RELIC: Investigating Large Language Model Responses using Self-Consistency

TL;DR

RELIC tackles the problem of LLM hallucinations by shifting from token-level confidence to semantic-level self-consistency across multiple samples. It introduces a self-consistency checking pipeline that decomposes long-form text into atomic claims, converts them into questions, and analyzes answers across samples using NLI and QA models to produce evidence-linked confidence. The RELIC system provides a multi-view interface (Response, Claim, Evidence) with keyword annotations, brushing for questioning, and editable generation to enable what-if analyses and targeted corrections. A formative study and a user study with ten participants demonstrate that RELIC improves users’ ability to verify and correct generated content, revealing design implications for collaborative, transparent, and user-empowered human-LLM interactions.

Abstract

Large Language Models (LLMs) are notorious for blending fact with fiction and generating non-factual content, known as hallucinations. To address this challenge, we propose an interactive system that helps users gain insight into the reliability of the generated text. Our approach is based on the idea that the self-consistency of multiple samples generated by the same LLM relates to its confidence in individual claims in the generated texts. Using this idea, we design RELIC, an interactive system that enables users to investigate and verify semantic-level variations in multiple long-form responses. This allows users to recognize potentially inaccurate information in the generated text and make necessary corrections. From a user study with ten participants, we demonstrate that our approach helps users better verify the reliability of the generated text. We further summarize the design implications and lessons learned from this research for future studies of reliable human-LLM interactions.

Figures (7)

• Figure 1: Pipeline of computing the self-consistency of individual claims (Algorithm \ref{['algo:algo_main']}). The input is a user prompt with which an LLM is invoked multiple times. Afterwards, the atomic claims are generated based on the top text response, then turned into questions and answered by all other generations. The answers are then clustered together based on their meaning (e.g., Spanish and from Spain).
• Figure 2: The Keyword Annotation uses a small word-scale visualization (left) to display the proportions of different categories of samples or alternatives. Upon clicking, users are able to view a list of alternatives and inspect their details (right).
• Figure 3: The alternative designs for the Keyword Annotation. The final design is on the right side of the second row.
• Figure 4: Questionnaire results. Overall, the 10 participants reported high usefulness and satisfaction.
• Figure 5: Increasing the number of additional samples improves Consistency Scores, but with diminishing returns.