At its annual Universe conference, GitHub last week announced a series of innovations for its AI-driven development tool: GitHub Copilot, which now integrates cutting-edge models from Anthropic, Google and OpenAI. GitHub also unveiled its latest advance to democratize software creation, allowing developers, whether new or experienced, to build micro-applications using natural language instructions: GitHub Spark.
GitHub Multi-model Copilot: towards more flexibility
Created by GitHub in collaboration with OpenAI, the AI-based programming assistance tool Copilot was initially powered by Codex, a version of OpenAI GPT-3 specifically designed for coding tasks. Copilot has evolved through updates to include GPT 3.5-turbo and, recently, GPT 4o and 4o-mini models, to meet different latency and quality requirements. GitHub has decided to integrate new cutting-edge AI models to allow developers and businesses to use it based on their internal policies, technology preferences, and security needs.
Claude 3.5 Sonnet from Anthropic, o1-preview and o1-mini from OpenAi have been deployed in Copilot Chat, Gemini 1.5 Pro from Google should join them in the coming weeks. Developers can switch between templates during a conversation.
Thomas Dohmke, CEO of GitHub, comments:
“In 2024, we have seen a boom in high-quality language models, large and small, that each individually excel at different programming tasks. There is no single model to govern all scenarios, and developers expect the agency to build with the models that work best for them. It is clear that the next phase of AI code generation will not only be defined by multi-model functionality, but also by multi-model choice. Today, this is exactly what we offer”.
Create natural language applications
GitHub has the ambition to reach 1 billion developers: Spark, developed in the GitHub Next laboratory, is its latest advance to democratize software creation, allowing users of all levels to design micro-applications or “Sparks” via instructions in natural language. With an intuitive interface and a real-time feedback loop, GitHub Spark allows you to iterate, test and share applications, providing maximum flexibility for both novices and experienced developers.
Spark is based on models from OpenAI and Anthropic. Experienced developers can directly make changes to the underlying code, while users or novice developers can iterate entirely in natural language. Once they’re happy with their spark, they can run it automatically on their desktop, tablet, or mobile device. They can also choose to share their micro-app with custom access control, as well as allow others to remix it and build on their creation.
Native AI experience extends across the entire GitHub platform
In addition to multi-modeling, GitHub introduces new features, such as:
- Copilot in VS Code : developers can now apply changes to multiple files simultaneously using natural language instructions;
- Copilot Extensions : they allow the integration of external tools and personalized functionalities into the development environment;
- Copilot Autofix and safety campaigns : Developers benefit from automated vulnerability remediation and new tools for managing security campaigns.
GitHub Octoverse 2024 : trends in AI and Python adoption
The Octoverse 2024 report shows that Python has overtaken JavaScript in usage, due to its versatility in AI projects. Additionally, GitHub is seeing a 98% increase in generative AI projects, a trend propelled by global AI adoption, including in India, Germany, and Japan.
The report also highlights significant growth in the developer community, particularly in Africa, Latin America and Asia.
