LUQ: Long-text Uncertainty Quantification for LLMs
This addresses the challenge of mitigating nonfactual outputs in LLMs for real-world applications requiring long responses, representing a domain-specific incremental advance.
The paper tackles the problem of uncertainty quantification (UQ) for long-text generation in LLMs, which is limited by existing methods focused on short text, and introduces LUQ and its variations, showing that LUQ outperforms baselines with a correlation coefficient of -0.85 for factuality scores and LUQ-Ensemble improves response factuality over standalone models.
Large Language Models (LLMs) have demonstrated remarkable capability in a variety of NLP tasks. However, LLMs are also prone to generate nonfactual content. Uncertainty Quantification (UQ) is pivotal in enhancing our understanding of a model's confidence on its generation, thereby aiding in the mitigation of nonfactual outputs. Existing research on UQ predominantly targets short text generation, typically yielding brief, word-limited responses. However, real-world applications frequently necessitate much longer responses. Our study first highlights the limitations of current UQ methods in handling long text generation. We then introduce \textsc{Luq} and its two variations, a series of novel sampling-based UQ approaches specifically designed for long text. Our findings reveal that \textsc{Luq} outperforms existing baseline methods in correlating with the model's factuality scores (negative coefficient of -0.85 observed for Gemini Pro). To further improve the factuality of LLM responses, we propose \textsc{Luq-Ensemble}, a method that ensembles responses from multiple models and selects the response with the lowest uncertainty. The ensembling method greatly improves the response factuality upon the best standalone LLM.