The latest from OpenAI comes to Copilot. Scheduling assistant evolves with a new AI model

In the last year, artificial intelligence has not only been behind image generators like DALL E and chatbots like ChatGPT, it has also played a leading role in software developmentAnd the progress continues. Copilot, which had surprised programmers with its capabilities, has just been updated.

The tool, which is developed by GitHub, a subsidiary of Microsoft, works with Codex, a language model originally powered by the prodigious GPT-3. Now, however, the company has announced which has just updated the underlying technology of the aforementioned Codex with a “more advanced OpenAI model”.

GPT-3.5 or GPT-4?

GitHub, taking the same stance as Microsoft with the new Bing, has preferred not to specify which AI model now powers Copilot, however there are some possibilities on the table. Officially, the latest version of GPT released is 3.5which became known late last year with the company’s viral bot run by Sam Altman.

So, the new heart of Copilot, and consequently of Codex, can be GPT-3.5 or, eventually, some other intermediate version or directly the future GPT-4. In any case, we see how those from Redmond do not sit idly by and continue to rely on OpenAI to continue feeding the applications that are under their orbit.

Now, this update is not a change of versions that simply stays on paper. The schedule assistant is now better. Developers, according to GitHub, will now receive more accurate code hints. And these will arrive much faster, which will help them improve their work dynamics.

With these improvements, the company hopes that a greater percentage of the code written by developers will be generated by this AI-powered tool. Currently, he notes, 46% of the code users upload to GitHub has been “chopped” by Copilot, and as we focus on Java that number rises to 61%.

In addition to the improved AI model, the tool now also has a new paradigm called Fill-In-the-Middle (FIM), which gives developers better pointers for code hints. This is complemented by an AI-based vulnerability filtering system that blocks insecure coding patterns.

Copilot for companies

GitHub continues to develop the Copilot business model. After some time in the testing phase, the paid version for individual users was launched in June of last year. This week the version for companies has just arrived, which basically allows employers to deploy and manage the tool in their work teams.

This possibility will allow new clients, in this case corporate, to become interested in Copilot. And to more customers, more billing for Microsoft subsidiary who, by the way, is an expert in managing business environments with their products. But there is more, they promise that the tool will continue to evolve.

“Over the next few years, we will embed AI into all aspects of the developer experience, from coding to pull requests to code deployments,” he says in a blog post. It remains, then, to continue observing closely the way in which AI also raises its own revolution in the world of programming.

Images: GitHub

In Xataka: I’ve been using GitHub’s Copilot for programming for a few days and this is my experience