Stack Overflow is evolving, and the internet is evolving. The way developers and technologists find information and learn online is not the same as it was a few years ago. Their expectations and behaviours have evolved since AI tools and products were made available. While core Q&A will always be a critical resource, we're working on making Stack Overflow the fastest way for developers to find exactly what they need and cultivate community, power learning, and unlock growth for every developer in this new age of technology (source). One concept we’re exploring is AI-powered search and discovery outside of Stack Overflow’s traditional question-and-answer format.

Yes, we’ve tried something similar before, so what’s different this time? Unlike past concepts that were RAG-based and only surfaced answers from Stack Overflow; this is instead a model-agnostic LLM + RAG model, where answers from Stack Overflow and the Stack Exchange Network are provided first, then LLM answers fill in any knowledge gaps. This solves the previous gap of a limited knowledgebase to pull from, that often gave irrelevant information or not enough context. We took past learnings, feedback, and research to build a different path forward.

We’ve launched the second phase of the beta experiment on stackoverfow.ai, a new AI tool that brings the power of community-verified knowledge and AI together, built separate from the public platform. In this phase of testing we're understanding functionality of the newly added features, like direct sources and attribution, and collecting feedback to iterate on the interface and user experience. If you’re a developer visiting Stack Overflow or the Stack Exchange Network, you will see a link on the left navigation of the site to try it out.

Why build it?

• A new way to get started on Stack Overflow - Stack Overflow is the largest community of technology enthusiasts, and now it’s easier than ever to get started. stackoverflow.ai offers a frictionless experience for finding answers and learning along the way, with trusted sources.
• An AI tool that prioritizes trusted, community-verified knowledge - While the adoption of AI tools continues to increase, so does developers’ lack of trust in the output. On stackoverflow.ai, human answers are prioritized and sourced, then the LLM fills in any knowledge gaps.
• Community is at the center of stackoverflow.ai - In the era of AI, the need for real human connection has never been more important. Developers can easily access the public platform to connect with the community, especially when you have complex questions, want to verify answers, or dive deeper into technical topics. 
• Get unstuck and learn along the way - Learn new coding techniques or programming languages with stackoverflow.ai. The tool’s responses don’t just provide an answer, the tool is designed to educate through its responses.
• An AI tool that stays relevant - Technology is constantly evolving. stackoverflow.ai has access to the latest questions and answers from the public platform, so you don’t have to worry about getting outdated answers. StackEval and StackUnseen leaderboards on ProLLM.ai show how LLMs can fall behind if they aren’t continuously trained on Stack Overflow’s ever growing database of human knowledge.

Feature set

• Attribution and sources - Answers from Stack Overflow and the Stack Exchange Network are provided first, then LLM answers fill in any knowledge gaps. Sources are presented at the top and expanded by default, with in-line citations and direct quotes from community contributions for additional clarity and trust.
• Post a question to the community: If developers are still not getting the solutions they need, the interface has a convenient path to bring their question to the community.
• Import chat history - Developers often visit Stack Overflow as a result of AI-related issues, now they can import their chat history from those other AI tools to get unstuck on stackoverflow.ai.
• ProLLM benchmarks - ProLLM Benchmarks evaluate LLMs on real-world interactions from Prosus Group companies, including Stack Overflow. StackEval and StackUnseen leaderboards on ProLLM.ai show how LLMs can fall behind if they aren’t continuously trained on human knowledge.