OpenAI, the company behind GPT-3, in collaboration with Microsoft and GitHub, released Codex in August 2021. And it’s touted as a victory for the programming community, as the technology will no longer require developers to learn how to code in different languages to create applications. But how did Codex come about, exactly?
- What Is Codex?
- Codex’s Journey into Becoming a Reality
- Codex Applications Seen So Far
- Does Codex Aim to Replace Programmers?
What Is Codex?
OpenAI’s Codex has been dubbed “a master of coding.” It can understand dozens of programming languages and natural language (English) and interpret them to create commands that a computer can perform.
Interested users can access Codex through a newly launched application programming interface (API), but they may have to wait to be allowed in.
OpenAI also launched a challenge where programmers can code against and with Codex. Two of its researchers, Greg Brockman and Ilya Sutskever, also conducted a live demo to show off Codex’s capabilities.
Codex’s Journey into Becoming a Reality
OpenAI treats Codex as GPT -3’s younger sibling.
GPT-3 was the computing industry’s master of language. It could mimic practically any writer’s style to make original compositions. It can also write articles from scratch using data obtained from various sources.
While using GPT-3, researcher Sharif Shameem noticed that GPT-3 could help with coding. That led to the development of GitHub Copilot, which aimed to help developers with their repetitive day-to-day tasks. Copilot can read documentation and write unit tests. It serves as a mighty autocomplete system (for coding, specifically).
You may be wondering what Copilot has to do with Codex by now. The answer is that Codex fuels copilot.
Codex isn’t a new version of GPT-3, though. Unlike GPT-3, which lacked memory, Codex has a memory storage capacity that is three times as big as the former. As such, Codex can access previous information without encountering errors. This capability allows Codex to understand what the task at hand requires better and fulfills it.
How Does Codex Work?
To create a program, developers need to identify tasks that a computer needs to perform in sequential order to produce the intended outcome. Those tasks, written in human-readable text (natural language in computing terms), need to be converted into machine-readable text (programming language or code). Only after the conversion can the computer do the tasks and produce the expected output.
OpenAI’s Codex does away with the conversion from human- to machine-readable language. It can make computers make sense of natural language to develop the desired end product like a game or website.
To see Codex in action, you can watch the OpenAI demo here:
Codex Applications Seen So Far
Despite its very recent launch, OpenAI’s Codex usage has been gaining traction and fast. To date, it has been utilized to:
- Instantly create memes
- Develop games
- Allowing contactless control for people with disabilities
- Create programs using languages other than English
- Solve math word problems
- Auto-generating captions for videos (speech-to-text)
- Building personal websites
- Creating login forms
- Analyzing data in natural language
- Change the way websites look when browsing online
- Creating web scrapers
- Making three-dimensional (3D) graphics
- Altering augmented reality (AR)
You can learn more about these applications from this video:
Does Codex Aim to Replace Programmers?
The rationale behind OpenAI’s Codex is not to make a computer in charge of programming. Codex should simply serve as a programmer’s assistant. That helper takes conceptual ideas and produces the code to turn them into software or any other technology.
According to OpenAI CTO and co-founder Greg Brokman, Codex is “a tool to multiply programmers.” He said programming has two components, and programmers and Codex will be responsible for one part each.
The first part requires the programmer to “think hard about a problem and try to understand it.” That essentially translates to conceptualizing and designing technology. Codex plays a role in the second part (which humans find repetitive and don’t want to do), “mapping those small pieces to existing code, whether it’s a library, a function, or an API.”
Given everything you’ve read so far, we can conclude that OpenAI’s Codex can transform programming. In fact, it may already be, as evidenced by its featured applications. Instead of fearing it, however, developers should take it as a boon. Now they have a trustworthy assistant that can take care of tedious tasks and focus more on what they like doing—dreaming up new technologies to fill the world with day in and day out.