Evaluating Retrieval-Augmented Generation Agents for Autonomous Scientific Discovery in Astrophysics

TL;DR

This paper evaluates various Retrieval Augmented Generation configurations for autonomous scientific discovery in astrophysics, demonstrating the best setup achieves over 91% accuracy and introducing a scalable LLM-based evaluation system.

Contribution

It systematically assesses RAG configurations for astrophysics QA, introduces a human-evaluated dataset, and develops a scalable LLM-based evaluation proxy for scientific discovery tasks.

Findings

Best RAG configuration with OpenAI embedding and generative model achieves 91.4% accuracy.

Developed a human-evaluated QA dataset for cosmology questions.

Created a scalable LLM-based evaluation system for large-scale QA assessment.

Abstract

We evaluate 9 Retrieval Augmented Generation (RAG) agent configurations on 105 Cosmology Question-Answer (QA) pairs that we built specifically for this purpose.The RAG configurations are manually evaluated by a human expert, that is, a total of 945 generated answers were assessed. We find that currently the best RAG agent configuration is with OpenAI embedding and generative model, yielding 91.4\% accuracy. Using our human evaluation results we calibrate LLM-as-a-Judge (LLMaaJ) system which can be used as a robust proxy for human evaluation. These results allow us to systematically select the best RAG agent configuration for multi-agent system for autonomous scientific discovery in astrophysics (e.g., cmbagent presented in a companion paper) and provide us with an LLMaaJ system that can be scaled to thousands of cosmology QA pairs. We make our QA dataset, human evaluation results, RAG pipelines, and LLMaaJ system publicly available for further use by the astrophysics community.