Hacker News Reader

Comment by josefx | Hacker News Reader

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.

View on HN

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.

Considering the country in question, is there any evidence that Mistral is not using the same data?

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

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.

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.

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.

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/...

I have no legal training, so my guess is as good as coin flip for how this is going to turn out.

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]).

Reply on news.ycombinator.com