Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

VERITAS: A Unified Approach to Reliability Evaluation

Rajkumar Ramamurthy, Meghana Arakkal Rajeev, Oliver Molenschot, James Zou, Nazneen Rajani

TL;DR

VERITAS is introduced, a family of hallucination detection models designed to operate flexibly across diverse contexts while minimizing latency and costs, and achieves state-of-the-art results.

Abstract

Large language models (LLMs) often fail to synthesize information from their context to generate an accurate response. This renders them unreliable in knowledge intensive settings where reliability of the output is key. A critical component for reliable LLMs is the integration of a robust fact-checking system that can detect hallucinations across various formats. While several open-access fact-checking models are available, their functionality is often limited to specific tasks, such as grounded question-answering or entailment verification, and they perform less effectively in conversational settings. On the other hand, closed-access models like GPT-4 and Claude offer greater flexibility across different contexts, including grounded dialogue verification, but are hindered by high costs and latency. In this work, we introduce VERITAS, a family of hallucination detection models designed to operate flexibly across diverse contexts while minimizing latency and costs. VERITAS achieves state-of-the-art results considering average performance on all major hallucination detection benchmarks, with $10\%$ increase in average performance when compared to similar-sized models and get close to the performance of GPT4 turbo with LLM-as-a-judge setting.

VERITAS: A Unified Approach to Reliability Evaluation

TL;DR

VERITAS is introduced, a family of hallucination detection models designed to operate flexibly across diverse contexts while minimizing latency and costs, and achieves state-of-the-art results.

Abstract

Large language models (LLMs) often fail to synthesize information from their context to generate an accurate response. This renders them unreliable in knowledge intensive settings where reliability of the output is key. A critical component for reliable LLMs is the integration of a robust fact-checking system that can detect hallucinations across various formats. While several open-access fact-checking models are available, their functionality is often limited to specific tasks, such as grounded question-answering or entailment verification, and they perform less effectively in conversational settings. On the other hand, closed-access models like GPT-4 and Claude offer greater flexibility across different contexts, including grounded dialogue verification, but are hindered by high costs and latency. In this work, we introduce VERITAS, a family of hallucination detection models designed to operate flexibly across diverse contexts while minimizing latency and costs. VERITAS achieves state-of-the-art results considering average performance on all major hallucination detection benchmarks, with increase in average performance when compared to similar-sized models and get close to the performance of GPT4 turbo with LLM-as-a-judge setting.

Paper Structure

This paper contains 30 sections, 1 figure, 5 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Reliability Evaluation
  4. Benchmarks and Metrics
  5. VERITAS Data Collection
  6. Textual Entailment
  7. Grounded Question Answering
  8. Summarization
  9. Generation of hallucination samples
  10. Grounded Dialogue Verification
  11. Spanish data
  12. Rationale Generation and Data Cleansing
  13. VERITAS
  14. Classifier models
  15. Generative models
  16. ...and 15 more sections

Figures (1)

  • Figure 1: VERITAS models provide a unified interface for hallucination detection using a multi-task approach comprising of three tasks. 1) NLI task in which the claim or the summary of the given document is checked/ verified 2) Grounded QA in which the answer is assessed for factuality 3) Grounded Dialogue in which the assistant responses are verified. In all these tasks, evaluation is performed based on the given document. Label 1 indicates no hallucination (content is factually consistent), while Label 0 denotes factual inconsistencies.