AI research firm OpenAI is releasing a new machine learning tool that translates English into code. The The software is called Codex and is designed to speed up the work of professional programmers as well as help enthusiasts start coding.
In Codex demos, OpenAI shows how software can be used to create simple websites and rudimentary games in natural language, as well as compile between different programming languages and answer data science queries. Users type English commands into the software, such as “create a web page with a menu on the page and a title at the top,” and Codex translates this into code. The software is far from infallible and requires some patience to work, but it can prove invaluable for speeding up and making encoding easier.
“We see this as a tool for multiplying programmers,” said Greg Brockman, chief technology officer and founder of OpenAI. Limit. “There are two parts to programming: you have to ‘consider the problem and try to understand it’ and ‘combine these little pieces with existing code, be it a library, a function, or an API. “” The second part is boring, he says, but Codex is the best at it. “It requires people who are already programmers and eliminates the rough work.”
OpenAI used an earlier version of Codex to build a tool called Flight attendant GitHub, a Microsoft-owned code repository that is itself a A close partner of OpenAI. Copilot is similar to Gmail’s autocomplete tools, and offers suggestions for refining lines of code as users type. However, the new version of OpenAI Codex is much more advanced and flexible, not just executing code, but creating it.
Codex is built on OpenTie’s GPT-3 language generation model, which is educated for much of the Internet, and as a result it can create and structure a written word in impressive ways. One application user found for GPT-3 created the code, but Codex enhances the capabilities of its predecessors and is specifically trained in open source repositories captured from the web.
This latter point has caused many coders to complain about OpenAI unfairly benefit from their work. For example, OpenAI’s Copilot tool often suggests code snippets written by others, and the entire program’s knowledge base is ultimately derived from open source work that is shared for the benefit of individuals, not companies. The same criticism is likely to be leveled at Codex, although OpenAI says the use of its data is legally protected by fair use.
When asked about these complaints, Brockman replies, “New technology is coming, we need this debate, and we’re doing things where the community has good points, and we’re taking feedback and doing things differently.” However, he argues that the wider coding community will ultimately benefit from the work of OpenAI. “The real net impact is a lot of value to the ecosystem,” Brockman says. “After all, I think these types of technologies can change our economy and create a better world for all of us.”
Codex is sure to create value for OpenAI as well as its investors. Although the company began its life as a non-profit laboratory in 2015, it did switched to the “covered profit” model in 2019 to attract outside funding, and although Codex will initially be released as a free API, OpenAI will start charging for access at some point in the future.
OpenAI says it doesn’t want to build its own tools using Codex because it has a better chance of improving the core model. “We realized that if we continued any of these, we would cut off all our other routes,” Brockman says. “You can choose as a startup to be the best together. And we have no doubt that it makes better versions of all these models. “
Of course, while Codex sounds very exciting, it’s hard to evaluate all of its features before the real programmers get their hands on it. I’m not a coder myself, but I saw Codex work and I had a few thoughts on the software.
Openai’s Brockman and Codex Director Wojciech Zaremba presented the program to me online and first created a simple website using Codex and then a rudimentary game. In the game demo, Brockman found a person’s silhouette in Google Images and told Codex to add that person’s image from the page before pasting it into the URL. The silhouette appeared on the screen, and Brockman resized it (“make the person a little bigger”) before it became controllable (“make it now controllable with the left and right arrow keys”).
Everything worked very smoothly. The chapter started to get mixed up around the screen, but we soon ran into a problem: it constantly disappeared off the screen. To end this, Brockman gave the computer additional instruction: “Constantly check to see if the person is off the page, and put it back on the page if there is.” This prevented it from going out of sight, but I was curious how accurate these instructions must be. I suggested we try another: “Make sure the person can’t leave the page.” This also worked, but for reasons that Brockman or Zaremba can’t explain, it also changed the width of the image and squeezed it flat on the screen.
“Sometimes it doesn’t know exactly what you’re asking,” Brockman laughs. He still has a few attempts, and then he issues a command that works without this unwanted change. “So you had to think a little bit about what was going on, but not very deeply,” he says.
This is great in our little demo, but it says a lot about the limitations of such a program. It’s not a magic gene that can read your brain and turn every command into error-free code – and OpenAI doesn’t claim that. Instead, using it requires thinking and a little trial and error. Codex doesn’t turn non-coders into expert programmers overnight, but it’s certainly much more readily available than any other programming language.
OpenAI is on the rise to Codex’s ability to change programming and computing more generally. Brockman says it could help address the shortage of programmers in the United States, while Zaremba sees it as the next step in the historical development of coding.
“What happens to Codex has happened a few times before,” he says. In the early days of computing, programming was done by creating physical punching cards that had to be fed into machines, then people invented the first programming languages and started fine-tuning them. “These programming languages began to resemble English using vocabulary like ‘print’ or ‘exit’, and so more and more people were able to program.” The next part of this path completely removes the specialized coding languages and replaces it with English commands.
Codex also has the ability to manage other programs. In one demo, Brockman shows how the software can be used to create a Microsoft Word interface. Because Word has its own API, Codex can enter instructions for it in code generated from user voice commands. Brockman copies the poem into a Word document and then tells Word (via Codex) to first remove all indents, number the lines, then calculate the frequency of certain words, and so on. It’s very smooth, though hard to say how well it would work outside the bounds of a pre-agreed presentation.
If it succeeds, Codex will not only help programmers but will become a new interface between users and computers. OpenAI says it has tested Codex’s ability to manage not only Word but also other programs like Spotify and Google Calendar. And while the Word demo is just proof of the concept, Brockman says, Microsoft is apparently already interested in exploring the possibility of the software. “They’re very excited about the model in general, and you should expect to see a lot of Codex apps,” he says.