8

Note: This post is primarily addressed to the company. When I say "you" in this post, I am referring to the company.

A couple weeks ago, Microsoft dropped "Copilot" / "New Bing" (which is confusingly not the same as GitHub Copilot (but sadly just typical)), and it seems to do very much what Overflow Labs and Prashanth's blog posts have been touting as the shining feature of OverflowAI's coming search feature. I gave it a try and played around with it. It performs web searches, and uses search results to generate chat responses, including its list of references, and sometimes pointing inline where those references were used. This is probably not the only such thing out there already, and there will be more to come.

How does the company think OverflowAI's search is going to compete with this?

You're not a big company with tons of money to get a bunch of top-tier researchers- at least- not at the scale of Microsoft. All I can think of is the particular source of the data.

  • For a SO for Teams, if the data is proprietary info locked in the SOfT and not accessible to another search tool, then sure, the SOfT will still hold some unique share over valuable data to its users... until someone makes it so their competing tool can scrape the SOfT or have some other integration.

  • For the public network sites, (if you actually care at all about them), if the value is our content (which we are the source of, and which you seem to think so), I feel stupid saying this over and over, but you're doing a poor job of treating us like we matter to you. Even still, our content is CC-BY-SA and on the public web. Any other tool can easily use it if they follow the license.

And why does the company continue to tout trust and attribution as if it's something super special and amazing when it's already being done by others?

If anything (and I'm not saying this from a genuine confidence in its value, or its value to the Stack Exchange network sites, but more so about its value to the company's product offerings like SO for Teams), this now seems like something you'll have to do just to be on par with your competition.

1

1 Answer 1

-1

How does the company think OverflowAI's search is going to compete with this?

With data. This is one reason why Stack Exchange Inc. is considering preventing companies from using SE data free of charge.

6
  • 5
    I've already stated that undeleted SE public site data is CC-BY-SA, which allows sharing and adaptation. I don't see how they're going to charge for the type of web search usage I described.
    – starball
    Commented Nov 15, 2023 at 19:34
  • 2
    even for non-web search and ML training, creative commons is saying that their licenses can allow for training under fair use, or have at least stated that they believe that it should be allowed: creativecommons.org/2023/02/17/fair-use-training-generative-ai, creativecommons.org/2023/08/18/…, creativecommons.org/2021/03/04/…
    – starball
    Commented Nov 15, 2023 at 19:34
  • 4
    @starball I agree with you, but you asked what the company thinks, and given their previous statements (e.g., "We are looking for ways to gate access to the Dump, APIs, and SEDE, that will allow individuals access to the data while preventing misuse by organizations looking to profit from the work of our community."), that's how I think they think. FYI Is it illegal for a firm to train an AI model on a CC BY-SA 4.0 corpus and make a commercial use of it without distributing the model under CC BY-SA? Commented Nov 15, 2023 at 19:38
  • so you've addressed the point about potentially competing in the space of data in LLM training, but not about data from web search
    – starball
    Commented Nov 15, 2023 at 19:59
  • 1
    @starball I don't know how SE thinks about that. IMHO, I don't see SE having any edge, aside from their user base. Does OverflowAI's coming search feature aim to do web searches beyond SE content? Commented Nov 15, 2023 at 20:12
  • @starball: It's not that their licenses allow or disallow training of generative AI under fair use; but that if training of generative AI constitutes fair use (which they think it does/ought to) you don't need a licence anyway. If it doesn't constitute fair use:-"To the extent you are relying on CC licenses to train AI, you will need to follow the relevant requirements under the licenses" Commented Nov 16, 2023 at 10:37

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .