OpenAI can translate English into code with new machine learning software Codex

AI research firm OpenAI launches a new machine learning tool that translates English into code. A program called Codex is designed to speed up the work of professional programmers and help hobbyists get started with programming.

In Codex demos, OpenAI demonstrates how the software can be used to create simple websites and rudimentary games using natural language, as well as translate between different programming languages ​​and process data science queries. Users type commands in English into the program, such as “Create a web page with menu on the side and address on top”, and Codex translates this into code. The program is far from flawless and requires some patience to run, but it can prove invaluable in making getting to the coding quicker and easier.

said Greg Brockman, chief technology officer of OpenAI and co-founder of subscriber. There are two parts to programming: you have to “think hard about a problem and try to understand it” and “assign these little bits to your existing code, whether it’s a library, a function, or an API.” Part two is boring, he says, but that’s the best thing about the constitution. food. People really take programmers and take the hard work away.”

OpenAI used an earlier version of Codex to create a tool called Second Pilot For GitHub, a code repository owned by Microsoft, which is itself a close partner of OpenAI. Copilot is similar to Gmail’s autocomplete tools and offers suggestions for ending lines of code as users type them. Despite this, the new OpenAI version of Codex is more advanced and flexible, and it is not only about completing the code, but also about creating it.

Codex is built on top of GPT-3, the OpenAI language generation model, which has been trained on much of the Internet, allowing for impressive generation and analysis of the written word. One of the uses that GPT-3 users have found is code generation, but Codex improves the capabilities of its predecessors and is specially trained in open source code repositories taken from the Internet.

This last point has led many programmers to complain that OpenAI is unfairly exploiting their work. OpenAI’s Copilot tool often suggests snippets of code written by others, for example, and the software’s entire knowledge base ultimately comes from open source work, which is shared for the benefit of individuals rather than companies. The same criticism will likely be leveled at Codex, although OpenAI says that use of this data is protected by law by fair use.

When asked about these complaints, Brockmann replied, “New technology is coming, we need this discussion, and there will be things that we do so that the community has good points about and we’ll take the feedback and do things differently. However, he argues that the broader crypto community will ultimately benefit.” From OpenAI’s work “The real net impact is very valuable to the ecosystem,” says Brockmann. “Ultimately, I believe these kinds of technologies can reshape our economy and create a better world for all of us.”

Codex will certainly create value for OpenAI and its investors. Although the company started its life as a non-profit lab in 2015, it switched to a “determined profit” model in 2019 to attract external funding, and while Codex initially launched as a free API, OpenAI will be launched at some point in the future. Going to charge for future access.

OpenAI says it does not want to build its own tools with Codex, because it is in a better position to improve the underlying model. “We knew that if we followed one of these paths, we would cut off one of our other paths,” Brockmann says. “You can pick a startup to be the best at one thing. And for us there is no doubt that this makes better versions of all these models.”

Of course, while Codex looks very exciting, it’s hard to judge the full scope of its capabilities before real programmers get to practice it. I’m not a programmer myself, but I’ve seen Codex in action and have some ideas about the software.

Brockman and Codex OpenAI Wojciech Zaremba showed me the software online, using Codex to first create a simple website and then a rudimentary game. In the game’s demo, Brockman found a silhouette of a person on Google Images and asked Codex to “add this image of a person from the page” before pasting the URL. The silhouette appeared on the screen and Brockmann made it smaller (“Make the person a little taller”) before it became controllable (“Now make it controllable with the left and right arrow keys”).

Everything works very smoothly. The song started spinning around the screen, but soon we had a problem: it kept disappearing from the screen. To stop this, Brockmann gave the computer additional instructions: “Continuously check that the person is off the page and put them back on the page if they are.” This kept it out of sight, but I was curious how accurate this instructable was. She suggested trying a different approach: “Make sure the person can’t get off the page.” This worked, too, but for reasons neither Brockman nor Zarimba can explain, he also changed the look of the figure and crushed it on screen.

“Sometimes you don’t know exactly what to ask,” Brockman laughs. I did a few tries and then came up with a command that works without this unwanted change. “So you had to think a little bit about what’s going on, but not too deeply,” he says.

This is fine for our little demo, but it tells us a lot about the limitations of this type of software. It’s not a magical ghost that can read your brain and turn any command into flawless code – and OpenAI doesn’t claim to be. Instead, you have to think and a little trial and error to use it. Codex won’t turn non-programmers into expert programmers overnight, but it’s definitely more intuitive than any other programming language.

OpenAI is optimistic about Codex’s ability to transform programming and computing in general. Brockman says it could help solve the programmer shortage in the United States, while Zarimba sees it as the next step in the historical development of coding.

“What happened to the constitution happened before,” he says. In the early days of the computer, programming was done by making punch cards that had to be inserted into machines, after which people created and refined the first programming languages. “These programming languages ​​are starting to look like English, with vocabulary like ‘print’ or ‘exit’ so that more people can program.” The next part of this path is to get rid of specialized programming languages ​​altogether and replace them with English commands.

“Each of these stages represents the programming languages ​​that are getting higher,” Zaremba says. “And we think Codex will bring computers closer to humans, allowing them to speak English instead of machine code.” Codex itself can speak more than a dozen coding languages, including JavaScript, Go, Perl, PHP, Ruby, Swift, and TypeScript. However, it is most effective in Python.

Codex also has the ability to control other programmes. In a demo, Brockman shows how the program can create an audio interface for Microsoft Word. Since Word has its own API, Codex can enter a code generated from a user’s voice commands. Brockman copies a poem into a Word document and tells Word (via Codex) to first remove all the dashes, then number the lines, then count the repetitions of certain words, etc. It’s very flexible, although it’s hard to say how well it works outside the confines of a pre-arranged demo.

If successful, Codex can not only help programmers, but will become a new interface between users and computers. OpenAI says it tested Codex’s ability to run not only Word, but also other programs like Spotify and Google Calendar. And while the Word demo is just a proof of concept, says Brockmann, Microsoft appears to be already interested in exploring the software’s capabilities. “They are very excited about the model in general and anticipate creating many more Codex applications,” he says.