Abstract

From topics such as" Religion and Philosophy" to" Health/Fitness" and" Technology," people rely on large language models (LLMs) daily to get information and answer questions [Costa-Gomes et al., 2025]. Such extensive use cases indicate a need to broaden the scope and applicability of these models, making them more and more ubiquitous. However, there are many open and ongoing challenges to reach such a point. This thesis introduces new methodologies and analyses that enable LLMs to process more data, handle more data types, and generate better-aligned responses.

Date

2026

Authors

Kian Ahrabian

Institution

University of Southern California