# Why Does ChatGPT Fall Short in Answering Questions Faithfully?

**Authors:** Shen Zheng, Jie Huang, Kevin Chen-Chuan Chang  
**arXiv:** [2304.10513](https://arxiv.org/abs/2304.10513)  
**Date:** 2023-04-20  
**Summary page:** https://www.summarizepaper.com/en/paper/2304.10513v1/

## Key points

- Large Language Models like ChatGPT have potential to impact various aspects of human life
- ChatGPT faces challenges in faithfulness
- In the paper "Why Does ChatGPT Fall Short in Answering Questions Faithfully?", authors analyze the failures of ChatGPT in complex open-domain question answering and identify abilities under the failures
- The authors categorize ChatGPT's failures into four types: comprehension, factualness, specificity, and inference
- Three critical abilities associated with QA failures are identified: knowledge memorization, knowledge association, and knowledge reasoning
- Experiments are conducted centered on these abilities to enhance faithfulness
- Furnishing the model with fine-grained external knowledge, hints for knowledge association and guidance for reasoning can empower the model to answer questions more faithfully
- This research provides insights into improving language models' performance in open-domain question answering tasks by addressing their limitations related to faithfulness.

## Summary

The recent advancements in Large Language Models, such as ChatGPT, have shown significant potential to impact various aspects of human life. However, ChatGPT still faces challenges in aspects like faithfulness. In this paper titled "Why Does ChatGPT Fall Short in Answering Questions Faithfully? ", authors Shen Zheng, Jie Huang, and Kevin Chen-Chuan Chang attempt to analyze the failures of ChatGPT in complex open-domain question answering and identify the abilities under the failures. The authors categorize ChatGPT's failures into four types: comprehension, factualness, specificity, and inference. They further pinpoint three critical abilities associated with QA failures: knowledge memorization, knowledge association and knowledge reasoning. The study conducts experiments centered on these abilities and proposes potential approaches to enhance faithfulness. The results indicate that furnishing the model with fine-grained external knowledge, hints for knowledge association and guidance for reasoning can empower the model to answer questions more faithfully. This research is crucial as it provides insights into improving language models' performance in open-domain question-answering tasks by addressing their limitations related to faithfulness.