Attention Satisfies: A Constraint-Satisfaction Lens on Factual Errors of Language Models
This work addresses the problem of factual errors in LLMs for users relying on accurate information, offering a mechanistic approach to enhance reliability, though it is incremental in building on existing probing techniques.
The paper investigates how Transformer-based LLMs generate factually incorrect text by modeling factual queries as constraint satisfaction problems, finding a strong positive relationship between attention to constraint tokens and factual accuracy. It introduces SAT Probe, a method that uses attention patterns to predict factual errors across 10 datasets with over 40,000 prompts on Llama-2 models, enabling early error identification.
We investigate the internal behavior of Transformer-based Large Language Models (LLMs) when they generate factually incorrect text. We propose modeling factual queries as constraint satisfaction problems and use this framework to investigate how the LLM interacts internally with factual constraints. We find a strong positive relationship between the LLM's attention to constraint tokens and the factual accuracy of generations. We curate a suite of 10 datasets containing over 40,000 prompts to study the task of predicting factual errors with the Llama-2 family across all scales (7B, 13B, 70B). We propose SAT Probe, a method probing attention patterns, that can predict factual errors and fine-grained constraint satisfaction, and allow early error identification. The approach and findings take another step towards using the mechanistic understanding of LLMs to enhance their reliability.