Can the Inference Logic of Large Language Models be Disentangled into Symbolic Concepts?
This work addresses the interpretability challenge for LLMs, which is crucial for researchers and practitioners in AI, though it appears incremental as it builds on prior findings about symbolic concepts in traditional DNNs.
The paper tackles the problem of explaining the inference logic of large language models (LLMs) by disentangling it into symbolic concepts, showing that these sparse concepts can accurately estimate inference scores across various input states and explain prediction errors.
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.