Meaning without reference in large language models

TL;DR

This paper argues that large language models likely capture important aspects of human-like meaning through their internal state relationships, despite skepticism about their understanding of concepts.

Contribution

It proposes that meaning in LLMs arises from internal conceptual roles, offering a new perspective on their cognitive capabilities beyond architecture and training data.

Findings

Meaning in LLMs is linked to internal state relationships.

This approach explains LLM success and guides making models more human-like.

Internal states' relationships are key to understanding LLMs' conceptual role.

Abstract

The widespread success of large language models (LLMs) has been met with skepticism that they possess anything like human concepts or meanings. Contrary to claims that LLMs possess no meaning whatsoever, we argue that they likely capture important aspects of meaning, and moreover work in a way that approximates a compelling account of human cognition in which meaning arises from conceptual role. Because conceptual role is defined by the relationships between internal representational states, meaning cannot be determined from a model's architecture, training data, or objective function, but only by examination of how its internal states relate to each other. This approach may clarify why and how LLMs are so successful and suggest how they can be made more human-like.