It was not that long ago when we shared with you our love story written with the help of AI. And yet, we have another news from this universe that keeps us sleeping with one eye open. Because part of our mission is to disclose to you emerging technologies, we would like to share our excitement about the Codex.
The latest version of this OpenAI system, which is able to translate the human language to code, was released in private beta in August this year. It is based on the famous GPT-3, but it is a deep learning language model which was trained on billions of lines of publicly available source code. The goal is simple: write (or say) the program instructions to the model and you will be provided with the code matching your needs. Not an easy task at all, but Codex can solve over 70% of the programming problems in the publicly available HumanEval test set. Not great, not terrible, one could think … but for sure very impressive compared to 0% success rate of “plain” GPT-3.
See it for yourself!
If you would like to see it in action, definitely rewatch the OpenAI Codex demo here. There are 3 examples in the video, starting with the Hello world email sent together with the current Bitcoin price using the Mailchimp API. The second task was to code a game, in which the person has to dodge a falling boulder. The last one included a speech recognition system that translates voice requests into text and instructs the model to update the Word document by showing Codex the documentation for JavaScript Microsoft Word API.
Sky is the limit
From those examples, the potential of such a technology is obvious even to non-programmers out there (for all Marvel fans: Yes, we are a step closer to J.A.R.V.I.S again!).
And if you are a programmer, just imagine! Getting rid of the boring parts of the job and focusing more on the tasks with higher added value. Codex could write unit tests, generate documentation or reverse-engineer the old, non-documented code and explain what it does. Translation from one programming language to another (or from the older language version to the new one) would be a great time saver as well. And there are many more use cases, for example for educational purposes, you name it!
The first tool which has adapted the technology has become GitHub Copilot. The name really fits, as this technology has the potential to become the common qualified friend who assists or relieves the “in command” programmers. Similarly like we in Optimics are qualified copilots to our clients, helping them to fly high in the digital sky.
