GPT Can Solve Mathematical Problems Without a Calculator
Yang, Zhen, Ding, Ming, Lv, Qingsong, Jiang, Zhihuan, He, Zehai, Guo, Yuyi, Bai, Jinfeng, Tang, Jie
–arXiv.org Artificial Intelligence
Previous studies have typically assumed that large language models are unable to accurately perform arithmetic operations, particularly multiplication of >8 digits, and operations involving decimals and fractions, without the use of calculator tools. This paper aims to challenge this misconception. With sufficient training data, a 2 billion-parameter language model can accurately perform multi-digit arithmetic operations with almost 100% accuracy without data leakage, significantly surpassing GPT-4 (whose multi-digit multiplication accuracy is only 4.3%). We also demonstrate that our MathGLM, fine-tuned from GLM-10B on a dataset with additional multi-step arithmetic operations and math problems described in text, achieves similar performance to GPT-4 on a 5,000-samples Chinese math problem test set. Our code and data are public at https://github.com/THUDM/MathGLM.
arXiv.org Artificial Intelligence
Sep-12-2023
- Genre:
- Research Report > New Finding (0.67)
- Industry:
- Education > Educational Setting (0.46)
- Technology: