MICROSOFT’s GitHub has agreed to bake artificial intelligence (AI) models from Anthropic and Alphabet’s Google into a coding assistant used by millions of software developers.
At first, customers will be able to use Google’s Gemini and Anthropic’s Claude 3.5 Sonnet to chat and ask questions, GitHub said on Tuesday (Oct 29) at its Universe conference in San Francisco. Eventually, the models will be incorporated into the main part of the GitHub Copilot assistant, which can spit out code with a few simple prompts.
After teaming up with OpenAI a few years ago, GitHub pioneered the use of generative AI to automate tedious parts of the coding process. The company will continue to make the OpenAI models the default setting, but developers will be able to switch to the Google and Anthropic models if they wish, GitHub chief executive officer Thomas Dohmke said.
Since acquiring GitHub in 2018, Microsoft has let the coding platform maintain a wide array of partnerships – even if they are technically Microsoft rivals. GitHub is keen to offer customers choice and to provide options to those who use Google’s and Amazon.com’s cloud services. The Anthropic model will run on the network operated by Amazon Web Services.
Microsoft also lets its Azure cloud customers use models other than OpenAI’s, including ones from Meta and Mistral, but does not offer models from Google or Anthropic.
GitHub also said it’s previewing an AI project called Spark that lets users create mini applications, say building a travel log with maps or an event RSVP tracker, using plain-English prompts rather than programming code.
Users who know how to program, can also open up the code generated by Spark and tweak it, toggling between coding and prompts. The idea is to give novices a way to create something and more experienced developers an easier way to test out ideas, Dohmke said. Developers who want to take their Spark projects further will have to move them to a fuller software development program, such as GitHub Copilot or Microsoft’s Power Platform.
Still, software development with AI is becoming more capable, he said.
“The size of the Lego blocks that Copilot on AI can generate has grown and it can assemble multiple Lego blocks together,” he said. “It certainly cannot write a whole GitHub or a whole Facebook, but the size of the building blocks will increase with AI getting better.” BLOOMBERG
