GPT-5.6 and Claude Fable Lack Critical Thinking in Research

Epoch AI and Anthropic discovered that current AI models like GPT-5.6 Sol and Claude Fable 5 can conduct experiments but lack scientific self-criticism and genuine creative thinking. Sol achieved only 15% of the human reference score, with that result stemming from already known methods. The models' primary weakness lies in their inability to critically question their own results, highlighting a significant limitation. These findings suggest that current models remain far from achieving autonomous research capabilities, raising concerns about the reliability of their results in future studies.