Robo-brainiac writes its own science papers and says it beats humans
Researchers led by Jaehyun Nam and Jinsung Yoon describe ScientistTwo, a fully autonomous multi-agent AI framework for scientific research. Given a problem, it sets baselines, forms hypotheses, runs experiments and ablation studies, and tests findings with a simulated peer-review rebuttal engine, all without human help. Benchmarked against papers accepted at ICLR, ICML and NeurIPS, it reportedly produced publishable papers and executable code. The authors say its solutions beat human state-of-the-art models and earned higher ratings than human papers from AI review agents.
Why it matters: It suggests AI could carry out research with little human help, which may speed up discovery and affect scientists and journals, though the results are self-reported and judged by AI reviewers, not yet humans.
Comments
No comments yet.