Sage-Taiyi leads scientific ideation benchmarks via metacognitive control
Original title: Sage-Taiyi: A Metacognitive Architecture for Scientific Discovery
It scores 54.0 overall on Taiyi-Bench against 47.9 for AI Scientist-v2, and chemistry work raised two yields to 84% and 69%.
Authors introduce Sage-Taiyi, a metacognition-inspired architecture that revises evidence, assumptions and open questions before selecting research targets and reusable methods. On Taiyi-Bench, covering 265 tasks from 50 papers across ten disciplines, it ranks first on all eight historical-rediscovery and open-ended ideation metrics. Five expert-guided collaborations connect proposals to experimental and computational validation in chemistry, materials science and engineering.
- Cobalt-catalyzed carbonylation yield rose from trace amounts to 84%.
- Photocatalytic allylic amination yield rose from 13% to 69%.
- The system also performs well on two external ideation benchmarks.
- Products
- Taiyi-BenchAI Scientist-v2
- AI models
- Sage-Taiyi
From the original article
1 min read at huggingface.co
We introduce Sage-Taiyi, a metacognition-inspired architecture that uses accumulated scientific understanding to guide what to investigate next and how to investigate it. Through explicit monitoring and control, the system revises its understanding of evidence, assumptions, and unresolved questions, then uses these assessments to adapt research targets and select reusable reasoning methods.
By Weida Wang, Benteng Chen, Yigeng Jiang · huggingface.co
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.