Mechanistic Understanding and Mitigation of Language Model Non-Factual Hallucinations
This addresses the issue of unreliable text generation for users of language models, but it is incremental as it builds on existing interpretability methods.
The paper tackled the problem of non-factual hallucinations in language models by identifying two mechanistic causes—insufficient knowledge enrichment and answer extraction failures—and proposed a targeted mitigation method that showed superior performance compared to baselines.
State-of-the-art language models (LMs) sometimes generate non-factual hallucinations that misalign with world knowledge. To explore the mechanistic causes of these hallucinations, we create diagnostic datasets with subject-relation queries and adapt interpretability methods to trace hallucinations through internal model representations. We discover two general and distinct mechanistic causes of hallucinations shared across LMs (Llama-2, Pythia, GPT-J): 1) knowledge enrichment hallucinations: insufficient subject attribute knowledge in lower layer MLPs, and 2) answer extraction hallucinations: failure to select the correct object attribute in upper layer attention heads. We also found these two internal mechanistic causes of hallucinations are reflected in external manifestations. Based on insights from our mechanistic analysis, we propose a novel hallucination mitigation method through targeted restoration of the LM's internal fact recall pipeline, demonstrating superior performance compared to baselines.