Listen to the Context: Towards Faithful Large Language Models for Retrieval Augmented Generation on Climate Questions

2025-05-21

AI Summary

Large language models (LLMs) in climate-domain retrieval-augmented generation (RAG) frequently exhibit factual hallucinations and produce outputs unfaithful to retrieved documents. Method: We propose a fidelity-oriented evaluation and optimization paradigm: (1) an automated atomic claim faithfulness metric; (2) a novel data purification strategy based on unfaithful sample filtering to significantly improve training data cleanliness; and (3) context-aware instruction fine-tuning built upon ClimateGPT. Contribution/Results: The resulting ClimateGPT Faithful+ model achieves a 57% atomic claim support rate on climate QA tasks—up 27 percentage points from the baseline 30%—demonstrating for the first time the decisive impact of data cleanliness on RAG fidelity. This work establishes a reproducible methodological framework for trustworthy climate AI.

Technology Category

• Natural Language Processing: Fact-Checking / Misinformation Detection (NLP Focus)
• Machine Learning: Calibration & Uncertainty Quantification
• Data Mining & Knowledge Management: Representing, Reasoning, and Using Provenance, Trust

Application Category

• Search and Retrieval-Augmented AI: Retrieval-Augmented Generation (RAG) and multi-modal RAG
• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions
• Economics, Online Markets and Human Computation: Trust and reliance of crowd workers and data experts on GenAI

Abstract

Large language models that use retrieval augmented generation have the potential to unlock valuable knowledge for researchers, policymakers, and the public by making long and technical climate-related documents more accessible. While this approach can help alleviate factual hallucinations by relying on retrieved passages as additional context, its effectiveness depends on whether the model's output remains faithful to these passages. To address this, we explore the automatic assessment of faithfulness of different models in this setting. We then focus on ClimateGPT, a large language model specialised in climate science, to examine which factors in its instruction fine-tuning impact the model's faithfulness. By excluding unfaithful subsets of the model's training data, we develop ClimateGPT Faithful+, which achieves an improvement in faithfulness from 30% to 57% in supported atomic claims according to our automatic metric.

Problem

• Assessing faithfulness of retrieval-augmented LLMs in climate science
• Improving model output fidelity to retrieved climate passages
• Enhancing ClimateGPT's faithfulness via data filtering and fine-tuning

Innovation

• Retrieval augmented generation for climate knowledge
• Automatic faithfulness assessment of model outputs
• Training data exclusion to improve model faithfulness