Probing Materials Knowledge in LLMs: From Latent Embeddings to Reliable Predictions

TL;DR

This study evaluates how large language models encode materials science knowledge, revealing that output modality affects reliability, and that embeddings from intermediate layers can improve numerical predictions, but model performance varies over time.

Contribution

The paper systematically assesses 25 LLMs across materials science tasks, highlighting the impact of output modality, introducing embedding extraction from intermediate layers, and analyzing temporal performance variation.

Findings

Symbolic tasks yield consistent, verifiable answers after fine-tuning.

Numerical predictions improve with embeddings from intermediate transformer layers.

Model performance varies significantly over time, affecting reproducibility.

Abstract

Large language models are increasingly applied to materials science, yet fundamental questions remain about their reliability and knowledge encoding. Evaluating 25 LLMs across four materials science tasks -- over 200 base and fine-tuned configurations -- we find that output modality fundamentally determines model behavior. For symbolic tasks, fine-tuning converges to consistent, verifiable answers with reduced response entropy, while for numerical tasks, fine-tuning improves prediction accuracy but models remain inconsistent across repeated inference runs, limiting their reliability as quantitative predictors. For numerical regression, we find that better performance can be obtained by extracting embeddings directly from intermediate transformer layers than from model text output, revealing an ``LLM head bottleneck,'' though this effect is property- and dataset-dependent. Finally, we present a longitudinal study of GPT model performance in materials science, tracking four models over 18 months and observing 9--43\% performance variation that poses reproducibility challenges for scientific applications.