# Solving math word problems with process- and outcome-based feedback

**Authors:** Jonathan Uesato, Nate Kushman, Ramana Kumar, Francis Song, Noah Siegel, Lisa Wang, Antonia Creswell, Geoffrey Irving, Irina Higgins  
**arXiv:** [2211.14275](https://arxiv.org/abs/2211.14275)  
**Date:** 2022-11-25  
**Summary page:** https://www.summarizepaper.com/en/paper/2211.14275v1/

## Key points

- Study examines supervision methods for language models in solving math word problems
- Researchers compare outcome-based and process-based approaches
- Investigate both final-answer and reasoning errors
- Experiments conducted on the GSM8K task
- Process-based supervision crucial for correct reasoning steps
- Results demonstrate improved performance with reduced final-answer and reasoning errors
- Significance of incorporating process-based feedback in training language models for math problem-solving tasks highlighted

## Summary

The study examines supervision methods for language models in solving math word problems. The researchers compare outcome-based and process-based approaches and investigate both final-answer and reasoning errors. They conduct experiments on the GSM8K task and show that process-based supervision is crucial for correct reasoning steps. The results demonstrate improved performance with reduced final-answer and reasoning errors. This highlights the significance of incorporating process-based feedback in training language models for math problem-solving tasks.