Research suggests that tutors should adopt a strategic approach when
addressing math errors made by low-efficacy students. Rather than drawing
direct attention to the error, tutors should guide the students to identify and
correct their mistakes on their own. While tutor lessons have introduced this
pedagogical skill, human evaluation of tutors applying this strategy is arduous
and time-consuming. Large language models (LLMs) show promise in providing
real-time assessment to tutors during their actual tutoring sessions, yet
little is known regarding their accuracy in this context. In this study, we
investigate the capacity of generative AI to evaluate real-life tutors'
performance in responding to students making math errors. By analyzing 50
real-life tutoring dialogues, we find both GPT-3.5-Turbo and GPT-4 demonstrate
proficiency in assessing the criteria related to reacting to students making
errors. However, both models exhibit limitations in recognizing instances where
the student made an error. Notably, GPT-4 tends to overidentify instances of
students making errors, often attributing student uncertainty or inferring
potential errors where human evaluators did not. Future work will focus on
enhancing generalizability by assessing a larger dataset of dialogues and
evaluating learning transfer. Specifically, we will analyze the performance of
tutors in real-life scenarios when responding to students' math errors before
and after lesson completion on this crucial tutoring skill.

该研究调查两个生成模型在评估真实教师在应对学生数学错误方面的表现能力，发现 GPT-3.5-Turbo 和 GPT-4 都能够熟练评估与学生犯错有关的标准，但在识别学生错误的情况下存在局限性。未来的研究将致力于通过评估更多对话数据集和评估学习转化来提升泛化能力，进一步分析教师在真实场景中应对学生数学错误的表现。

利用大型语言模型评估导师应对学生数学错误的表现

Using Large Language Models to Assess Tutors' Performance in Reacting to  Students Making Math Errors

Research suggests that providing specific and timely feedback to human tutors
enhances their performance. However, it presents challenges due to the
time-consuming nature of assessing tutor performance by human evaluators. Large
language models, such as the AI-chatbot ChatGPT, hold potential for offering
constructive feedback to tutors in practical settings. Nevertheless, the
accuracy of AI-generated feedback remains uncertain, with scant research
investigating the ability of models like ChatGPT to deliver effective feedback.
In this work-in-progress, we evaluate 30 dialogues generated by GPT-4 in a
tutor-student setting. We use two different prompting approaches, the zero-shot
chain of thought and the few-shot chain of thought, to identify specific
components of effective praise based on five criteria. These approaches are
then compared to the results of human graders for accuracy. Our goal is to
assess the extent to which GPT-4 can accurately identify each praise criterion.
We found that both zero-shot and few-shot chain of thought approaches yield
comparable results. GPT-4 performs moderately well in identifying instances
when the tutor offers specific and immediate praise. However, GPT-4
underperforms in identifying the tutor's ability to deliver sincere praise,
particularly in the zero-shot prompting scenario where examples of sincere
tutor praise statements were not provided. Future work will focus on enhancing
prompt engineering, developing a more general tutoring rubric, and evaluating
our method using real-life tutoring dialogues.

研究发现，提供具体及时的反馈可以提高人类导师的表现，但由于评估导师表现的时间消耗性质，给出具体及时的反馈存在挑战，然而使用大语言模型 (如 AI-chatbot ChatGPT) 来为实际应用中的导师提供建设性反馈有潜力。 这项工作在导师 - 学生模式下评估了由 GPT-4 生成的 30 次对话，并将两种不同的提示方法进行比较：零 - shot 思维链和少 - shot 思维链，以识别基于五个标准的有效赞扬的具体组成部分，并通过与人类评分员的结果进行比较来评估 GPT-4 是否能够准确识别每个赞扬标准。研究发现，零 - shot 和少 - shot 思维链方法产生了相似的结果。 GPT-4 在识别导师提供具体和即时赞扬的情况下表现良好，但在识别导师提供真诚赞扬的能力方面表现不佳，尤其是在没有提供真诚导师赞扬语句的零 - shot 提示场景中。未来的研究将着重于增强提示工程，开发更普遍的导师评分表，并使用实际的导师对话来进行评估。