GPT Takes the CPA Exam
modelInitial repository for research evaluating GPT models on the CPA exam, later developed into the comprehensive "GPT as Knowledge Worker" project
══════════════════════════════════════════════════════════════════
The initial repository for research on evaluating GPT language models’ performance on the Uniform CPA Examination. This project served as the foundation for the more comprehensive “GPT as Knowledge Worker” research.
Project Status
This repository represents the early stages of the CPA exam evaluation research. The authors recommend referring to the gpt-as-knowledge-worker repository for the complete and current version of this research.
Research Context
Part of the broader investigation into AI capabilities for professional knowledge work, specifically focusing on:
- Accounting and financial reasoning
- Regulatory understanding
- Ethical decision-making
- Professional certification requirements
Publication
The research evolved into the paper:
- GPT as Knowledge Worker: A Zero-Shot Evaluation of (AI)CPA Capabilities
- Authors: Jillian Bommarito, Michael James Bommarito II, Daniel Martin Katz, Jessica Katz
- Published: January 2023
- Also available on arXiv
Key findings: GPT models achieved up to 57.6% accuracy on CPA exam questions with optimal prompting, significantly outperforming random guessing (25%) and showing strong non-entailment with 82.1% top-2 accuracy.