CosmoAgent：用 LLM 智能体搜索暗能量与修正引力模型
CosmoAgent: AI Agent for Theoretical Cosmology
物理学家让 8 个大模型当宇宙学研究员，自动发明引力理论并用 DESI 和 DES 数据打分，最好的模型拿到了 ΔAIC=-3.328。
论文提出 CosmoAgent，一个基于 LLM 的智能体框架，用于搜索暗能量和修正引力理论。作者定义了协变语法，让 LLM 能在修正引力拉格朗日量空间中探索，并通过计算流水线推导运动方程、求解背景动力学，用 DES 5 年 Ia 型超新星样本和 DESI DR2 BAO 数据评估似然，再按 AIC 排序。团队测试了 ChatGPT、Claude、Gemini、Kimi、DeepSeek、Qwen 的 8 个前沿模型，均能找到比平坦 ΛCDM 名义 AIC 更低的模型，最好提升 ΔAIC = -3.328，其中 7 个模型找回了二次精质解。作者指出这些 AIC 提升幅度不足以构成对 ΛCDM 的观测偏好。
CosmoAgent: AI Agent for Theoretical Cosmology
We present CosmoAgent, an agentic AI framework using large language models (LLMs) to search for dark-energy and modified-gravity models. We define a covariant grammar that allows LLM agents to explore the space of modified-gravity Lagrangians. An in-house computational pipeline derives the equations of motion, solves the background dynamics, and evaluates the likelihood using late-time expansion data, including the DES 5-year type Ia supernova sample and the DESI Data Release 2 BAO measurements. Theories are ranked by the Akaike Information Criterion (AIC) to balance simplicity and likelihood. We benchmarked 8 frontier LLMs from ChatGPT, Claude, Gemini, Kimi, DeepSeek, and Qwen, and all LLMs are able to find theories with a lower nominal AIC than flat $Λ$CDM, with a best improvement of $Δ$AIC = -3.328. Seven recover quadratic quintessence. Exploratory searches also return induced-gravity and scalar-Gauss-Bonnet structures. The modest AIC improvements do not establish an observational preference for these models over flat $Λ$CDM. The code is publicly available at wsunamoutlook-prog/AI-Modified-Gravity.