I find it a bit odd that "senior leadership is working on a strategy to protect Stack Overflow data from being misused by companies building LLMs", but historically has only cared about sites that are proxies to content.

I agree that there are a lot of open questions and concerns - legally and ethically - with how LLMs (and other applications of generative AI) are trained and implications around their output. However, the company does not own the content that we provide. Anyone who posts content on the network retains ownership but grants the company an irrevocable license to take certain actions with that content. As owners, we are responsible for contacting people who have copied our content or, if we choose, seeking consultation with lawyers to pursue legal actions.

Cutting off the data dumps and any future steps - which I suspect will involve changes to the Data Explorer and/or the API - doesn't help us as content creators and owners. The vast majority of Stack Exchange content is already out there in the existing data dumps. The web pages can (and I presume are) being crawled - Google's Bard, for example, can summarize any Stack Exchange page and use posts that are minutes old as part of the summarization.

Restricting access to the open data does more harm to the communities than it does to the LLM developers. There are far better investments in time, money, and effort.

  • 2
    Possibly related? law.stackexchange.com/q/91785/323
    – Kevin
    Commented Jun 10, 2023 at 6:48
  • 1
    @Kevin Probably not. My understanding is that your answer there is still correct and it's an open legal question. This is more about the stance of SE, regardless of the legal question(s). Previously, it's up to me to take the necessary actions to protect my work. Now, SE is saying that they will take action to protect my work, but only in a specific situation that may not be applicable, and in a manner that is more harmful than good, without any input or feedback from the owners of the work. Commented Jun 10, 2023 at 9:53
  • They can't defend your copyright for you in the first place. Not unless you sign a much more restrictive contract than the current ToS, anyway.
    – Kevin
    Commented Jun 10, 2023 at 19:29
  • @Kevin That's exactly my point. I don't see what "Stack Overflow data" means, since it's owned by the contributors and I don't see what they can do to protect things that other people own. So something else is going on. Commented Jun 10, 2023 at 19:35
  • 1
    Related post by wizzwizz4: "That ship has sailed. OpenAI's models famously have a knowledge cut-off date of 2021, and few people care. Stack Exchange data is gathered, and fed into these models, via the Common Crawl project. Nobody is using Stack Exchange Data Dumps to train AI models."
    – starball
    Commented Jun 10, 2023 at 20:52


You must log in to answer this question.

Browse other questions tagged .