2023-What Makes Good In-context Demonstrations for Code Intelligence Tasks with LLMs?

[skyleaworlder]

https://arxiv.org/pdf/2304.07575

Shuzheng Gao1†, Xin-Cheng Wen1, Cuiyun Gao1∗, Wenxuan Wang2, Hongyu Zhang3, Michael R. Lyu2
1 School of Computer Science and Technology, Harbin Institute of Technology, Shenzhen, China
2 Department of Computer Science and Engineering, The Chinese University of Hong Kong, China
3 School of Big Data and Software Engineering, Chongqing University, China
szgao98@gmail.com, xiamenwxc@foxmail.com, gaocuiyun@hit.edu.cn, hyzhang@cqu.edu.cn, {wxwang,lyu}@cse.cuhk.edu.hk

ASE '23, cite 30+

from ASE '23 research track. I need to graduate on time!

本文认为 few-shot 的选择、顺序、个数三个方面，会影响 llm 在 code summary、program repair、program synthesis 三个领域的能力。