A study by Researchers from Brown University and Chinese unis tested if ChatGPT could effectively run a fictitious software company called ChatDev. Tasks were divided into planning, coding, testing, and writing instructions.

Take aways:

  • Automated Roles: ChatGPT instances took roles like CEO, CTO, and programmers. They communicated autonomously to complete software tasks, from language selection to error detection.
  • Efficiency and Cost: ChatDev completed software development in less than seven minutes and for less than a dollar. About 86.66% of the software was “perfectly done.”
  • Challenges and Biases: The study noted that language models like ChatGPT can have mistakes or biases, which can be challenges in software development.
  • Implications: The study suggests that AI can be efficient and cost-effective in software development and could help new programmers or engineers.