GitHub and OpenAI have launched a new AI tool that generates their own code

[ad_1]

GitHub and OpenAI have released a technical preview of a new AI tool called Copilot, which lives inside the Visual Studio Code editor and automatically completes code snippets.

According to GitHub, Copilot does more than just bring back the parrot’s code that’s been seen before. Instead, it analyzes the code you’ve already written and generates new matching code, including specific functions that were previously called. Examples on the project’s website include writing code automatically to import tweets, drawing a scatter plot, or getting a Goodreads rating.

It works best with Python, JavaScript, TypeScript, Ruby, and Go, according to a blog post from GitHub CEO Nat Friedman.

GitHub sees this as an evolution in pair programming, where two programmers will be working on the same project to catch each other’s errors and speed up development. With Copilot, one of those programmers is by default.

This project is the first major result of Microsoft’s $1 billion investment in OpenAI, the research firm now led by Y Combinator President Sam Altman. Since Altman took over, OpenAI has switched from a non-profit status to a “maximum profit” model, taken over Microsoft’s investment, and began licensing the GPT-3 script generation algorithm.

Copilot is built on a new algorithm called OpenAI Codex, which CTO Greg Brockman at OpenAI describes as a descendant of GPT-3.

GPT-3 is a key language generation algorithm from OpenAI, which can generate text that is sometimes indistinguishable from human writing. It is capable of writing very convincingly due to its massive size of 175 billion parameters, or adjustable knobs that allow the algorithm to correlate relationships between letters, words, phrases, and sentences.

While GPT-3 generates the English language, OpenAI Codex generates code. OpenAI plans to release a version of Codex through its API later this summer so developers can build their own apps using the technology, an OpenAI representative told The Verge in an email.

Codex was trained on terabytes of publicly available code taken from GitHub, plus English language examples.

While testimonials on the site rave about the productivity gains that Kopilot provides, GitHub notes that not all of the code used has been scanned for errors, unsafe practices, or personal data. The company wrote that it has put in some filters to prevent the co-pilot from generating offensive language, but it may not be ideal.

“Due to the pre-release nature of the underlying technology, GitHub Copilot may occasionally produce undesirable output, including biased, discriminatory, abusive or offensive output,” the Copilot website says.

Given criticism of GPT-3 bias and offensive language patterns, it appears that OpenAI has not found a way to prevent algorithms from inheriting the worst elements of their training data.

The company also warns that the form can suggest email addresses, API keys, or phone numbers, but this is rare, and the data has been found to be artificial or randomly generated by the algorithm. However, the code generated by Copilot is largely original. A test conducted by GitHub found that only 0.1 percent of the generated code could be found literally in the training set.

This is not the first project that tries to generate code automatically to help hard-working programmers. The startup Kite offers very similar functionality, with over 16 code editors available.

Now, Copilot is in restricted tech preview, but you can register on the project’s website for a chance to get access to it.

[ad_2]