Microsoft Corp. GitHub A surprise emerged today with the emergence of the popular productive AI coding assistant. GitHub CopilotIt will no longer be supported solely by OpenAI’s GPT models.

Instead, the company said at its annual GitHub Universe event that it is taking a multi-model approach, giving developers the opportunity to leverage models from rival companies such as Anthropic PBC and Google LLC.

This announcement came alongside a number of other announcements via GitHub initialization A new AI tool called Spark that can automate much of the work of building web applications and more general updates to GitHub Copilot in VS Code and Xcode.

But the biggest announcement to date is the decision to make GitHub Copilot a multimodal tool. One blog postGitHub Chief Executive Officer Thomas Dohmke said users can now use Anthropic’s Claude 3.5 Sonnet extended language model. Copilot Chat’s web and VS Code interfacesGoogle’s Gemini 1.5 Pro will also arrive as another option in the coming weeks.

Additionally, GitHub Copilot will support an expanded range of OpenAI models that are said to focus on more advanced reasoning than GPT-4o, including more specialized models such as GPT o1-preview and GPT o1-mini. currently just using it.

The idea is that developers can switch between various models depending on what they’re trying to achieve, recognizing that in the AI ​​world there are different horses for different tracks and some models are better suited than those. others depending on the task.

The choice won’t be up to developers alone. Organizations will also be able to choose which models they want to make available to various team members. GitHub Copilot Enterprise.

Dohmke said the approach makes sense because it has become clear that there is no single model that will dominate every scenario. “The next phase of AI code generation will be defined not only by multi-model functionality, but also by multi-model selection,” he wrote.

Dohmke said Anthropic and Gemini will be available first in the GitHub Copilot web and VS Code chat interfaces, but the company plans to eventually bring them to all surface areas and functions. Copilot Work AreaGitHub’s command line interface and various other tools.

A multi-model future?

Considering Microsoft’s influence on GitHub, this move has led to: speculation The Redmond-based company itself may decide to end its exclusivity agreement with OpenAI and allow users to access other models through its Copilot tools.

While GitHub kicks off the co-pilot craze by launching its generative AI coding assistant in 2021, Microsoft hosts its own Co-Pilots On platforms such as Windows and Office.

Aspect OpenAI’s biggest supporterMicrosoft has taken full advantage of this relationship, using GPT models as the engine behind various Copilot tools, and has given no indication that it is unhappy with their performance. However, with GitHub moving to a multi-model approach, it’s likely that Microsoft will at least consider doing the same.

Whether he will actually do this is far from certain. Microsoft acquired GitHub in 2018 but, unlike many other acquisitions, has allowed the company to operate more or less independently since then. So the same thinking that is prevalent in GitHub may not be present in Microsoft.

For a long time, Microsoft praised OpenAI’s GPT models and presented them as the best of the best, so customers would think that privileged access gave it an advantage over its competitors.

But it became clear to many that the OpenAI models were not actually that superior. Google’s Gemini And Anthropic Claude Both models consistently show off some impressive talents of their own. However, competition in the AI ​​industry has changed and Microsoft is giving businesses more tools and platforms they need Building productive AI applications. One of the things their customers want is flexibility in model selection, so Azure AI service now offers multiple different LLMs from various vendors.

Additionally, the choice of which model to use is less obvious in some applications than in others. The complexity of writing code means that GitHub Copilot would definitely benefit from having more options, as some models are more proficient at certain programming languages ​​than others. However, this may not be the case for Copilots who are tasked with writing newsletters or correcting the user’s grammar. So we’re not quite sure yet whether Microsoft will reduce OpenAI’s exclusivity advantage.

GitHub Spark: Building apps with words

While the tech world watches with bated breath to see if Microsoft follows GitHub’s lead, there were plenty of other announcements at the GitHub Universe event. Perhaps most importantly, Spark launchIt is an artificial intelligence tool for creating web applications using natural language.

In a demonstration, GitHub showed that an initial prompt (essentially just a description of the app) generates a series of live previews of what that app could potentially look like. Users can compare different previews, choose the one they like, and then enter more prompts to change the look and feel of the app.

Of course, experienced developers will be able to change the code if they wish, while inexperienced users will be able to continue directing the code with their natural language until they get the design they want.

Once the user is satisfied with the application, GitHub Spark can deploy it wherever the user wants, such as a desktop, tablet, or smartphone. They will also be able to share their app with others so their colleagues can make changes to the app if they wish.

According to GitHub, Spark will help the company realize its vision of creating 1 billion developers in the world.

“For too long, there has been an unscalable barrier to entry that has separated the vast majority of the world’s population from developing software,” Dohmke said. “With Spark, we will enable more than 1 billion PC and mobile phone users to create and share their own micro applications directly on GitHub.”

Featured image: GitHub