Are Human Conversations Special? A Large Language Model Perspective

arXiv:2403.05045v1 Announce Type: cross
Abstract: This study analyzes changes in the attention mechanisms of large language models (LLMs) when used to understand natural conversations between humans (human-human). We analyze three use cases of LLMs: interactions over web content, code, and mathematical texts. By analyzing attention distance, dispersion, and interdependency across these domains, we highlight the unique challenges posed by conversational data. Notably, conversations require nuanced handling of long-term contextual relationships and exhibit higher complexity through their attention patterns. Our findings …

abstract analyze arxiv attention attention mechanisms cases code conversations cs.ai cs.cl cs.lg human humans interactions language language model language models large language large language model large language models llms natural perspective study type use cases web