Dear Stack Overflow denizens, thanks for helping train OpenAI's billion-dollar LLMs

Stack Overflow, a community-driven Q&A site, and OpenAI, maker of AI models, have agreed to work to improve each other's products, the latest deal in a series of tie-ups to feed machine learning models' thirst for data.

The two organizations characterized the partnership on Monday as a way to "strengthen the world's most popular large language models," by which they mean OpenAI's.

The pact, which follows a similar Stack Overflow snuggle with Google Cloud, may also strengthen the developer site's business prospects by ensuring it receives some consideration from OpenAI for the super lab's scraping of advice of the Stack Overflow community.

We also presume these agreements are in place to head off any copyright or other legal brouhaha with the forum super-site. Stack Overflow content tends to show up a lot in LLM training datasets, including those used by OpenAI. Rather than scrape the Q&A website for content, OpenAI (like Google) can now directly extract that info from a dedicated API, it seems.

As to how generative AI will make things better for Stack Overflow visitors seeking help, that's not entirely clear. As Pulumi's recent experience with AI Answers has shown, not everyone welcomes machine-generated content.

Stack Overflow frowns upon generative AI, the output of which is banned from user-submitted posts. And the ineffectiveness of that ban can be inferred from the apology offered four days ago by "VonC," a Stack Overflow contributor, who admitted to making about 1,850 posts over the past year or so, two thirds of which were based on AI-generated text.

"I would like to sincerely apologize for the series of answers I posted on Stack Overflow over the past 13 months, which were based on outputs generated from an AI tool (before being reworked, researched, and sourced)," wrote VonC. "I understand that this action was against Stack Overflow's community guidelines, which explicitly ban the use of generative AI for creating posts."

Last June, a Stack Overflow moderator explained to The Register that the site was in crisis because people with programming questions were increasingly asking ChatGPT for answers, to the detriment of visitor traffic, participation, and revenue.

"ChatGPT and GenAI in general have become a serious existential threat to Stack Overflow's revenue model, and therefore the company has become desperate to try to find some way to get back the lost users (and lost revenue) they have lost to ChatGPT," our source said. "And the way they seem to have chosen to try to lure them back is by 'pivoting' into turning the site into a kind of elaborate ChatGPT clone of their own."

• Stack Overflow to charge LLM developers for access to its coding content
• ChatGPT's odds of getting code questions correct are worse than a coin flip
• Google says public data is fair game for training its AIs
• Stack Overflow bans ChatGPT as 'substantially harmful' for coding issues

That gambit became apparent last July when Stack Overflow introduced OverflowAI, an initiative that aims to help software developers by enlisting generative AI in the question-and-answer process. It encompasses efforts to make search more conversational, to ingest enterprise knowledge and integrate with Slack, among other things.

For example, OverflowAI might offer a summary answer to a question – remixed from human-authored replies with cited sources – instead of a community-ranked list of different responses that vary in quality. "We want to make it possible for public platform users to receive instant, trustworthy, and accurate solutions to problems using conversational search powered by GenAI," said CEO Prashanth Chandrasekar at the time.

Chandrasekar offered similar sentiment in a statement accompanying the OpenAI deal: OverflowAPI, which but for one letter might be confused with last year's more general OverflowAI project.

"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 users," he opined.

But specifics about the partnership are scarce. According to the announcement, "OpenAI will utilize Stack Overflow's OverflowAPI product and collaborate with Stack Overflow to improve model performance for developers who use their products." Also, "Stack Overflow will utilize OpenAI models as part of their development of OverflowAI and work with OpenAI to leverage insights from internal testing to maximize the performance of OpenAI models."

Members of the Stack Overflow community have asked for further details in various responses to the partnership announcement, with no success.

"I understand there is a desire for more specifics, but we are just announcing the partnership today," said staff moderator Rosie in response to a request for more details. "As work begins in the future, we will have more to share about how integrations with our partners will work."

OpenAI and Stack Overflow did not immediately respond to requests to explain the arrangement further. ®