Knowledge Mechanisms in Large Language Models: A Survey and Perspective

TL;DR

This paper surveys the mechanisms of knowledge in Large Language Models, introducing a new taxonomy to analyze how knowledge is utilized and evolves, and discusses challenges like knowledge fragility and dark knowledge.

Contribution

It presents a novel taxonomy for understanding knowledge mechanisms in LLMs, covering knowledge utilization and evolution, and offers insights into knowledge fragility and dark knowledge.

Findings

Knowledge utilization involves memorization, comprehension, application, and creation.

Knowledge evolution examines how knowledge changes within and across LLMs.

Fragility of parametric knowledge and dark knowledge pose significant challenges.

Abstract

Understanding knowledge mechanisms in Large Language Models (LLMs) is crucial for advancing towards trustworthy AGI. This paper reviews knowledge mechanism analysis from a novel taxonomy including knowledge utilization and evolution. Knowledge utilization delves into the mechanism of memorization, comprehension and application, and creation. Knowledge evolution focuses on the dynamic progression of knowledge within individual and group LLMs. Moreover, we discuss what knowledge LLMs have learned, the reasons for the fragility of parametric knowledge, and the potential dark knowledge (hypothesis) that will be challenging to address. We hope this work can help understand knowledge in LLMs and provide insights for future research.