|
SimpleBench, a new multiple-choice benchmark, demonstrates that high school-level human reasoning (83.7% accuracy) outperforms all tested frontier AI models, including Claude Fable (81.9%), across questions involving spatio-temporal reasoning, social intelligence, and linguistic adversarial robustness. The benchmark reveals that current LLMs' reliance on memorized knowledge and approximate reasoning retrieval is insufficient for answering certain basic questions that non-specialized humans can solve.
Read Full Article →
← Back to latest