Stack Overflow to provide coding knowledge for OpenAI models – Business

Stack Exchange Inc., the operator of the popular Stack Overflow coding forum, today announced a content licensing partnership with OpenAI.

Under the deal, the large language model developer will gain access to posts from Stack Overflow. When a ChatGPT user asks a coding question that has been answered on Stack Exchange’s forum, the chatbot will be capable of fetching information from relevant posts. OpenAI says it plans to “provide attribution to the Stack Overflow community” within ChatGPT responses.

The LLM developer will collaborate with Stack Overflow to improve its models. According to the companies, the effort will place an emphasis on making OpenAI’s LLMs more useful for developers. This hints that the goal is to enhance ChatGPT’s code explanation and generation capabilities.

“Stack Overflow is the world’s largest developer community, with more than 59 million questions and answers,” said Stack Overflow Chief Executive Officer Prashanth Chandrasekar. “Our goal with OverflowAPI, and our work to advance the era of socially responsible AI, is to set new standards with vetted, trusted, and accurate data that will be the foundation on which technology solutions are built and delivered to our user.”

OpenAI will access Stack Overflow data through OverflowAPI, an application programming interface that Stack Exchange debuted last June. It enables companies to license the more than 59 million coding questions and answers on the forum by buying a subscription. In addition to existing content, the subscription offers access to the new posts that users upload to Stack Overflow over time.

One major selling point of OverflowAPI is that it provides “feedback signals” an AI model can use to evaluate the accuracy and usefulness of a given post. Those feedback signals take the form of the upvote and downvote counters next to users’ content submissions. By evaluating the number of upvotes a given post has received, neural networks can determine whether it should be incorporated into prompt responses.

Stack Exchange says that OverflowAPI can noticeably improve foundation models’ coding capabilities. In an internal evaluation, the company used the API to train a version of Llama 2 with 34 billion parameters. The data from OverflowAPI increased the percentage of coding tasks the model performed successfully during the test by nearly 20%.

OpenAI is not the only customer of OverflowAPI. In late February, Stack Exchange announced that Google LLC will make content from its forum available through the Gemini for Google Cloud chatbot. Stack Exchange, in turn, will use the search giant’s public cloud as its “platform of choice” for hosting its forum.

The deal with OpenAI likewise extends beyond content licensing. Stack Exchange will use the LLM developer’s models to support the development of OverflowAPI. Furthermore, it’s working on integrations between Stack Overflow and OpenAI models that will begin rolling out by the end of June.

It’s possible Stack Exchange plans to integrate OpenAI models into one or more of the AI features it’s currently developing for its platform. Last year, the company previewed an enhanced version of the Stack Overflow search bar that can generate answers to coding questions based on information in forum posts. The technical level of the automatically-generated answers is customizable.

Stack Exchange also intends to add AI tools to Stack Overflow for Teams, a commercial product that software teams can use to share technical knowledge internally. One of the planned features is capable of automatically asking questions about code and other technical data stored in third-party platforms such as GitHub. According to Stack Exchange, those questions can help developers more quickly familiarize themselves with an application’s code base.

Image: Stack Overflow