This is on top of an earlier €500m fine in 2021. And I'm guessing the total bill of €750m does at least far exceed what it would have cost to comply in the first place. Regardless of percentages, I'm guessing this does change Google's approach, at least in France.
If they want to access that market they have to play by their rules. Along the same tangent, the EU appears to be concerned with the intent/spirit of their laws and will amend them when anyone tries to side step them. It seems Apple and Google are not used to that.
As inconsistently interpreted by the member states, each with their own political biases and concerns. It’s a complicated, increasingly risky, and unpredictable market for tech firms.
Ultimately that’s fine, it’s not a necessity the EU be a leader in AI (except perhaps in defense).
Because most of those comments are unoriginal, one lined, and low on substance. Realize that “boooooo! mega corp” is not a valuable opinion in any discussion.
> 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.
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.
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.
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]).
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.
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.
I see more and more AI solutions to be simply not available in European Union. EU is a big market, true, but apparently not big enough to justify all the hassle, risks of handling random bureaucrats who want to become famous because they managed to fine some rich company. EU managed to enforce GDPR to some extent (still many parties are practically ignoring it by making refusal for tracking really annoying), question how it will play out with AI.
I tried to sign up for Claude 3 the other day but was met with a message that it’s not available in Sweden (which is a EU member state). It was available in Suriname though, so it’s not just a “US only for now” policy.
The saddest part is that many legislations in other parts of the world try to blindly copy the european legislation, without any effort in thinking if it makes sense or not.
that's not that sad, especially when talking abt gdpr. I mean gdpr can be better by enforcing respecting some request headers but other than that, gdpr is good, just like punishing unauthorized use of other's work
I'd say that's a gray area. Several of these schemes have been found in violation (Heise's implementation of it got them sued for example and they had to modify it). The bigger point here is that ONLY European websites have been going that route whereas American websites mostly just try to circumvent the GDPR by being clever (e.g. abusing the legal term "legitimate interest" to add a tedious layer of opt-outs), which hilariously backfires because it demonstrates that the violation is intentional.
Note that the point with the compliant "accept or pay" ones is that they still can't coerce you to "accept all" - they still have to only bundle the minimum necessary and justify it, and they then have to opt out paying users from those. Keep in mind that the ePrivacy Directive explicitly requires having an "refuse all" (or in this case "accept minimum necessary") button with equal weight and visibility as an "accept all" if the latter exists. Most consent dialogs are still in violation of this as they didn't update since the GDPR initially came into effect.
What is it with making this a bureaucracy problem before addressing the fact that AI models are often fed data that doesn't belong to the company training the model, which has usually been gathered with methods that would be considered outright theft or corporate espionage in other industries?
Of course there are bureaucracy issues, and there always will be, but I think it's absolutely not the main topic here.
If they harvest massive amounts of copyrighted and suspect data and send the worst parts of it to strangers over the Internet, everyone calls it "ML" and "training" and investors give them billions in funding but when I do it they call me a pirate and a pervert and throw me in jail.
> risks of handling random bureaucrats who want to become famous because they managed to fine some rich company
Do list the names of famous bureaucrats that got famous because they fined rich companies. There's Margrethe Vestager who is the commissioner for competition, and is neither famous nor famous for finning rich companies.
Also, do you have any actual criticism of specific legislation or is this a generic "laws bad because they make it harder for business which is bad" rant without any substance?
The guy who was a CEO of a top 10 digital services (?) company worldwide, Atos, and before that of the biggest French telecom (France Telecom/Orange). He was named 3 times in the top 100 CEOs by Harvard Business Review. He was a minister of economics, finances and industry for 2 years. Also has a few sci-fi books.
And for you, he became popular when he was named European Commissioner? Which company did he fine to become famous with his new position?
The only actual example I can think of is Schrems and he didn't fine anyone, he merely dismantled the shoddily constructed EU-US privacy law circumventions (Privacy Shield etc). And I had to look up his first name (Max) because I only knew his name from the court cases (notably Schrems II).
Heh, great illustration of how hollow the "politician acting for fame" accusation is: the only name you could think of is of an EFF-like activist, he's not a politician at all :D
Turns out I had him down as a "politician" because I conflated him with the Greens politician who was doing elaborate flow charts on Brexit on social media. Incidentally I also forgot his name too.
I saw he's only listed as a "lawyer" (not a political position or party membership) when looking up his full name but given that I know literally nothing else about him (the face doesn't even feel familiar) I left it at that.
It's not the EU's fault that the US government has been completely useless for so long that big tech thinks they have a right to trample all over society to make an extra buck.
Couldn’t one argue that the current system is expressly in the interest of the US government? We are entering a new technological age with the US at the forefront and the EU falling further behind while busily erecting more obstacles. Economically, at least.
If it’s not US companies, it will just be Chinese ones. Homegrown isn’t an option the EU places much value on.
The US was stricter on its homegrown giants of the past, so I don't really buy into the idea that the US has to let big tech impose all of its dystopian ideals upon people to retain a lead over China.
The current approach may be benefiting the US economically, but it's also ticking up a time bomb in the form of an increasingly pissed off public.
The EU is also going to miss out on new AI companies, so no only are they at a competitive disadvantage because they're using less AI, they also aren't getting the job and tax benefits from the new companies that will form.
I wonder through what mechanism they were able to fine them for bard/ Gemini, since it's not really clear that training an LLM is a copyright violation. Is it through those prior commitments made in 2019 or whatever?
Its because the EU has something called "neighbouring rights" which are an additional right to the usual form of copyrights.
> To tackle this, the EU created a form of copyright called “neighbouring rights” that allows print media to demand compensation for using their content.
In some EU countries, like Austria for example, there are royalties to pay on non-volatile storage memory[1], who's funds go to the local mafia...errr I mean music/movie rights holders association, because get this, storage can be used for storing reproductions of copyrighted content, so if you buy anything from external HDDs to phones with flash memory, then you're automatically a potential pirate and you owe them money upfront to compensate them for the piracy induced losses.
It's a tax that started from the 1980's first on audio and VHS cassettes, and kept getting updated over the years to include MP3 players, flash memory, smart phones and smart watches.
I expect the exact same thing will happen to LLM bots. The juggernauts like Ringier or Axel Springer will lobby in their respective countries or at EU level to get their pound of flesh whenever people use something like ChatGPT, because they'll claim it was trained on their copyrighted data and so they deserve to get their cut, otherwise the service will have to be banned in the EU.
> In some EU countries, like Austria for example, there are royalties to pay on non-volatile storage memory[1]
In France it's the same but with an even more insane version. The tax is so high that it's skewing the EU average and you have to take France out of the data.
As an example, taxes on mobile phones are at 14EUR for a phone with a drive greater than 64GB (so most of them nowadays)
I notice Austria an France seem to have a lot of such dumb taxes in common. I guess it must be to feed those generous welfare systems. But these taxes are basically robbing Peter to pay ... Peter.
That's the idea yeah, the economic strategy in France is very simple: if it works then tax it, it if doesn't work then give more subsidies.
That's a terrible long term strategy because the good companies are subsidizing failing ones, destroying the reward function of the market.
The second terrible side effect to this policy is that the subsidies system is too complex for small companies to handle so the small companies have an increased failure rate compared to a normal economy.
There are a lot of laws that are seemingly universal like patents, copyright, privacy (right to your own likeness etc) but the details can vary widely internationally and even between EU countries.
As an example: it's infamously prohibited to commercially publish photographs taken of the Eiffel tower at night because the light installation from 1990 onward is protected by copyright law as a work of art (as is the glass pyramid of the Louvre btw). HOWEVER it's legal to take and use pictures of permanetly publicly displayed works of art in some European countries like Austria, Germany and Switzerland ("Panoramafreiheit"). So you could sell post cards of unlicensed photographs of the Eiffel tower lights in those countries but not France. But if you commercially use them on a Germany-based website available in French you might still get sued from France.
Here's a German map of this specific law across Europe to illustrate the point. You don't have to speak German to understand that the laws vary wildly despite all being based in copyright:
https://www.parismalanders.com/wp-content/uploads/2022/06/Pa...
It's easy to see how one could argue that using copyrighted data as "training material" for a commercial LLM is commercial use. It's also easy to see how one could argue that this is not "reproduction" of significant portions of the copyrighted work and therefore copyright law does not apply. But it's also easy to see how one could argue that this still qualifies as "citation" the way Google regurgitates snippets from news sites and is thus covered by the "citation" provision for protecting newspapers. But it's also easy to see how one could argue that's not the case. Different jurisdictions will interpret these things differently.
This does nothing for real people, it's just big companies fighting to get money while doing as little as possible. It would be much better to see a crackdown on all the harmful stuff big tech does to individuals rather than use public resources to shift around who benefits from rent seeking.
Interesting, the title should probably be changed to something more related to them being fined for training on copyrighted data.
That's the more interesting part here, they've essentially banned developing ai in France with this (though maybe they just won't go after their own ai companies).
Developing AI with news articles. It doesn't stop you from e.g. licensing reddit data and training on that.
And it's unclear if other companies would actually run afoul of this at all, or whether this is just applicable because of the terms Google agreed to in the 2022 settlement about using news articles in their search engine. For a settlement agreement to make sense at all they obviously had to agree to stricter terms than what the regulation itself demands.
A few large media companies own the copyright on the majority of news articles. AI companies can just pay to license the content (or give an equity stake in exchange).
Anyone is welcome to pay a publication or an author (for books/magazines etc) in order to use their data for training. Companies are also welcome to request consent from individuals to use their online posts for this purpose.
Otherwise, what OpenAI/Gemini are doing is, simply put, "theft".
The public blog post is offered as exactly that, public. The copyrighted content that Google trained on was not public, not paid for, not permitted to be used in training, and not used to train people, only AI.
There's no relation between the post and the hypothetical you invented - it's misdirection at best and manipulation at worst to compare them.
> they've essentially banned developing ai in France
This is hyperbole. Intellectual property laws have existed longer than AI, they're being enforced on AI companies. These companies can pay to use the data, just like I would have to pay for server time with those same companies. They are richer than some countries, not students experimenting in a basement.
And don't give me the "nobody lost anything" bullshit argument. I pirate heavily myself but I'm not a hypocrite pretending that I'm not taking something I should've paid for.
“Which way the wind will blow” — as if it ever blew in the direction of ripping people off. Google still says you can’t use their ML stuff for anything related to ML, which smells like a noncompete foisted on customers.
I'm in two minds with these kinds of fines (maybe not super relevant for this specific fine).
1. Do these big megacorps just not care about rules, and they look at fines as a cost of doing business while the steam roll through industries.
2. Have lawmakers completely failed for the past decades to effectively referee tech, and so instead of having sane regulation and clear rules everything has to get decided in courts with fines.
Probably both are true, and the real losers are small businesses trying to get started.
The laws around training an LLM on publicly accessible IP have never been developed and haven't been developed, so the judge has to decide from common law. It's the same issue as crypto <> SEC. Companies ask lawmakers to set the rules, but they drag their feet, they release their products, and end up getting sued.
It has been decided already in more technically knowledgeable countries that training on data is not the same as copying the data. It's obvious if you understand the concept of data compression, but judges don't always see it that way.
Or that we already train humans on copyrighted data and then employ that human intelligence for commercial use.
You can argue that AI is a categorically different scale that requires special regulation to protect our societal values (and I think I do) but that is a different claim and implementation than copyright enforcement.
It should be viewed as derivarive work. It equally applies to humans. It shouldn't be classified as fair use because the works have been used in whole to train the model.
'A "derivative work" is a work based upon one or more preexisting works, such as a translation, musical arrangement, dramatization, fictionalization, motion picture version, sound recording, art reproduction, abridgment, condensation, or any other form in which a work may be recast, transformed, or adapted.'
I'm French. It's important to understand that France has been very bitter about missing the tech train with its failed "Minitel." And since, instead of making necessary changes to become more business-friendly and technologically relevant, they've just remained bitter and try to tax/punish US giants any way they can.
Thus, instead of becoming the type of places where the next Google or OpenAI could be incorporated, they're promoting the brand of a country that loves playing the Robinhood for its corporate friends in the newspaper business. I get the logic of the fine, and it's also sad to see them be proud of fining these foreign companies, instead of actually doing anything to change the fact that the country is becoming increasingly irrelevant in the modern technological age.
I doubt they'll be able to survive in the long term to be honest.
They'll either be forced to cooperate with the slow IT movers megacorps like Orange or Capgemini or buried in the legal framework by the copyright mafias. (Or a mix of both)
I'd love to be proven wrong of course but I know too much how it works there to be optimistic.
punishing tech giants for breaking the laws can be done regardless of creating a better less bureaucratic tech environment in france. Why you are trying to correlate both? Tbh I think the fine for google for this case is too small.
What does this really accomplish, will the newspapers get any of that money or will it just be thrown into country's budget pot, as a side effect alienating France further from possible tech investments? The French Google office is tiny for such a big country, at some point it's just not worth to navigate these political waters I assume.
it is interesting to see for which parties these actions are taken on. they appear more like the protectionist actions that governments take in other business domains.
102 comments
[ 3.3 ms ] story [ 179 ms ] threadAs inconsistently interpreted by the member states, each with their own political biases and concerns. It’s a complicated, increasingly risky, and unpredictable market for tech firms.
Ultimately that’s fine, it’s not a necessity the EU be a leader in AI (except perhaps in defense).
*this is a joke, just in case
The concept of having to pay somebody for linking out to their content is hostile to the open internet.
Imagine if you had to pay Pearson every time you cited ‘their’ research?
The official EU page says mere hyperlinks aren’t covered[1] but this legal source says headlines & images are [2] (making links de facto covered)
[1] https://digital-strategy.ec.europa.eu/en/faqs/copyright-refo...
[2] https://www.companydebt.com/articles/will-the-eus-link-tax-c...
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.
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.
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.
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
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.
Really irritating. But yes, we deserve it.
as always, it helps to maintain an identity in several geopolitical 'blocks', this time Montenegro worked :)
Note that the point with the compliant "accept or pay" ones is that they still can't coerce you to "accept all" - they still have to only bundle the minimum necessary and justify it, and they then have to opt out paying users from those. Keep in mind that the ePrivacy Directive explicitly requires having an "refuse all" (or in this case "accept minimum necessary") button with equal weight and visibility as an "accept all" if the latter exists. Most consent dialogs are still in violation of this as they didn't update since the GDPR initially came into effect.
Of course there are bureaucracy issues, and there always will be, but I think it's absolutely not the main topic here.
Do list the names of famous bureaucrats that got famous because they fined rich companies. There's Margrethe Vestager who is the commissioner for competition, and is neither famous nor famous for finning rich companies.
Also, do you have any actual criticism of specific legislation or is this a generic "laws bad because they make it harder for business which is bad" rant without any substance?
Thierry Breton is another name that comes to mind.
And for you, he became popular when he was named European Commissioner? Which company did he fine to become famous with his new position?
The only actual example I can think of is Schrems and he didn't fine anyone, he merely dismantled the shoddily constructed EU-US privacy law circumventions (Privacy Shield etc). And I had to look up his first name (Max) because I only knew his name from the court cases (notably Schrems II).
I saw he's only listed as a "lawyer" (not a political position or party membership) when looking up his full name but given that I know literally nothing else about him (the face doesn't even feel familiar) I left it at that.
If it’s not US companies, it will just be Chinese ones. Homegrown isn’t an option the EU places much value on.
The current approach may be benefiting the US economically, but it's also ticking up a time bomb in the form of an increasingly pissed off public.
If not training, then possession perhaps?
> To tackle this, the EU created a form of copyright called “neighbouring rights” that allows print media to demand compensation for using their content.
It's a tax that started from the 1980's first on audio and VHS cassettes, and kept getting updated over the years to include MP3 players, flash memory, smart phones and smart watches.
I expect the exact same thing will happen to LLM bots. The juggernauts like Ringier or Axel Springer will lobby in their respective countries or at EU level to get their pound of flesh whenever people use something like ChatGPT, because they'll claim it was trained on their copyrighted data and so they deserve to get their cut, otherwise the service will have to be banned in the EU.
[1] https://www-wko-at.translate.goog/oe/handel/maschinen-techno...
In France it's the same but with an even more insane version. The tax is so high that it's skewing the EU average and you have to take France out of the data.
As an example, taxes on mobile phones are at 14EUR for a phone with a drive greater than 64GB (so most of them nowadays)
That's a terrible long term strategy because the good companies are subsidizing failing ones, destroying the reward function of the market.
The second terrible side effect to this policy is that the subsidies system is too complex for small companies to handle so the small companies have an increased failure rate compared to a normal economy.
There are a lot of laws that are seemingly universal like patents, copyright, privacy (right to your own likeness etc) but the details can vary widely internationally and even between EU countries.
As an example: it's infamously prohibited to commercially publish photographs taken of the Eiffel tower at night because the light installation from 1990 onward is protected by copyright law as a work of art (as is the glass pyramid of the Louvre btw). HOWEVER it's legal to take and use pictures of permanetly publicly displayed works of art in some European countries like Austria, Germany and Switzerland ("Panoramafreiheit"). So you could sell post cards of unlicensed photographs of the Eiffel tower lights in those countries but not France. But if you commercially use them on a Germany-based website available in French you might still get sued from France.
Here's a German map of this specific law across Europe to illustrate the point. You don't have to speak German to understand that the laws vary wildly despite all being based in copyright: https://www.parismalanders.com/wp-content/uploads/2022/06/Pa...
It's easy to see how one could argue that using copyrighted data as "training material" for a commercial LLM is commercial use. It's also easy to see how one could argue that this is not "reproduction" of significant portions of the copyrighted work and therefore copyright law does not apply. But it's also easy to see how one could argue that this still qualifies as "citation" the way Google regurgitates snippets from news sites and is thus covered by the "citation" provision for protecting newspapers. But it's also easy to see how one could argue that's not the case. Different jurisdictions will interpret these things differently.
That's the more interesting part here, they've essentially banned developing ai in France with this (though maybe they just won't go after their own ai companies).
And it's unclear if other companies would actually run afoul of this at all, or whether this is just applicable because of the terms Google agreed to in the 2022 settlement about using news articles in their search engine. For a settlement agreement to make sense at all they obviously had to agree to stricter terms than what the regulation itself demands.
Good luck competing with AI that were able to train on all data instead of just a small subset.
Japan has already codified it into law, I expect if other countries pile in on calling it a violation of copyright, many will set up an entity there.
Otherwise, what OpenAI/Gemini are doing is, simply put, "theft".
There's no relation between the post and the hypothetical you invented - it's misdirection at best and manipulation at worst to compare them.
This is hyperbole. Intellectual property laws have existed longer than AI, they're being enforced on AI companies. These companies can pay to use the data, just like I would have to pay for server time with those same companies. They are richer than some countries, not students experimenting in a basement.
And don't give me the "nobody lost anything" bullshit argument. I pirate heavily myself but I'm not a hypocrite pretending that I'm not taking something I should've paid for.
1. Do these big megacorps just not care about rules, and they look at fines as a cost of doing business while the steam roll through industries.
2. Have lawmakers completely failed for the past decades to effectively referee tech, and so instead of having sane regulation and clear rules everything has to get decided in courts with fines.
Probably both are true, and the real losers are small businesses trying to get started.
It has been decided already in more technically knowledgeable countries that training on data is not the same as copying the data. It's obvious if you understand the concept of data compression, but judges don't always see it that way.
You can argue that AI is a categorically different scale that requires special regulation to protect our societal values (and I think I do) but that is a different claim and implementation than copyright enforcement.
'A "derivative work" is a work based upon one or more preexisting works, such as a translation, musical arrangement, dramatization, fictionalization, motion picture version, sound recording, art reproduction, abridgment, condensation, or any other form in which a work may be recast, transformed, or adapted.'
(US Copyright Act of 1976)
Thus, instead of becoming the type of places where the next Google or OpenAI could be incorporated, they're promoting the brand of a country that loves playing the Robinhood for its corporate friends in the newspaper business. I get the logic of the fine, and it's also sad to see them be proud of fining these foreign companies, instead of actually doing anything to change the fact that the country is becoming increasingly irrelevant in the modern technological age.
They'll either be forced to cooperate with the slow IT movers megacorps like Orange or Capgemini or buried in the legal framework by the copyright mafias. (Or a mix of both)
I'd love to be proven wrong of course but I know too much how it works there to be optimistic.
And that may be part of the issue. Milking billions and having a postal box as office doesn’t look good for the government.