Agent4Sci (Anthropic Fellows Program): AI agents beat the best published results on each of 20 scientific benchmarks across 11 fields — Claude Opus by 10.4% on average relative to them — but, the authors conclude, almost entirely through machine-learning tricks rather than scientific insight
translated by AI, reviewed by the editors
· Science
- technology
- Machine learning
- source
- arxiv.org ↗
¶ cprx5f84m