ComputeGPT: A computational chat model for numerical problems

TL;DR

ComputeGPT enhances language models' ability to solve numerical problems by converting questions into code, executing it locally, and returning accurate answers, thus overcoming traditional probabilistic limitations.

Contribution

This paper introduces ComputeGPT, a novel chat model that integrates on-demand code execution for improved numerical problem-solving accuracy.

Findings

Achieves state-of-the-art efficiency on numerical problems

Provides a safe, browser-based environment for code execution

Successfully converts questions into executable code for accurate answers

Abstract

Language models are not accurate in numerical problems. Their architecture does not allow for anything less than a probabilistic next word. This paper introduces ComputeGPT: an approach of creating a chat model able to answer computational problems through running on-demand code. ComputeGPT converts each question to relevant code, runs the code, and returns the computed answer as part of the chat. We combine this approach with a local browser-based Python interpretation and fine-tuned prompts in order to achieve state-of-the-art efficiency on numerical problems and provide a suitable front-end and safe environment for the code to be executed in.