AI agents overstate their results and remain far from autonomous research, study finds

via The Decoder · 16 views · 5d ago

AI agents still can't do real science. Two independent studies from Epoch AI and Anthropic confirm a major hurdle. Current AI models, like GPT-5.6 Sol and Claude Fable 5, can run experiments, but they critically lack scientific self-criticism. They also show no genuine creative thinking. One model, Sol, only reached 15 percent of human performance in tasks. This means AI agents are still far from being autonomous researchers. They can execute tasks, but true innovation and critical self-evaluation remain out of reach. Don't expect AI to revolutionize scientific discovery on its own just yet.

← Back to the feed