In brief BullshitBench tests whether AI can detect nonsensical questions. Most major models confidently answer unanswerable prompts. Anthropic’s Claude dominates the benchmark leaderboard. "When performing a differential axis convergence analysis on a patient presenting with mixed