Can the Inference Logic of Large Language Models be Disentangled into Symbolic Concepts?

TL;DR

This paper investigates whether the inference process of large language models can be broken down into symbolic concepts, demonstrating that sparse symbolic concepts can explain and transfer across inputs and clarify prediction errors.

Contribution

It proposes a method to disentangle LLM inference scores into symbolic concepts and shows their transferability and explanatory power for model predictions.

Findings

Sparse symbolic concepts can accurately estimate LLM inference scores.

Symbolic concepts transfer well across similar input sentences.

Concepts help explain LLM prediction errors.

Abstract

In this paper, we explain the inference logic of large language models (LLMs) as a set of symbolic concepts. Many recent studies have discovered that traditional DNNs usually encode sparse symbolic concepts. However, because an LLM has much more parameters than traditional DNNs, whether the LLM also encodes sparse symbolic concepts is still an open problem. Therefore, in this paper, we propose to disentangle the inference score of LLMs for dialogue tasks into a small number of symbolic concepts. We verify that we can use those sparse concepts to well estimate all inference scores of the LLM on all arbitrarily masking states of the input sentence. We also evaluate the transferability of concepts encoded by an LLM and verify that symbolic concepts usually exhibit high transferability across similar input sentences. More crucially, those symbolic concepts can be used to explain the exact reasons accountable for the LLM's prediction errors.