OpenAI, the artificial intelligence research and deployment venture that co-developed the GitHub Copilot AI-based coding tool, has released an API version of the model that powers Copilot.
On August 10, OpenAI released an improved version of its OpenAI Codex AI system, which translates natural language into code, through the OpenAI API in a private beta. Codex is the model that powers Copilot, which was launched as a Visual Studio Code extension. OpenAI is inviting developers and businesses to build on top of OpenAI Codex through the API and has set up a Codex waitlist to access the API.
The Codex model is proficient in more than a dozen programming languages, OpenAI said. Codex can interpret simple commands in natural language and execute them on a user’s behalf, enabling development of a natural language interface to existing applications. OpenAI Codex is a descendant of OpenAI’s GPT-3 AI API, which has been used for applications such as search, conversation, and text completion.
Although OpenAI Codex is most capable in Python, it also is proficient in JavaScript, Go, PHP, Ruby, Swift, and TypeScript, OpenAI said. OpenAI Codex has much of the natural language understanding of GPT-3 while also producing working code. Featuring a general purpose programming model, OpenAI Codex can be applied to essentially any programming task. OpenAI has used it for transpilation as well as for explaining and refactoring code.
GitHub Copilot, while still in a technical preview stage, has drawn criticism since being unveiled in early-July. The Free Software Foundation has called the technology “unacceptable and unjust,” arguing that it requires using software that is not free and constitutes a service as a software substitute. The Free Software Foundation also raised questions about fair use and copyright infringement.