From Linguistic Nuance to Social Simulation: Evaluating the Capabilities and Fidelity of Large Language Models as Human Proxies
MCML Authors
Bolei Ma
Dr.
Abstract
Bolei Ma
Dr.
Abstract
This dissertation investigates the potential and limitations of large language models as proxies for human behavior, examining their linguistic understanding, multilingual and cultural competence, and ability to simulate social perspectives. It identifies limitations in handling linguistic ambiguity, cultural nuances, and diverse human opinions, while proposing improved evaluation and data alignment methods to make LLM-based social simulations more representative and reliable. (Shortened.)
BibTeXKey: Ma26