The first version of the OpenAI Codex aims to save developers time by interpreting spoken comments and coding themselves to suggest lines of code. However, the model will be equipped with more impressive capabilities.
The new version in development (currently in private beta) should already be able to generate complex code from simple natural language instructions. In an announcement, OpenAI states that Codex, which has a lot of natural language understanding of GPT-3, has been specifically trained to produce working code. In addition, it is now possible to send commands in English to any program via the API. “OpenAI Codex is a universal programming paradigm, which means it can be applied to any programming task (even if results may vary)”, As described in the OpenAI blog.
When it was introduced, GitHub Copilot solved the problem in the community of AI developers and experts due to copyright issues and the quality of the automatically generated code. Controversy as you can read here.
In its latest announcement, OpenAI states that the primary purpose of Codex is to “do the most mundane thing a developer can do,” which is to map simple problems in existing code (libraries, APIs, or functions) so that developers can focus on more creative tasks .
In the following video, OpenAI demonstrates how the new capabilities of Codex can be used to create a simple video game:
“Subtly charming coffee scholar. General zombie junkie. Introvert. Alcohol nerd. Travel lover. Twitter specialist. Freelance student.”