• woelkchen@lemmy.world
    link
    fedilink
    English
    arrow-up
    4
    arrow-down
    2
    ·
    4 months ago

    As I understand it, this is only about using search results for summaries. If it’s just that and links to the source, I think it’s OK. What would be absolutely unacceptable is to use the web in general as training data for text and image generation (=write me a story about topic XY).

    • elrik@lemmy.world
      link
      fedilink
      English
      arrow-up
      10
      ·
      4 months ago

      If it’s just that and links to the source, I think it’s OK.

      No one will click on the source, which means the only visitor to your site is Googlebot.

      What would be absolutely unacceptable is to use the web in general as training data for text and image generation.

      This has already happened and continues to happen.

      • woelkchen@lemmy.world
        link
        fedilink
        English
        arrow-up
        5
        arrow-down
        1
        ·
        4 months ago

        No one will click on the source, which means the only visitor to your site is Googlebot.

        That was the argument with the text snippets from news sources. Publishers successfully lobbied for laws to be passed in many countries that required search engine operators to pay fees. It backfired when Google removed the snippets from news sources that demanded fees from Google. Their visitors dropped by a massive amount, 90% or so, because those results were less attractive to Google users to click on than the nicer results with a snippet and a thumbnail. So “No one will click on the source” has already been disproven 10 or so years ago when the snippet issue was current. All those publishers have entered a free of charge licensing agreement with Google and the laws are still in place. So Google is fine, upstart search engines are not because those cannot pressure the publishers into free deals.

        This has already happened and continues to happen.

        With Gemini?

        • elrik@lemmy.world
          link
          fedilink
          English
          arrow-up
          2
          ·
          4 months ago

          The context is not the same. A snippet is incomplete and often lacking important details. It’s minimally tailored to your query unlike a response generated by an LLM. The obvious extension to this is conversational search, where clarification and additional detail still doesn’t require you to click on any sources; you simply ask follow up questions.

          With Gemini?

          Yes. How do you think the Gemini model understands language in the first place?

          • woelkchen@lemmy.world
            link
            fedilink
            English
            arrow-up
            1
            arrow-down
            1
            ·
            4 months ago

            The context is not the same.

            It’s not the same but it’s similar enough when, as the article states, it is solely about short summaries. The article may be wrong, Google may be outright lying, maybe, maybe, maybe.

            Google, as by far the web’s largest ad provider, has a business incentive to direct users towards the web sites, so the website operators have to pay Google money. Maybe I’m missing something but I just don’t see the business sense in Google not doing that and so far I don’t see anything approximating convincing arguments.

            Yes. How do you think the Gemini model understands language in the first place?

            Licensed and public domain content, of which there is plenty, maybe even content specifically created by Google to train the data. “the Gemini model understands language” in itself hardly is proof of any wrongdoing. I don’t claim to have perfect knowledge or memory, so it’s certainly possible that I missed more specific evidence but “the Gemini model understands language” by itself definitively is not.