Better tools for scientific discovery.
We want AI to be useful in scientific discovery. RSI4Science brings together tasks, data and checks that researchers can use to study how their systems learn from results and improve.
Shared resources help people build on one another’s work. Recursive self-improvement (RSI) uses a system’s current capabilities to improve itself, then brings the updated capabilities into further rounds of improvement. Our datasets provide scientific tasks and feedback for researchers testing that loop. Why study RSI?
A shared effort
Initiated by Xiongye Xiao and Yue Zhao. The website is maintained by DUAL Lab.
Researchers, students and volunteers across disciplines are welcome.
A growing community collection
The molecular teaching example and local tools are available to explore. The wider collection is in development. You can suggest a source or prepare a dataset and contact us for review. These teaching materials are not evidence of a completed RSI experiment.
Take part
Share a useful resource, try an example, improve a guide or help review a scientific task. A source link or a small, practical contribution is a good place to start.
Editorial and scientific review
RSI4Science / DUAL Lab publishes the research briefs. Initial briefs used AI-assisted research, writing and source checks. Articles identify sources, assistance and material corrections; independent reproduction or human editorial review is claimed only when recorded.
Scientific and engineering reviews are separate. There is no established review panel or completed independent task validation yet. Contributors cannot approve their own work alone. Links to other projects do not imply partnership or endorsement.
Prepare a correction draft using public material. The form saves to your device and does not submit it.