Solving Quantitative Reasoning Problems with Language Models

TL;DR

This paper introduces Minerva, a large language model trained on technical content that significantly improves quantitative reasoning performance on scientific problems without external tools.

Contribution

The paper presents Minerva, a novel language model trained on technical data, achieving state-of-the-art results in scientific quantitative reasoning tasks.

Findings

Achieves state-of-the-art performance on technical benchmarks

Correctly answers nearly one-third of undergraduate science problems

Performs well without external tools

Abstract

Language models have achieved remarkable performance on a wide range of tasks that require natural language understanding. Nevertheless, state-of-the-art models have generally struggled with tasks that require quantitative reasoning, such as solving mathematics, science, and engineering problems at the college level. To help close this gap, we introduce Minerva, a large language model pretrained on general natural language data and further trained on technical content. The model achieves state-of-the-art performance on technical benchmarks without the use of external tools. We also evaluate our model on over two hundred undergraduate-level problems in physics, biology, chemistry, economics, and other sciences that require quantitative reasoning, and find that the model can correctly answer nearly a third of them.