一句话：通过近万份生成程序，评估 LLM 温度参数对编程作业作弊代码可提交性的影响。
生成式人工智能的快速发展使自动代码生成成为大学编程课程中的新型作弊策略。本文关注 LLM 的温度参数如何影响生成代码在必修编程作业中的可提交性。作者针对四个编程任务，通过改变温度生成近 10,000 个程序，并对这些程序是否适合提交进行综合评估。该工作为理解温度设置与 AI 生成代码用于作业作弊之间的关系提供了实证依据。
In light of the rapid evolution of generative artificial intelligence, automatic code generation constitutes a novel cheating strategy for university programming courses. Varying the temperature parameter of LLMs, we generate nearly 10,000 programs for four programming tasks and present a comprehensive evaluation of their suitability for submission to mandatory assignments.
@inproceedings{paperbot11637,
title = {Too Hot To Handle: The Effects of Temperature on AI-Generated Code Used to Cheat in Programming Assignments},
author = {Robin Maisch and Timur Sağlam and Larissa Schmid and Sebastian Hahner and Tobias Hey},
booktitle = {ICSE},
year = {2026},
doi = {10.1145/3774748.3795678}
}