5 ms·

> The competition authority said Google’s AI-powered chatbot Bard – since rebranded as Gemini – was trained on content from publishers and news agencies without

> The competition authority said Google’s AI-powered chatbot Bard – since rebranded as Gemini – was trained on content from publishers and news agencies without notifying them. An important reason—amongst others—why projects like ChatGPT cannot be conceived inside Google anymore. Yeah, it's sad how legal, privacy, and compliance teams have expanded to obstruct any significant engineering project, but there are genuine reasons why.

josefx 3y ago
So why does Google seem to have such significant issues trying to catch up despite violating these laws and agreements at every chance it can? Apparently using all this data does not help them all that much.
- numbchuckskills 3y ago
  I think the point is that others are using this data as well, but not a target given there's no gigantic piggy bank to extract $ from.
  - wongarsu 3y ago
    Or rather, the others don't have preexisting agreements due to previous grievances. For OpenAI this would be the first violation of regulations about news content reproduction. That would be lengthy court case with an uncertain outcome. But Google already went through such a court case (before Gemini was a thing), settled, and now with Gemini violated the settlement agreement. Showing that they violated the settlement agreement is much easier and quicker, so that's the news you get today.
    - ben_w 3y ago
      Indeed. I was about to link to the news from December that the New York Times is suing OpenAI for using their news articles. https://fingfx.thomsonreuters.com/gfx/legaldocs/byvrkxbmgpe/OPENAI%20MICROSOFT%20NEW%20YORK%20TIMES%20mtd.pdf https://fingfx.thomsonreuters.com/gfx/legaldocs/byvrkxbmgpe/... I have no legal training, so my guess is as good as coin flip for how this is going to turn out.
      - seanmcdirmid 3y ago
        That one seems to be NYTimes not understanding how LLM's work, and will hinge on technical clarifications (either OpenAI used NyTimes articles in training their model [fault], or the NyTimes seeded their prompt with enough article snippets to reproduce it [not at fault]).
- ajross 3y ago
  Is there any evidence that OpenAI is not using the same data? They're just not big enough for the regulators to aim at yet.
  - fakedang 3y ago
    Considering the country in question, is there any evidence that Mistral is not using the same data?
  - leononame 3y ago
    Or maybe there just isn't evidence they've used it? Whereas for Google, we do have evidence they used that data. I mean, I'm all for these companies disclosing their training data. But just assuming OpenAI isn't fined because of their site is (unless you know more) pure speculation. Same goes for the sibling comment suggesting Mistral should also be looked at
    - az09mugen 3y ago
      One difference between ChatGPT & Gemini on one hand and Mistral on the other hand, is that Mistral is an open source model. Can't say the same thing for OpenAI and Google LLMs.
nickpsecurity 3y ago
A big part of the problem might be that Google tries to do everything in house. Many companies use contractors or third-party software to reduce the risk in these areas. Most of the innovative work is also being done outside of Google. Cheaper, too. One of Google’s own AI researchers said as much. Imagine Google investing in a huge ecosystem of 3rd party tools that could solve their problems with the deal being OSS or shared-source deliverables they could use. That they’re independent puts most liability on them. For whatever is shared, Google could still support their managers, lawyers, etc in situations of mutual interest. If Google wasn’t so focused on control and ego, then they could take advantage of such strategies to receive all their benefits. Google is suffering strictly because of their operational choices that increase liability, not E.U. regulations. They could suffer for E.U. regulations in a future situation where they’re doing their best but E.U. bullied them anyway. We’re not there yet.

josefx 3y ago

So why does Google seem to have such significant issues trying to catch up despite violating these laws and agreements at every chance it can? Apparently using all this data does not help them all that much.

numbchuckskills 3y ago
I think the point is that others are using this data as well, but not a target given there's no gigantic piggy bank to extract $ from.
- wongarsu 3y ago
  Or rather, the others don't have preexisting agreements due to previous grievances. For OpenAI this would be the first violation of regulations about news content reproduction. That would be lengthy court case with an uncertain outcome. But Google already went through such a court case (before Gemini was a thing), settled, and now with Gemini violated the settlement agreement. Showing that they violated the settlement agreement is much easier and quicker, so that's the news you get today.
  - ben_w 3y ago
    Indeed. I was about to link to the news from December that the New York Times is suing OpenAI for using their news articles. https://fingfx.thomsonreuters.com/gfx/legaldocs/byvrkxbmgpe/OPENAI%20MICROSOFT%20NEW%20YORK%20TIMES%20mtd.pdf https://fingfx.thomsonreuters.com/gfx/legaldocs/byvrkxbmgpe/... I have no legal training, so my guess is as good as coin flip for how this is going to turn out.
    - seanmcdirmid 3y ago
      That one seems to be NYTimes not understanding how LLM's work, and will hinge on technical clarifications (either OpenAI used NyTimes articles in training their model [fault], or the NyTimes seeded their prompt with enough article snippets to reproduce it [not at fault]).
ajross 3y ago
Is there any evidence that OpenAI is not using the same data? They're just not big enough for the regulators to aim at yet.
- fakedang 3y ago
  Considering the country in question, is there any evidence that Mistral is not using the same data?
- leononame 3y ago
  Or maybe there just isn't evidence they've used it? Whereas for Google, we do have evidence they used that data. I mean, I'm all for these companies disclosing their training data. But just assuming OpenAI isn't fined because of their site is (unless you know more) pure speculation. Same goes for the sibling comment suggesting Mistral should also be looked at
  - az09mugen 3y ago
    One difference between ChatGPT & Gemini on one hand and Mistral on the other hand, is that Mistral is an open source model. Can't say the same thing for OpenAI and Google LLMs.

nickpsecurity 3y ago

A big part of the problem might be that Google tries to do everything in house. Many companies use contractors or third-party software to reduce the risk in these areas. Most of the innovative work is also being done outside of Google. Cheaper, too. One of Google’s own AI researchers said as much. Imagine Google investing in a huge ecosystem of 3rd party tools that could solve their problems with the deal being OSS or shared-source deliverables they could use. That they’re independent puts most liability on them. For whatever is shared, Google could still support their managers, lawyers, etc in situations of mutual interest. If Google wasn’t so focused on control and ego, then they could take advantage of such strategies to receive all their benefits. Google is suffering strictly because of their operational choices that increase liability, not E.U. regulations. They could suffer for E.U. regulations in a future situation where they’re doing their best but E.U. bullied them anyway. We’re not there yet.