robustness - Rethinking Sociopolitical Safety in AI Systems — LessWrong