Exploring How LLMs Capture and Represent Domain-Specific Knowledge

Mirian Del Carmen Hipolito Garcia, Camille Couturier, Daniel Madrigal, Ankur Mallick, Robert Sim, Anastasios Kyrillidis, Victor Rühle, Saravan Rajmohan

Keywords: Large Language Models, domain-trajectories, hidden states, prefill-phase, model selection.

TL;DR: We investigate whether LLMs capture domain-specific nuances in natural language. We test the domain sensitivity of LLMs by examining their ability to distinguish queries from different domains using hidden states generated during the prefill phase.

Abstract: We study whether Large Language Models (LLMs) inherently capture domain-specific nuances in natural language. Our experiments probe the domain sensitivity of LLMs by examining their ability to distinguish queries from different domains using hidden states generated during the prefill phase. We reveal latent domain-related trajectories that indicate the model's internal recognition of query domains. We also study the robustness of these domain representations to variations in prompt styles and sources. Our approach leverages these representations for model selection, mapping the LLM that best matches the domain trace of the input query (i.e., the model with the highest performance on similar traces). Our findings show that LLMs can differentiate queries for related domains, and that the fine-tuned model is not always the most accurate. Unlike previous work, our interpretations apply to both closed and open-ended generative tasks.

Primary Area: unsupervised, self-supervised, semi-supervised, and supervised representation learning

Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.

Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.

Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.

No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.

Submission Number: 11324