Publication
arXiv
Stage
Preprint
What we read
Summary of the paper
Authors
Hejia Geng, Zesen Huang, Haoyang Li, Wenbin Li, Koutian Wu, Zihan Zhou, Yuanbo Pang, Weihao Liu, Zigong Xu, Zhiping Li, Zongzheng Zhang, Chuanfei Dong, Jiankai Sun, Tianzhe Zheng, Fengyu Xie, Yue Ma, Yueheng Shi, Tong Xie, Zonglin Di, Xianrong Liu, Qucheng Gao, Yimin Liu, Jiaming Pan, Sheng Huang, Xiao-Han Ma, Lanqing Yuan, Zhenlin Zhu, Ziang Liu, Ziyang Xu, Junkai Wang, Kangkai Liang, Jiayi Xian, Zehong Zhao, Liuwei Xu, Jingxu Xie, Peijin Zhang, Qiang Gao, Chengyi Xing, Zhe Zhao, Xi Wang, Yaopeng Xing, Xing Meng, Zhenfei Yin, Yingcheng Wu, Ling Yang
Universities and research institutions
Not yet supplied in verified metadata; the Brief does not guess.

What they did and found

ScienceIDE grouped large sets of scientific code into tasks that machines can try to run. In 85 difficult tasks across 18 environments, top AI models fixed some defects and learned from examples, but most tasks remained unsolved within one hour.

Why it matters

It shows where AI can help with routine fixes, while human judgment remains essential for tricky parts and safety checks in real work settings.

ScienceIDE turns scientific code into runnable tasks, showing AI can repair simple issues and understand code, yet it struggles with complex cross‑dependency setups today too.

Teams may use AI for routine fixes or adding missing features, while scientists keep key decisions and handle hard cases that require careful review often.

Read the paper PDF ↗

Original sources · 1
  1. ScienceIDE: Turning World's Scientific Codebase into Agent Learnable Environments ↗arXiv · 2026-09-16

Check the original paper for its authors, methods, version and access terms.

What remains uncertain

Study focuses on verified physics and geoscience tasks; results may not extend to broad open-ended discovery or long-term outcomes.