[ad_1]
OpenAI has already made big changes to Codex, the AI-based coding assistant announced by the company last month. The system now accepts plain English commands and generates live working code, allowing someone to create a game or web application without even naming a variable. A few lucky coders (and, presumably, non-coders) will be able to run the tires on this new Codex API in a free private beta.
Codex is best viewed as OpenAI’s general-purpose language engine, GPT-3, but trained only on code instead of ordinary written material. This allows it to do things like full lines of code or entire sections, but when it was announced it wasn’t really something a non-coder would be able to interact with easily.
That has changed with this new API, which interprets ordinary, everyday requests such as “bounce the ball off the sides of the screen” or “download this data using the public API and sort it by date” , and publishes the functional code in one of a dozen languages.
I was treated to a live demo in which OpenAI co-founders Greg Brockman (CTO) and Wojciech Zaremba (Codex manager) built a simple game from scratch and explained what was going on behind the curtain.
“Programming is about having a vision and breaking it down into pieces, and then creating code for those pieces,” Brockman explained. The intention with Codex is to let coders spend more time on the first part than on the second. After all, a huge amount of code is duplicating or outright copying what others have done before – it can be creative, of course, but no one is going to exercise their imagination to do basic things like deploy a web server to test a web server. little code. That’s exactly what Brockman did with a single line – “create a web page that says that” or something like that.
Image credits: OpenAI
A second later, there were a dozen lines of JavaScript doing just that in a completely standard way.
“It’s the worst part of the programming,” Brockman said. “I’ve written this kind of code probably a dozen times, and I always forget exactly how it works. I don’t know about these APIs, and I don’t have to. You can just do the same things more easily, with fewer keystrokes or interactions.
Because Codex is trained on virtually all public code on GitHub, among other repositories, it is aware of all standard practices, the 50 or 100 times that a person includes a web server, keyboard commands, or manipulations. ‘objects and animations in their code. And because the natural language side has all the usual understanding of GPT-3, it gets when you say “shrink it and crop it” and then “have its horizontal position controlled by the left and right arrow keys,” you refer to the same “that”.
He also keeps in mind his own work, several kilobytes of coding context for himself.
He is also aware of the generalities built into the code corpus. For example, when Brockman told him to “drop the rock from the sky”, the system did not ask what “sky” is, even though it had not been defined on the canvas for the most part. Virgin. Not only did the boulder fall from the top of the screen, but the falling speed accelerated like an object normally would – because that’s her best guess of what “fall” and “sky” mean. from other uses and contexts.
Image credits: OpenAI
“We think this offers a new way to interact with existing software,” said Zaremba, who built a limited version of it for his doctoral thesis years ago, as they demonstrated a Codex plug-in for Microsoft Word. Automation exists for many tasks in word processors, sure, but what if you have a weird formatting issue and want to fix 100 different instances? Type “make all text the same size and font, and double spaces to be unique” and it will do so, removing spurious styles and choosing the most likely size and font considered ” normal ”. Then type “make all titles 24 point and bold” and it zooms in doing that, and so on.
It should be noted here that this sort of thing is convenient for many, but crucial for those who do not have the ability to do these things due to disabilities. If you are using your word processor using voice commands or a joystick, being able to perform complex tasks like the ones above is extremely useful. A blind coder, like anyone else, can patch a standard public test server, but the process of skimming Stack Overflow, getting the best snippet, checking the syntax, changing the relevant variables, etc. will almost certainly take longer.
And for those who work within the limits of the syntax and conventions passed down from above, Codex can easily be made to reflect these by exposing the model to the documentation. Codex can also convert and transfer code from one language to another, much like a translation engine turns Spanish into French.
Brockman said that, like with GPT-3, they’re only scratching the surface of what’s possible and hope to be surprised by what the developers come up with (after all, OpenAI didn’t predict AI Dungeon). The beta will be private, like that of GPT-3, but developers can apply and describe their project and the Codex team will review them for inclusion. Eventually the API will be a paid public API, but the timing and price is yet to be determined on this.
[ad_2]
Source link