Reproduce “A Self-Evolving AI Agent System for Climate Science” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Zijie Guo, Jiong Wang, Fenghua Ling, Wangxu Wei, Xiaoyu Yue, Zhe Jiang, Wanghan Xu, Jing-Jia Luo and 14 more
EarthLink is a self-evolving AI agent system designed to automate climate science research workflows, integrating diverse data and tools from planning to analysis and discovery. The system demonstrated human-like analytical proficiency across various tasks and autonomously identified new precursors for Atlantic Ni˜no, achieving a forecast skill (Temporal Correlation Coefficient) of 0.46, an improvement over the 0.39 baseline.
Key insights
- A modular multi-agent architecture, leveraging foundation LLMs like GPT-5, can effectively integrate diverse functionalities to automate complex scientific workflows in climate science.
- Combining a domain-specific knowledge base with dynamic planning and code generation enables an AI system to achieve analytical proficiency comparable to an early-career researcher.
- Implementing self-evolution mechanisms, where successful executions refine knowledge and tool libraries, allows the AI agent system to continually improve its reasoning and efficiency over time.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.