Our work demonstrates that large language model (LLM) pre-trained on texts
can not only solve pure math word problems, but also physics word
problems-problems to be solved by calculation and inference based on some prior
physical knowledge. We collect and annotate the first physics word problem
dataset-PhysQA, which contains over 1000 junior high school physics word
problems (on Kinematics, Mass&Density, Mechanics, Heat, Electricity). Then we
use OpenAI' s GPT3.5 to generate the answer of these problems and found that
GPT3.5 could automatically solve 49.3% of the problems on zero-shot learning
and 73.2% on few-shot learning. This result show that by using similar problem
and its answer as prompt, LLM could solve elementary physics word problems
approaching human level. Besides automatically solving problems, GPT3.5 could
also summarize the knowledge or topic examined by the problem, generate the
relevant explanation, and synthesis new physics word problems according tothe
input problems.Our work is the first research on automatically solving,
explaining and generating physics word problems of multiple types and scenes,
and we gain an acceptable and state-of-art accuracy, which demonstrates the
potential of LLM's further application in the field of secondary education.

我们的研究表明，基于预训练文本的大型语言模型（LLM）不仅可以解决纯数学问题，还可以解决物理问题 - 即基于某些先前的物理知识进行计算和推理的问题。我们收集和注释了第一个物理问题数据集 - PhysQA，其中包含 1000 多个初中物理问题（关于运动学、质量和密度、力学、热学、电学）。然后我们使用 OpenAI 的 GPT3.5 生成了这些问题的答案，发现 GPT3.5 能够以零样本学习解决 49.3％的问题，以少样本学习解决 73.2％的问题。这个结果表明，通过使用类似问题及其答案作为提示，LLM 可以解决接近人类水平的初级物理问题。除了自动解决问题，GPT3.5 还可以总结问题涉及的知识或主题，生成相关的解释，并根据输入问题合成新的物理问题。我们的研究是首次在多种类型和场景下自动解决、解释和生成物理问题，我们获得了令人满意和领先水平的准确性，这表明了在中等教育领域中 LLM 进一步应用的潜力。