# Can ChatGPT Understand Too? A Comparative Study on ChatGPT and Fine-tuned BERT

**Authors:** Qihuang Zhong, Liang Ding, Juhua Liu, Bo Du, Dacheng Tao  
**arXiv:** [2302.10198](https://arxiv.org/abs/2302.10198)  
**Date:** 2023-02-19  
**Summary page:** https://www.summarizepaper.com/en/paper/2302.10198v1/

## Key points

- Study titled "Can ChatGPT Understand Too? A Comparative Study on ChatGPT and Fine-tuned BERT" by Qihuang Zhong, Liang Ding, Juhua Liu, Bo Du, and Dacheng Tao
- Evaluation of ChatGPT on the GLUE benchmark and comparison with four fine-tuned BERT-style models
- ChatGPT struggles with paraphrase and similarity tasks but excels in inference tasks
- Comparable performance to BERT in sentiment analysis and question answering tasks
- Highlighted bad cases from inference tasks indicating potential limitations of ChatGPT
- Comparative analysis provides insights into strengths and weaknesses of ChatGPT compared to fine-tuned BERT models
- Contributes to understanding language model capabilities and suggests avenues for further exploration in natural language processing technologies

## Summary

In their study titled "Can ChatGPT Understand Too? A Comparative Study on ChatGPT and Fine-tuned BERT," authors Qihuang Zhong, Liang Ding, Juhua Liu, Bo Du, and Dacheng Tao delve into the understanding ability of ChatGPT. The study evaluates ChatGPT on the popular GLUE benchmark and compares it with four fine-tuned BERT-style models to provide a quantitative analysis of its understanding capability. <br>
Their findings reveal several key insights: Firstly, ChatGPT struggles in handling paraphrase and similarity tasks but excels in inference tasks. It also demonstrates comparable performance to BERT in sentiment analysis and question answering tasks. However, the study highlights some bad cases from inference tasks that showcase potential limitations of ChatGPT. <br>
Overall, this comparative analysis provides valuable insights into the strengths and weaknesses of ChatGPT in terms of understanding abilities compared to fine-tuned BERT models. It contributes to a deeper understanding of the capabilities of language models like ChatGPT and opens up avenues for further exploration and improvement in natural language processing technologies.