Probing Materials Knowledge in LLMs: From Latent Embeddings to Reliable Predictions
This work addresses reliability and reproducibility issues for researchers applying LLMs to materials science, highlighting incremental insights into model behavior and limitations.
The study evaluated 25 large language models on materials science tasks, finding that fine-tuning improves consistency for symbolic tasks but not for numerical ones, where models remain unreliable due to inconsistency across runs. It also identified an 'LLM head bottleneck' where embeddings from intermediate layers outperform text outputs for numerical regression, and tracked GPT models over 18 months, showing 9-43% performance variation that challenges reproducibility.
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.