ScienceRSI: Recursive Scientific Self-Improvement
- 1HKUST
Abstract
Scientific AI can improve cumulatively when capabilities developed in one model become useful learning signals for the next. We introduce ScienceRSI, a framework for recursive scientific self-improvement organized around specialize, verify, transfer, and reuse. Compact experts develop complementary scientific capabilities; shared evaluation identifies their useful contributions; a target model integrates those contributions and can serve as the starting point for further specialization. This organization connects reasoning, model scaling, and iteration through reusable scientific capabilities. We evaluate its transfer step in 4B[->]4B, 4B[->]8B, and 8B[->]27B configurations. ScienceRSI ranks first among the compared methods on all five science benchmarks for the 4B and 8B students. Relative to rubric-based reinforcement learning, scientific-reasoning averages improve by 2.0 and 1.9 percentage points, and domain-science averages by 3.3 and 2.1 points, respectively. In a separate cross-generation 27B extension, reported scores improve over Vanilla on all seven benchmarks, with the unweighted mean rising from 87.3% to 94.0%. These results support verified capability transfer across model scales as a practical foundation for ScienceRSI. Accumulating these gains over repeated rounds is the next step toward sustained recursive scientific improvement.
Related
- ECG-RSI: Recursive Self-Improvement through Targeted Electrocardiogram Waveform Evolution
- From Problems to Solutions: How Self-Questions Drive Strong Scientific Reasoning
- ADMET-EvO: a self-evolving scientific agent for sustained research across heterogeneous tasks
- biHash: bisulfite-aware recursive nucleotide hashing
- Large-scale analysis of transcript data reveals thousands of recursive splicing events in human introns
BibTeX
@misc{xu2026sciencersi,
title = {{ScienceRSI: Recursive Scientific Self-Improvement}},
author = {X. Xu and J. Yan and Z. Wang and M. Xu and Z. Zhang},
howpublished = {bioRxiv},
year = {2026},
month = oct,
doi = {10.64898/2026.10.20.760442},
url = {https://www.biorxiv.org/content/10.64898/2026.10.20.760442v1},
note = {CC BY},
}
RIS
TY - UNPB
TI - ScienceRSI: Recursive Scientific Self-Improvement
AU - X. Xu
AU - J. Yan
AU - Z. Wang
AU - M. Xu
AU - Z. Zhang
PY - 2026
DA - 2026/10/20
PB - bioRxiv
DO - 10.64898/2026.10.20.760442
UR - https://www.biorxiv.org/content/10.64898/2026.10.20.760442v1
AB - Scientific AI can improve cumulatively when capabilities developed in one model become useful learning signals for the next. We introduce ScienceRSI, a framework for recursive scientific self-improvement organized around specialize, verify, transfer, and reuse. Compact experts develop complementary scientific capabilities; shared evaluation identifies their useful contributions; a target model integrates those contributions and can serve as the starting point for further specialization. This organization connects reasoning, model scaling, and iteration through reusable scientific capabilities. We evaluate its transfer step in 4B[->]4B, 4B[->]8B, and 8B[->]27B configurations. ScienceRSI ranks first among the compared methods on all five science benchmarks for the 4B and 8B students. Relative to rubric-based reinforcement learning, scientific-reasoning averages improve by 2.0 and 1.9 percentage points, and domain-science averages by 3.3 and 2.1 points, respectively. In a separate cross-generation 27B extension, reported scores improve over Vanilla on all seven benchmarks, with the unweighted mean rising from 87.3% to 94.0%. These results support verified capability transfer across model scales as a practical foundation for ScienceRSI. Accumulating these gains over repeated rounds is the next step toward sustained recursive scientific improvement.
ER -