Frontier language models struggle to analyze experimental spectral data
CME-SpectrumBench: Can LLMs Analyze Condensed Matter Spectral Data?
Artificial Intelligence
Summary
Analyzing experimental data in advanced physics experiments can be tricky, and large language models (LLMs) are usually tested on math and theory problems rather than real data. The authors created a new benchmark called CME-SpectrumBench to see how well these models handle noisy, complex experimental spectral data from condensed matter physics. They found that LLMs alone often miss details or make mistakes with this data, but do better when they can write and run code to analyze it. This means future models might need better ways to interact with complex data directly.
What this means in practice
- •For experimental physicists: Use language models combined with coding tools to assist in analyzing complex spectral data from condensed matter experiments.
- •For software developers for scientific tools: Develop AI-powered tools that generate and run code to extract insights from noisy research datasets in material science.$Commercial implications: Enables building advanced analysis software for research labs that reduces manual coding and speeds up spectral data interpretation.
Authors
Jin Gene Wong, Anjney Midha, Joseph Mathew Tennyson, Wei-Lin Chiang, Ion Stoica, Zhi-Xun Shen
Abstract
Large Language Models (LLMs) are presently well assessed by expert-level mathematics and physics benchmarks which focus on theoretical, symbolic reasoning. However, we lack experimental benchmarks that evaluate data analysis in frontier research problems, a capability that future domain-specific foundation models must have. To this end, we present CME-SpectrumBench, an expert-designed benchmark consisting of 190 questions that measures how reliably frontier LLMs analyze research-level spectral scientific data in condensed matter experiments (CME). Using simulated CME spectral data, we prompt models to identify structures from 2D intensity arrays in the face of noise and limited instrumental resolution, a test of dense context reasoning. Models are scored by how accurately their numerical answers approach the ground truth. Under direct query, scores generally pale in comparison to handwritten code due to clear failure modes: models can find simple features but struggle to trace patterns or compute derived quantities, and performance typically degrades with increased data resolution and noise/convolution levels. LLM performance becomes comparable to handwritten code by allowing tool use (access to a Python interpreter) with/without ReAct prompting, in effect mapping questions onto agentic code generation problems. Our findings show that frontier models can serve as research assistants by interfacing with data through code, but still face difficulties processing information directly from large datasets - a key barrier future foundation models must overcome. The code and data are publicly available at https://huggingface.co/datasets/jingene/CMESpectrumBench.