OpenAI Codex: Exploration and Review of the Platform & API
Done in context of a miniproject in the course AI applications at university of applied sciences OST, written by Nick Wallner, Mirio Eggmann
1 Introduction
The goal of this project is to use and evaluate OpenAI Codex. Codex is a natural language processing model, based on OpenAI GPT-3, that can generate working code in over a dozen of programming languages. It enables a user to write code just by giving instructions in a natural language like English, French or even Swiss dialect. We will use Codex to find instructions on how to create a working Pong-Game in OpenAI’s provided JavaScript Sandbox. Furthermore, we will integrate it into a minimal terminal application to enable a user to get detailed information about how a particular function works.
2 Installation and Setup
Codex is still in closed beta. We have one account with access to the OpenAI Codex beta-platform and can review it on the following two sites or through the API:
- https://beta.openai.com/playground
- https://beta.openai.com/codex-javascript-sandbox
- https://api.openai.com/v1/
We also got access to GitHub Copilot, which is implemented using Codex, but our focus is on Codex itself.