This is the point. EU weenies can congratulate themselves on their regulations against big bad evil companies all they want but they're going to get no AI jobs.
It is deeper than that working pretty big EU tech-firm. Our product is basically bot that uses GPT-4o and RAG and we are having lots of those eu-regulation talks with customers. It probably would be nightmare if we fine tuned our models or some more complex stuff.
A simple approach to compliance:
[https://artificialintelligenceact.eu/assessment/eu-ai-act-compliance-checker/](https://artificialintelligenceact.eu/assessment/eu-ai-act-compliance-checker/)
What is your opinion on Mario Draghi’s report?
[Report link](https://commission.europa.eu/document/download/97e481fd-2dc3-412d-be4c-f152a8232961_en?filename=The%20future%20of%20European%20competitiveness%20_%20A%20competitiveness%20strategy%20for%20Europe.pdf)
“With the world on the cusp of an Al revolution, Europe cannot afford to remain stuck in the “middle technologies and industries” of the previous century. We must unlock our innovative potential. This will be key not only to lead in new technologies, but also to integrate Al into our existing industries so that they can stay at the front.”
Does this influence your thinking at all?
It's a mixed bag. Draghi does make some good points, but in my view, he doesn't focus on the biggest issue: Capital Markets and state funding.
The US Inflation Reduction act has had significant economic impact, but Europe is utterly incapable of matching it. Meanwhile private capital is very conservative and fractured. For me that is the key issue we face.
Nonetheless, I will say the following: Europe should focus on not *weakening*, but *simplifying* its regulations. Having worked on many, I can't think of many EU laws I'd like to see repealed, but I can think of many cases where they are convoluted and too complex.
We either need to draft simpler, better laws, or we need to create tools for businesses to feel confident they are compliant more easily.
The GDPR is a great example: many people still don't understand that you don't need to ask for cookies if the cookies you are using are *necessary* for the site to work (login cookies, dark mode preference etc...). There are thousands of commercial services and tools that help people work out if they are GDPR compliant or not, it shouldn't be that hard.
I don't see any plausible positive effect for Europe. I know the press releases hyping it up, but the product doesn't deliver. People mock shady companies that ride the AI hype wave. The AI Act is that sort of thing.
Give me one example where it is supposed to benefit the average European. Then we look under the hood and see if it will work that way.
In fairness, the bigger problems lie elsewhere. Information, knowledge, data is becoming ever more important and Europe reacts by restricting it, and making it more expensive. It's a recipe for poverty. Europe should be reforming copyright to serve society instead of applying the principle to other areas with the GDPR or the Data Act.
Are you trying to say that you believe that the AI Act will do anything about "algorithmic bias"?
But you're not able to explain how it would achieve that magic. You notice?
I'd highly recommend reading the AI act before making statements like this. When I get home I'll happily provide an explanation. Essentially, in tandem with other EU legislation it allows victims of algorithmic bias to investigate, prove, and be compensated for the bias they have faced.
Obviously, I have read the AI Act. Well, I think I skipped some bits that aren't relevant for me.
How else would I know that it's bad? Not from the press releases, right? You're not at work here, so it's ok to use your brain. You aren't at work here, right?
Any plans to provide a public figure exception on the biometric sections? I suspect most vision models won’t be available in the EU until that is refined.
The Biometric categorisation ban concerns biometric categorisation systems that categorise individually natural persons based on their biometric data to deduce or infer their race, political opinions, trade union membership, religious or philosophical beliefs, sex life or sexual orientation.
It wouldn't apply to the case you describe
>religious or philosophical beliefs,
Question:
If I show the AI an image of pope and ask "who?". It can not say that he is the head of the catholic church?
No, again, because the AI would be deducing his job, not his religion. The Human then deduces his religion from his job title. I don't think we need AI to tell us the pope is catholic.
You appear to be confusing "rationality" and "law."
Telling me someone is in the catholic church doesn't mean I then need to **deduce** they are Catholic. That is implicit in the original statement.
By the letter of the law, that is illegal.
Sure, you can apply rational arguments to this, but the law says what the law says. This is why many of us are complaining.
I think this is exactly the problem. In a field that is as early as AI, it is essentially impossible to have a tightly worded law that covers exactly the right areas. As a result you get a very vague law that where no one really understands what it means. I have seen first hand that this uncertainty causes companies to decide to move to other regions.
I'll go one step further: it is almost Impossible to have watertight laws on a fast moving topic like AI, therefore we rely on people using common sense. To claim, like some previous commenters have, that the law is rigid and binary, is totally incorrect. If it were, we wouldn't need lawyers.
And I will reassert again that we are talking about he use of biometric categorisation, which Is not what this is.
Exactly. But this is why it is so incredibly damaging if laws like this one are passed. The people creating these laws may feel good that they did something, but the result is you destroy businesses and force people to move to other countries in order to build companies. Over time, the EU becomes a technological backwater that has zero impact on tech. This is causing massive damage. It's everyone's responsibility to stop these laws from happening in the future and going after the people who create them.
I think there are very few cases where this uncertainty will remain when the AI act comes into force. (Codes of Practice, which explain how to apply the AI act to LLMs are coming, until they are published, the AI act does not apply, several months will be given to companies for compliance)
It's also worth noting that the AI act will impose few to no obligations on the vast majority of AI systems.
Finally, returning to our previous discussion, I'd like to again highlight that the biometric categorisation prohibition refers to a system using biometric traits to infer religious beliefs. That is absolutely not what LLMs currently do. Identifying a well known figure and pointing out his job is not the same as using a persons biometric data to guess their religion (if you ask ChatGPT to do that, it will refuse, btw).
Also race. Any multimodal model can determine race quite accurately, and right now the common reading of the AI act is that makes it illegal. Again, this idea of regulating the capabilities of a model that emerges naturally from training data just shows that the people who wrote this don't understand the very fundamentals of AI. The EU AI Act as it stands now massively hurts European the EU economy. The EU missed the Internet Revolution due to regulation, and as a result there are almost no successful EU internet companies. We are now doing the same thing with AI, and it will have the same result.
It would be pretty simple to prohibit an AI system from infering race. In the same way OpenAI have stopped ChatGPT from doing certain things.
We didn't miss the Internet Revolution due to regulation, we missed it due to lack of investment.
It’s essentially impossible to create a multi-modal modal that can’t detect race. And OpenAI hasn’t stopped ChatGPT from doing it at all, they maybe made it slightly harder. But that’s done at the app layer, not for the system.
And I personally left Germany for the US during the internet boom, and studied there in part because of regulation. Private modems were illegal in Germany and I couldn’t afford an official one from the Post. I had friends whose house was searched by the police for operating an illegal internet connection.
In the US, they were legal and early ISPs thrived, and they enabled early BBS systems and then Internet sites. By the time Germany deregulated and loosened the Fernmeldemonopol of the Post, it was much too late. I ended up starting a company in Silicon Valley.
Today, I see this as an investor helping startups. Why was Uber created in the US and not Europe? Regulation. Why do I have driverless taxis here and not in Germany? Regulation. Why can my Tesla drive me without me touching the steering wheel in the US but not Europe? Regulation. How about SpaceX? Regulation. Could you build Anduril in Germany? Not a chance. I think you really don’t understand what is happening day to day for startups. Capital is globally mobile, talent is globally mobile, it’s regulation (and to some degree culture) that drives where companies are being built.
https://preview.redd.it/qje0qurtw5rd1.jpeg?width=344&format=pjpg&auto=webp&s=f35351cb03a78b21938444419e5186564f5a6607
“Tell me about him”
Most normal answers (say echoing the Wikipedia page) involve violating the statute, no?
You can’t know that for sure. It’s all projected into a dense space. Useful to hear that you think the line should be “inferences made beyond data available in the text corpus” though. Maybe write that into the law?
It's not a question of wanting to: the EU itself can't legally regulate military use of AI.
But there are plenty of highly dangerous non-military applications
> the EU itself can't legally regulate military use of AI.
It sounds like you are half-way to a decent solution.
> there are plenty of highly dangerous non-military applications
Such as?
I am interested in high danger to individuals, e.g. in a specific scenario.
The AI act is focused on impact on individuals rights: AI powered CV analysis, AI powered justice (in the US, for example, their recidivism AI), Biometric Mass surveillance etc...
Again, a specific scenario with high danger to an individual?
All of those technologies could make society drastically better, and existing laws prevent the obvious cases for severe danger to individuals.
Again, our focus is on addressing fundamental rights risks, because we have existing regulation to address physical harm (Machinery regulation for example).
If you can't see the risks AI can pose to fundamental rights, then you shouldn't be doing AI
So you acknowledge that this isn't about grave danger to individuals but about more abstract issues.
I do see the risk, and there are real concerns there.
But the choice isn't between a favorable status quo and derisked adoption of beneficial technologies.
It's between a deeply non-ideal status quo (e.g. extensive human bias in CV analysis and justice) and adoption of beneficial technologies along with very real risks.
If we get this right the benefit greatly outweighs the harm.
The EU is already seeing the costs of not doing so. The world won't play along with your fantasies about risk-free progress, and global tech companies won't put their heads on the axeman's block and be subject to arbitrary interpretations of ambiguous laws. With the penalty being up to 7% of annual revenue if I recall correctly they would have to be insane to do so.
Id just like to go back to what you said about the highly non-ideal status quo: here you seem to imply AI will make more ethical decisions than humans.
This concerns me a lot, because it's a flawed argument I've heard so many times. My response would be, how are you training your AI? Where will you find the perfect, unbiased training data?
Take Amazon's attempt to use AI to pick developers on the basis of their CVs: it only picked men, because Amazon's dev workforce is predominantly male, and the AI was trained on their CVs. You could say the same of the US recividism Ai.
In this case, and in a great many like it, AI doesn't guarantee things will get better, it guarantees they will stay the same. It will repeatedly reproduce the status quo.
I don't want to live in a deterministic society where choices about peoples futures are made by machines with no understanding of the impact of their decisions. That's why we need to regulate.
There is room for improvement: I do not doubt that. The EU is currently preparing codes of conduct for GPAI (LLMs for instance), which will clarify what LLM Devs and providers have to do to comply.
Finally, a small counter-argument: I do not doubt that for now, the AI act is a barrier to some degree, but as codes of conduct come out, compliance will become simpler and edge cases will be ironed out. Then suddenly a benefit of regulation arises: certainty. Certainty about what is allowed and what could end up with you facing a lawsuit (because even if there is no AI legislation in the US (California may change that, and they are to some extent following the AI act), it doesn't mean AI companies can't face legal action for what their AI does.
I was merely pointing out that the existence of flaws does not necessarily mean it is worse than the status quo.
Notably Amazon did not use their failed experiment, so it is a poor example of actual harm.
On the technical level, more advanced AI systems that can understand the causal structure of the world are able to correctly reason about *why* something is the case. These are not doomed to replicate surface level statistical distributions if used in the ways you fear.
You might be interested to read [*The Book of Why*, by Judea Pearl](https://www.goodreads.com/book/show/36204378-the-book-of-why).
For example such a system can understand that the task in CV selection is to uncover causally relevant predictors such as educational attainment and job experience rather than purely correlational predictors such as gender.
You might object that this can still favor socio-economically privileged applicants and not redress historical inequities. This is true. The same has been observed of human CV screeners.
And at this point we go beyond the statistical meaning of bias to a social one. Our ethical intuitions on such matters are often not coherent when required to express them concretely, I think we should not blame the machine for our divided hearts.
> Finally, a small counter-argument: I do not doubt that for now, the AI act is a barrier to some degree, but as codes of conduct come out, compliance will become simpler and edge cases will be ironed out. Then suddenly a benefit of regulation arises: certainty. Certainty about what is allowed and what could end up with you facing a lawsuit (because even if there is no AI legislation in the US (California may change that, and they are to some extent following the AI act), it doesn't mean AI companies can't face legal action for what their AI does.
This is a good and compelling argument.
But how do you achieve that in practice? This isn't just about minor clarifications and helpful documentation, the laws will need to be substantively changed to allow adoption of mainstream technologies. For example it appears that OpenAI's Advanced Voice is illegal in the EU since it involves interpretation of the user's emotions. It might even be the case that the entire upcoming generation of multimodal models will be illegal in the EU for general use as designed because they have this capability.
I'm sure such an outcome wasn't the intention when making the law in question, but it was the effect.
And that will no doubt happen ever more frequently as the pace of development accelerates. With legislative timelines being what they are heavy proscriptive regulation seems like it must inevitably constrain adoption even if ambiguity can be solved.
I ran my idea through it. I see no path to make sure that I would be able to pass this.
>Ensure that the outputs of the AI system are marked in a machine-readable format and detectable as artificially generated or manipulated.
The idea would be for the system to mimic human responses closely, text and maybe audio and there's no room for disclaimers after someone accepts API terms or opens the page and clicks through a disclaimer.
Everything I want to do is illegal I guess, thanks.
I mean.. OpenAI are already finding a way to do this in the EU market, so it isn't impossible.
If you are building a chatbot, it doesn't have to remind you in every response, it just needs to be clear that the user is not talking to a human.
Nah, it's not reasonable at all. Technically possible? Maybe, with enough capital to pay off people researching what really needs to be a bar to cross off some fearmongering career asshole's wishlist as a requirement.
Maybe it's silly, but I have an artistic vision for a product like this. Those requirements make it inauthentic and I wouldn't be happy to introduce something with a goal of giving authentic feeling but with a backdoor. I'll stay a hobbyist, you aren't able to take away things I can do locally.
People deserve to know when they are speaking to a human being and when they are not. Misleading them is not ethical, and the fact that this is your goal is precisely why career assholes like me have to exist.
Users wouldn't be mislead. They open a website/app, they click OK on a pop up that informs them that they talk with a machine learning model. And from that point on, experience is made to be as similar to interacting with a human being as possible, getting user to be immersed.
When you go to cinema, do you see reminders that story shown on the screen is a fiction every 10 minutes?
This is what I meant in my previous comment: just saying once at the beginning of the conversation that the user is speaking to an AI is enough to comply with the transparency rules of the AI act, so your project will be fine!
I updated my previous comment for clarity.
I am not sure how that could get around the requirement of content being "detectable as artificially generated or manipulated" but I hope you're right.
I think here you have to focus on the goal, which is ensuring that people who are exposed to AI generated content know it is AI generated.
To do do, we should differentiate between conversational and "generative": for conversational AI, there is likely only one recipient, hence a single warning at the beginning of the conversation is perfectly fine.
For "generative" (I know it's not the best term, but tldr ai that generated content that id likely to shared on to others), some degree of watermarking is necessary so that people who see the content later on still know it is generated by AI.
If you reread my above comment, you will find that your proposed solution, which I described in the comment, would be completely adequate to comply with the AI act.
That a well-funded Microsoft-backed multibillion dollar company with a massive head-start can fulfill regulatory requirements is exactly what you'd expect, though. Regulatory Capture is going to be the way the big players maintain market share and seek monopoly.
As are MistralAI, a french startup.
Half the people commenting on Reddit about AI act compliance have no actual experience or knowledge of AI act compliznce.
I played around with that app and calling it "simple" is... an interesting take.
As someone who works in this field, with shit like this I can see why there's almost no AI work going on in Europe compared to the US and Asia.
I don't see it as too complex. It gives you a basic overview of what you need to do depending on your situation. What are you struggling with in particular? I'd be happy to explain.
As for the European Industry, we aren't doing too bad. We have a MistralAI, and a reasonable number of AI startups, most of which are (thankfully) not just ChatGPT wrappers. When OpenAI inevitably either increases its usage costs to a level of profitability, or simply collapses, I'm pretty sure a large number of "AI startups" built with ChatGPT in the US will go bust.
We are undoubtedly behind, but not because of regulation: it's because of lack of investment, and lack of European Capital markets.
It's also worth noting that the profitability at scale of LLMs as a service versus their potential benefits are yet to be proven (especially given the fact that most big LLM as a service providers, OpenAI included, are operating at a significant deficit, and their customers (in particular microsoft) are struggling to find users willing to pay more money for their products.
If it were up to me, I would not have Europe focus on LLMs at all, and instead focus on making anonymised health, industrial and energy data available to build sector-specific AI systems for industry. This would be in line with Europe's longstanding focus on Business-to-business solutions rather than business-to-consumer.
I am working in venture capital, and that's absolutely not true. We are investing globally, but the EU's regulation (AI but also other areas) causes many founding teams to move to locations like the US that are less regulated. I have seen first hand examples where this is happening with AI start-ups as well. And as a US VC, we are actually benefitting from this. But its still a poor outcome for Europe.
Issue is that we know we are regulatory compliment but still very often customer meeting goes on phase where we speak about 5-20 minutes regulatory stuff.
Even if the data is have been anonymized ? My assumption is if you comply with RGDP regulations your data would be valid be use as fine tune material, but I guess that in theory, practically in forcing RGDP might be mote costly.
I mean, not using GPT-4o would be the first step IMO. I thought closed source models a big no no in regulated industries. Unless, you consume it via Azure.
Yeah we are using Azure and our big part of our company is build top of Azure so it is not that big issue. Open models have pretty abysmal language support especially with smaller European languages so that is why we still using OpenAI.
>Unless, you consume it via Azure.
which is consumed a shit ton. we basically only do Azure stuff, because for 100 projects on azure we have only 5 projects on AWS....
I mean that's why Microsoft's big mantra is now being AI their company's center piece. "we aren't a cloud company, we are an AI company" is something you often hear Nadella is saying.
The same way any law is enforced.
But if you are wondering what you will have to do to comply, let me reassure you: you don't have to for personal use cases.
https://preview.redd.it/kwpcrsgtf3rd1.png?width=1080&format=pjpg&auto=webp&s=edb96a464d8c8704f7b33aa5f446f75a683c4d36
I live and work in the Netherlands
Backlash from Meta about EU regulation making it very hard for them to train on image data from EU citizens. Zuck said a few months back that those limitations would result in Meta not launching AI models in EU, and now we see that play out.
You are right! Let's see if Zuck continues this redemption arc and complies with the rules in my region
https://preview.redd.it/58olyl19f4rd1.png?width=1080&format=pjpg&auto=webp&s=b6a1ba041eb18be33cbf78cde15abe9afa0acfba
Hah I am touched really, but no need, I do not have any use case I cannot already do with Florence, and if I need it anyways I'll just proxy somewhere else. Main problem would be if I wanted to use it at work (which I do not at the moment) but just in case, I opened a ticket at their repo ([https://huggingface.co/meta-llama/Llama-3.2-11B-Vision-Instruct/discussions/28](https://huggingface.co/meta-llama/Llama-3.2-11B-Vision-Instruct/discussions/28))
Sure and this is completely fine if you’re a individual dev, clearly Meta are not going to sue you
But you can’t get away with this if you’re a business, especially a big one. Too much legal risk.
So what do you do? You accept the EU is just determined to over regulate and move your primary company out, denying the EU the tax
EU is literally regulating itself poorer
You could but if you try to build a product round it, the gubbmint will shit all over you.
Means like the cartoon says: there will be no AI tech companies in Europe.
Dumbasses.
Meta would be forced to choose between either:
* licensing Llama as Open Source software (removing restrictions, and likely complying with the minimum requirements set out in the OSI's upcoming Open Source AI definition)
* Keeping Llama as it is, but having to comply with the AI act
Comply with the ai act in this case means either not offering it in Europe or train the model again but this time without any data that was collected from EU citizens without their consent?
No, the AI act does not regulate use of personal data, but as far as I am aware, llama is not trained on EU citizens data (in any case, not without their consent).
The AI act is more about the risks the AI poses itself, if you Google AI act and go on the European commissions website it explains it well
I mean, I literally wrote the amendments that added it to the [final text](https://artificialintelligenceact.eu/?s=open+source&et_pb_searchform_submit=et_search_proccess&et_pb_include_posts=yes&et_pb_include_pages=yes), but sure, random internet person, I'm sure you're an expert.
Well given that, perhaps you could actually go and read the AI act before confidently claiming that I am wrong about a text that I myself contributed to writing?
Just a suggestion
“Article 53, 2. The obligations set out in paragraph 1, points (a) and (b), shall not apply to providers of AI models that are released under a free and open-source licence that allows for the access, usage, modification, and distribution of the model, and whose parameters, including the weights, the information on the model architecture, and the information on model usage, are made publicly available. This exception shall not apply to general-purpose AI models with systemic risks
Looks clear to me.”
You said it will matter if they consider it “open source” or not. Which is not true. Open source is a defined term not an undefined term.
They have to figure out if it aligns with the specifications set in that paragraph
I'm not the guy but to me, prohibiting manipulative or deceptive use, distorting or impairing decision-making. Like fuck. That's a wildly high bar for 2024's hallucinating AI's.
The rules apply when the AI system is \*designed\* to do these things. If they are \*found\* to be doing these things, then the issues must be corrected, but the law regulates the intended use.
On issues like biometric categorisation, social scoring and manipulative AI, the issues raised are fundamental rights issues. Biometric categorisation is a shortcut to discrimination, social scoring is a shortcut to authoritarianism, and manipulative AI is a means to supercharge disinformation.
> If your AI is too complex to fix, then citizens rights come first
Okay, let's think practically about this. So EU effectively bans AI. What do you think the outcome of this is?
> If they are \*found\* to be doing these things, then the issues must be corrected
I'm an AI engineer. How on earth would you correct for such a thing?
Right now I could go to chatgpt, and ask it to do social scoring and it will. So say I found that - how would you, as the AI engineer, now "correct that"?
> Creating such a system would be illegal.
Okay, so if I collected such data on a person and fed it to ChatGPT who then gave me the social score, who would be liable for that?
If you do that you are not creating an AI system, so not you. I expect OpenAI could be responsible in theory (in fact if you did try this, I'm not sure it would work), but in practice the application of the law requires common sense: the goal of the provision is to go after businesses and governments that are racking up information on their citizens and using it to rank them.
However I question the ability of LLMs to do this sort of reasoning in any case.
> OpenAI could be responsible in theory
Okay, great. Can you see the chilling effect that would have on OpenAI in the EU, and what would you expect OpenAI to do to "correct" that?
> but in practice the application of the law requires common sense
So you would expect the OpenAI lawyers to say "Oh, we're breaking the law as it's written, but it's okay because hopefully they'll have common sense to not sue us" ?
And again, what exactly would you expect OpenAI to do to "correct" it?
> However I question the ability of LLMs to do this sort of reasoning in any case.
I think you're greatly underestimating LLMs. I've fed huge text files into LLMs and asked it to pull out patterns, inconsistencies, etc. They are getting absolutely amazing at it.
As of now, the AI act does not apply to General Purpose AI, as we are in the process of drawing up a code of practice to give guidance on how to follow the AI act.
You raise an interesting question: **do providers of General Purpose AI have to prevent their GPAI from doing banned things?**
I'll be working on the drafting of the Code of Practice, it's a question I'll be sure to raise, so that GPAI providers get clear instructions on what they have to do. Thanks for raising a really challenging question.
I suspect that they (OpenAI) will be expected to do the same thing they have done with other uses they classify as unethical (to have ChatGPT respond that it can't do this thing). To some extent they have already done this with religion (ChatGPT outright refuses to try to identify a persons religion on the basis of their facial features)
> I suspect that they (OpenAI) will be expected to do the same thing they have done with other uses they classify as unethical (to have ChatGPT respond that it can't do this thing).
You know how trivial it is to get around that?
Just google jailbreak prompts. I use them to do taboo sexual roleplay with chatgpt.
> To some extent they have already done this with religion (ChatGPT outright refuses to try to identify a persons religion on the basis of their facial features)
meh, I played with it, and found it pretty trivial to work around. Would this now make OpenAI liable and I could sue them with this law?
https://preview.redd.it/xe94cr1u8drd1.png?width=1674&format=png&auto=webp&s=b45beb7d3e8510b24693a7ac267edcdacc85f0a2
> Biometric categorisation is a shortcut to discrimination
And yet, a general-purpose vision-language model would be able to answer a question like "is this person black?" without ever having been *designed* for that purpose.
If someone is found to be using your general-purpose model for a specific, banned purpose, whose fault is that? Whose responsibility is it to "rectify" that situation, and are you liable for not making your model *safe* enough in the first place?
Most cameras can do that as well, as part of their facial recognition software - yet cameras are legal in the EU.
The entire thing is a non-issue... and the fact that Meta claims it is an issue implies they either don't know what they are doing, or that they are simply lying, and are using some prohibited data (i.e. private chats without proper anonymization) as training data.
If you use it your self hosted GPVL and ask this question, nobody is coming after you, it a company starts using one for this specific purpose, hey can face legal consequences.
> nobody is coming after you
To be clear, are you saying that the law exempts you, or are you in favor of passing laws in which lots of use cases are illegal but don't want enforced?
In the past such laws have been abused to arrest and abuse people that you don't like.
Yeah, the way the regulation is written, it affects how the AI system is used, not whether it is fundamentally capable of something - otherwise a simple camera would be illegal, considering it is "able" to store information about someones race or gender.
The process of “finding” is very one sided and impossible to challenge. Even providing something that may be perceived as doing it is an invitation for massive fines and product design by bureaucrats.
From Steven Sinofsky’s substack post regarding building products under EU regulation:
> By comparison, Apple wasn’t a monopoly. There was no action in EU or lawsuit in US. Nothing bad happened to consumers when using the product. Companies had no grounds to sue Apple for doing something they just didn’t like. Instead, there is a lot of backroom talk about a potential investigation which is really an invitation to the target to do something different—a threat. That’s because in the EU process a regulator going through these steps doesn’t alter course. Once the filings start the case is a done deal and everything that follows is just a formality. I am being overly simplistic and somewhat unfair but make no mistake, there is no trial, no litigation, no discovery, evidence, counter-factual, etc. To go through this process is to simply be threatened and then presented with a penalty. The penalty can be a fine, but it can and almost always is a change to a product as designed by the consultants hired in Brussels, informed by the EU companies that complained in the first place. The only option is to unilaterally agree to do something. Except even then the regulators do not promise they won’t act, they merely promise to look at how the market accepts the work and postpone further actions. It is a surreal experience.
Full link: https://hardcoresoftware.learningbyshipping.com/p/215-building-under-regulation
You misunderstand, you are not expected to proactively search for cases of your AI doing something illegal. You are expected to rectify its behavior if any are found by yourself or your users, and you are expected to evaluate the potential risk of your AI doing illegal things.
As a reminder, Open Source AI is exempted from the AI act, and the AI act only applies to AI that is "on the market" (so not your backroom usage)
Open-source AI is not exempt from the AI Act if it meets the "systemic risk" requirement.
> 1. A general-purpose AI model shall be classified as a general-purpose AI model with systemic risk if it meets any of the following conditions:
> - (a) it has high impact capabilities evaluated on the basis of appropriate technical tools and methodologies, including indicators and benchmarks;
> - (b) based on a decision of the Commission, ex officio or following a qualified alert from the scientific panel, it has capabilities or an impact equivalent to those set out in point (a) having regard to the criteria set out in Annex XIII.
> 2. A general-purpose AI model shall be presumed to have high impact capabilities pursuant to paragraph 1, point (a), when the cumulative amount of computation used for its training measured in floating point operations is greater than 1025.
> 3. The Commission shall adopt delegated acts in accordance with Article 97 to amend the thresholds listed in paragraphs 1 and 2 of this Article, as well as to supplement benchmarks and indicators in light of evolving technological developments, such as algorithmic improvements or increased hardware efficiency, when necessary, for these thresholds to reflect the state of the art.
The 10^25 FLOPs threshold most likely means that Llama3-405B is alread presumed to have systemic risk with its [4*10^25 FLOPs](https://x.com/karpathy/status/1781047292486914189) of training budget. Meta hasn't released an official figure, I believe, but it's ultimately up to the Commission to adjust or ignore that number anyway. The objective is to associate anything *impactful* with liability for whatever impact it may have. That's absolutely going to deter research.
It won’t deter research, in an R&D setting in a company or university you can explore and test these things freely, but if you release them to the public you are and should be accountable for them.
And when it comes to the Digital Markets Act and this article, it is UTTER bullshit.
The EU passed a law, with the aim of opening up Digital Markets, and preventing both Google and Apple from abusing their dominant positions in the mobile ecosystem (the fact that they get to decide what runs on their platform).
There were clear criteria on what constitutes a "gatekeeper": companies with market dominance that meet particular criteria. Apple objectively meets these criteria. Given that, they have to comply with these rules.
Should apple feel they do not meet the criteria for compliance, they can complain to the regulator, should the regulator disagree, they can take it to the European Court of Justice, **as they have done on a great many occasions up until now.**
Condescension and straw manning from a euroweenie? No surprise there.
There are multitudes of reasons why the EU sucks and most of it has to do with over-regulation.
From what I've read, this is basically it. It's not really AI related, it's data privacy related, which the EU is quite strict on (GDPR).
Honestly, I would tend to agree. I mean I'm pro-AI (Obviously, I mean I'm posting here!) but still, you can't just use people's personal data to train your model without asking them...
The right to privacy isn’t absolute, you have a right to privacy in your home but it is totally reasonable for the police to violate your privacy and come into your house with a warrant.
Now how you implement this for end to end encryption is a more complicated issue and has to balance other things but the base principle is valid.
I agree with this. But what they have in mind is completely different. What they want to do is similar to Apple's CSAM. They want to make phone manufacturers include an AI which scans all your pictures/text messages to check whether if they contain "illegal" content. At the same time, they want to exclude government employees from it for "security"
There was a whole video on this from multiple people, I'd recommend you check it out
[https://www.youtube.com/watch?v=SW8V\_pZxmq4](https://www.youtube.com/watch?v=SW8V_pZxmq4)
there's a huge difference between getting a warrant through proper channels for probable cause and violating people's privacy as a matter of course because they think it might impede their ability to investigate.
Man, it's extremely simple. Not sure what your level of education is but pretty much anyone literate can understand it.
https://www.facebook.com/terms.php?ref=pf
This is like someone getting into a fight over being caught in someone's video in the park. If you put stuff in public, then it's in public and the expectation of privacy. I can't get over how people putting stuff in public for public use and then get made when the public takes them up on the offer.
You present this example as if it's univerally accepted that you can film someone in public, in this case focusing on those in a fight, without concent. It's not universal. The U.S. legal bubble is not universal.
The example you used is very much not universal.
Read up on EU laws, and e.g. local variations like France and Germany.
I get what you're saying, and it's a good point, but we're talking about a company using the data, not just someone's boss seeing their employee goofing off on facebook and firing them.
It might be *legally* ok to use someones public photos like this, but there are ethical considerations with it.
I would say the same thing if someone took someone's facebook photos and used them commercially in some way. It might be "public" but it's still someone's personal data, it's not really "fair game" to use it anyway you want.
> cheaper and less regulated. we are at a point where more regulation kills far more people than it saves.
I have not seen a more hilarious statement than this for a while. **claps** well done!
You may wish to have a peek at history to find out why we have regulations on things like, [I dunno... flour](https://www.thermofisher.com/blog/food/a-history-of-food-fraud/).
> It became routine for flour to be extended with chalk or gypsum dust, and for milk to be watered down to the point that chalk was used to re-color it white.
Arguing that medicine is too expensive due to regulations is fucking hilarious, and one can only assume that you're American.
Do you think this is because the EU regulation would forbid the usage of LLama 3.2 or because Meta is anti regulation and is doing a political move here? I mean Llama 3 is still available and the EU regulations mostly affect high risk models, what could have happend between 3.0 and 3.2 that changed the models so rapidly they cannot be made available anymore? Which part/paragraph of the EU regulation is it that prevents us from using the LLama3.2 models.
Thanks for thr help!
I don't really think so, as that would be extremely stupid by them...
Apples recent AI-feature move makes at least some sense, as it will directly affect some end users, which might put some pressure on the EU. However, no end user will be directly affected by EU customers not being able to use the Llama model, since the corresponding businesses will just use some alternative instead, without the end user even noticing...
The model was trained by illegally (in EU) scraping user data from the photos posted on facebook.
In europe you can't consent to something that doesn't exist yet and most facebook accounts were created before the rise of language models.
Does that mean, everyone in Asia, Russia and America etc. will be able to ask detailed questions about a Facebook user from Europe, just Europeans will not?
Sadly yes. Facebook hopefully did its best to scramble the input data but the model can be tricked into spitting out personal details anyway.
It's called "regurgitation" if you are interested.
[https://privacyinternational.org/explainer/5353/large-language-models-and-data-protection](https://privacyinternational.org/explainer/5353/large-language-models-and-data-protection)
Right, others think it is more important to win the AI race for max profit as looking on such critical things that bring them no money. Instead, it could cost them a lot of money.
EU lost on AI with that, because it's clear that some countries will do anything to be ahead in AI, so if you put obstacles in your own way, don't be surprised if you stumble.
And that's why I feel caught between two stools here, I can absolutely understand both sides, but they are not compatible with each other...
> EU lost on AI with that
Well, Mistral Large 2 is the most efficient large LLM, Flux is the best image generator AI, and DeepL is the best translator. The EU is arguably doing very well.
Well, yeah, that is right. I like Mistral myself very much since their first release, especially because they train it also for German and finetunes based on their model they always was the best on that language and on top much less censored. I also use DeepL since it exists (but it begs more and more for money). Didn't used Flux yet, but heard about how good it should be compared to SD(XL).
So, yes, in that point that is right. When it comes to AI itself, it looks very good for us in EU.
But that is not the problem here. The EU regulation is more about using this AIs in your own product and this is where companies are being slowed down.
And especially here Meta is a special case. They have a tough standing in the EU in general, because of various things in the recent past as well.
> The EU regulation is more about using this AIs in your own product and this is where companies are being slowed down.
I am not sure about that... Much of the regulation equally affects American and European companies when they want to service European customers.
So, which aspect do you believe really puts European companies at a disadvantage?
EU companies cannot take initiative on anything. Its a systemic problem spawning from lack of talent pool, language barriers, and confusing government relations discouraging new companies to even try to succeed.
+1 from me mate. I am pro GDPR but there are a lot of inherently other issues that cripple tech companies across Europe. Except if you are in Germany where a nice corporate bribery will solve everything.
EU citizens can use the model, the license is worldwide.
But Meta will not deploy the model in their EU services because the AI act requires disclosing the source of the training data, and proving that it's not trained on illegal data.
Note that if the model was trained on EU data without consent, then by the GDPR, legal action can be taken to force meta to remove that data. Irrelevant of where that data is stored. Its just very hard to prove that if Meta does not disclose its data source ;)
The AI Act is not yet active for LLM (classified as General Purpose AI - aka GPAI). The regulation for GPAI should be enforced from (may?) 2025, and in practice after the AI Office of the EU is operational.
Here's a summary of the requirement, they are more severe for closed AI. It applies to any AI service trained or deployed in the EU, including OpenAI (which engaged itself to comply sooner than required)
>**General purpose AI (GPAI):**
>All GPAI model providers must provide technical documentation, instructions for use, comply with the Copyright Directive, and publish a summary about the content used for training.
>Free and open licence GPAI model providers only need to comply with copyright and publish the training data summary, unless they present a systemic risk.
>All providers of GPAI models that present a systemic risk – open or closed – must also conduct model evaluations, adversarial testing, track and report serious incidents and ensure cybersecurity protections.
The exact quote for the data source is:
>[Article 53](https://artificialintelligenceact.eu/article/53/), 1.(d) draw up and make publicly available a sufficiently detailed summary about the content used for training of the general-purpose AI model, according to a template provided by the AI Office.
I don't think the template exist yet.
>will be able to ask detailed questions about a Facebook user
You will not be able to ask that anywhere, at all.
It's part of the training data of the model, you can't query it specifically for individual information.
Thanks! But what about the 1B and 3B text models? If they are just derived by distiallation of the 8B and 70B models it should not be a problem, right? Are they available in the EU? Sorry cant check atm, I'm on holiday in Asia :D
The smaller 3.2 text models are available here in Italy.
The text part of the bigger 3.2 models didn't change from the 3.1 version. A text-only 3.2 70b and the 3.1 70b are the same.
To me this is pretty transparently a part of the ongoing feud between Facebook and the EU, that rules that the main source of income from Facebook is actually illegal. This is a punishing move, not a cautious one.
I see regulations as a symptom of a deeper cause: an average European is more risk-averse and values work-life balance.
And as a person working in software development with a touch of AI, I am actually questioning the actual value of these products, at least in their current form.
Well, also the EU can protect their own industries with regulation (tariff barriers being the other main mechanism). The danger then is that those industries can become lazy and rely on that protection instead of innovating or investing in newer technologies.
Preventing countries from selling their products under market value and competing unfairly is a legitimate thing to do.
As for our own industry, they have to follow ever stricter regulations, and are actively innovating to meet those requirements.
There are a number of EU manufacturers with decent electric cars available, and prices are dropping. Allowing Chinese manufacturers to flood the market with vehicles sold under the cost of production, and not necessarily meeting EU safety standards, would be utter insanity.
"Under market value" is a bit subjective. There are economies of scale and lower labor costs to consider. Additionally the EU has provided various subsidies for EVs including infrastructure, research etc.
The Norwegians seem to be taking full advantage of the competitively priced Chinese vehicles.
Core problem is that only protecting existing industry is a failing strategy long-term. We need an explosion of new AI startups that explore the possibilities, try and fail, so that at least a couple of them end up growing into big new industries in Europe. With the current regulation that's not going to happen and the explosion of AI startups will happen elsewhere. Leaving EU to have a slowly declining industry with limited potential. Not great for the longevity of European society.
It is already happening with cars now EU is pushing more regulation because German carcompanies cannot build proper software and batteries for their cars.
Chinese EVs will get heavily taxed because they're much cheaper than European ones: https://www.sneci.com/blog/eu-to-impose-taxes-on-chinese-electric-vehicles/
Being supervised "Chinese" style like in UK and US is not something people are longing for. If AI companies aren't able to make money without supplying tools for opression they have no right to live.
There are viable companies for AI out there
Not really? If the regulation sucks everyone will just move to using the cheaper, freer, and actually available alternatives. Eu is basically just giving away its future to china in one more way with this.
I don't think they the regulations are perfect.... But at least we have them.
They can be refined. My main use for ai these days has been for spelling corrections when i need to reply to tickets to clients on my Jira board...
And yes I work in software dev as well
> They can be refined.
Sure, but once the EU gets to that point it'll be left long behind. The regulations will be refined so that EU users can make use of American and Asian AI products.
>At this point the EU is creating regulations based on hypotheticals from the imaginations of its bureaucrats, not observed issues.
They always have done, that's a large part of it's existence and how it justifies itself. And EU zealots will defend it.
>At this point the EU is creating regulations based on hypotheticals from the imaginations of its bureaucrats, not observed issues.
I mean, that's exactly what the US is doing. The US is operating under the hypothetical presumption that regulations would do more harm than good, whereas the EU is moving forward with the opposite hypothesis. It's impossible to know which will be more beneficial for integrating this new technology into society except with hindsight. It went one way with the tech boom the last few decades, with AI it could absolutely go another.
I agree, regulation is not perfect. Yet, having a discussion about what should be regulated and how exactly is very different from saying "all regulation bad".
I am working with LLMs and there is simply no economical need for better models aka improved zero shot performance. Even with performance boost, I would never change the model in a production environment, because everything else is built around the model and it's behavior.
When you consider OpenAI is making a multibillion dollar loss and has no path to profitability, you start to realise precisely how fucked the situation is.
That's a bad example though, because OpenAI is still technically a nonprofit/capped-profit company. When they shift gears to being fully for profit, you're likely going to see some big changes in their monetization strategy.
At a guess, They'd have to multiply their current pricing by 4 to get anywhere near profitability, and that is with the discount compute they already get from microsoft.
I'm worried that when they do, an entire ecosystem of AI Startups will die, and a large chunk of their customer base will leave.
But the reason they are moving to a for profit status is to attract investment. The problem is that the issue isn't the non profit status, it is that they really don't have a workable pathway to monetisation
That entirely depends on whether you believe they can create autonomous agents or AGI and what kind of value people place on those things. That's the big gamble for all AI companies right now, right?
You make a good point: if OpenAI can deliver the technical leap required to reach that stage, then the investment may have been worth it (although I do wonder what applications for AGI are worth the likely insane compute cost), but to be honest, given the recent releases, I'm not convinced there is a pathway from LLMs to AGI. I could be wrong, but I just don't see it happening. In the meantime OpenAI continue to make their LLMs more and more complex, and more and more energy-demanding solely in order to imitate AGI. That isn't a good sign.
To be fair, OpenAI has been simplifying their LLMs and making them more compute optimized ever since GPT 4. That's reflected in the pricing as well. Even o1 is not more expensive than GPT 4. My take on that is that they learned their lesson on compute for inference with GPT 4 and will make sure that each model from now on requires less at inference time even if it's a better quality.
Risking what? That companies use people's private data, including pictures, to train AI models? That companies start using models to make automated decisions for critical things (like an insurance provider deciding if you get a payout or healthcare)?
Seriously, the EU AI regulations are not that hard to understand. Main things are controls around how you use AI, and how critical their decisions are (to avoid e.g. Airbus using ChatGPT to design a plane and it falling out of the sky afterwards); and use of private data.
If Americans want their healthcare to depend on AI decision making, and for mass layoffs from various sectors, and their private data to be used to train models, it's fine by me to be left behind on those.
You are writing on an US platform using an US phone with US OS. You found the platform using an US search engine. The platform is financed by US ad companies and hosted by US cloud services. Meanwhile our car industry is burning down. Remember when we were leading the solar industry? Now China is dominating the market, because instead of coming up with regulations they just built a ton of solar power plants.
That's the risk, and our kids will have to pay for it. I think everyone should start a company in Germany once in their lives just to experience the current state of regulations. Doesn't have to be that long, 1-2 years is enough. It's easy to talk if you are not the one who has to pay the compliance cost for all this nonsense.
> You are writing on an US platform using an US phone with US OS. You found the platform using an US search engine. The platform is financed by US ad companies and hosted by US cloud services.
Running on servers manufacturered all around the world, on chips manufactured in Taiwan, designed in the UK, built on tech from the Netherlands and Germany. From a phone designed in Taiwan and manufacturered in China with a screen built in South Korea. We can go on all day.
Germany != the EU.
I've started companies in two different EU countries (not Germany) and the experience has been very fluid and intuitive. The terrible experience with German bureaucracy has no relation with compliance to EU regulations about privacy or AI. You're just mixing random complaints you have together and blaming the amorphous blob for everything.
>Meanwhile our car industry is burning down
Whose car industry? And let me guess, EU AI regulations are guilty for that as well?
Right, but wait, if AI is brought into the “right” form, then the country/company that has the best product in this area and therefore has a lot of power over others will win. And that's where the AI race comes from, no one wants to lose to a competitor here and therefore anything goes.
Everyone knows how much power there is in the potential of an AI and therefore also a whole lot of potential to earn a really insane amount of money with it. Human greed knows no bounds.
Anyone who seriously believes that AI is about improving humanity is underestimating the greed for power. You can already see it in OpenAI's proposal to regulate hardware, which they only want to use to eliminate the open source competition that is dangerous for them. Of course, they claim otherwise to the outside world, logically. But as I said, in the end it's always about power, wanting to dominate the market, striving for a monopoly.
I mean, I don't work in tech but I happen to be in a bubble of people who do and a single high quality dev can now do the work of a single dev and 3 junior devs together. That's substantial imo.
If I understand it correctly, you seem to be saying that higher productivity leads to a better work-life balance? If this is the case, then the U.S. should be among the top countries in terms of work-life balance, since it has one of in not the most productive workers in the world. As far as I know, this simply isn't true.
No, that's not what I'm saying. I'm saying it enables such an end case: you still need things like legislation, unionization, and cultural/social traditions that value such things to make that a reality.
Higher productivity, in Europe, has this far meant steadily decreasing work hours. Germany has an extraordinary economy, for example, and works very little compared with most of the planet, thanks to this consistently high productivity.
To be fair, the first steam engines weren't *that* valuable or productive outside of very niche cases... fortunately the steam engine wasn't regulated then.
And it resulted in horrible explosions that killed a lot of people, after which the invention of a *steam governor* was a cruicial step to making it safer. :3
Like how most European companies are in violation of GDPR, Mistral almost certainly uses illegal training data. The fact that they won't be investigated, but the threat of prosecution is so high American companies can't even release in the continent should let you know whats going on.
Or maybe American companies are just incompetent at following regulations, since they are used to buying legislators when needed rather than actually doing what the regulation requires them to do.
For example, the Claude models were not available in the EU for a long time, despite them being available in the UK... presumably because the people at Claude didn't even know that the EU and UK are using the same regulation!
Or, why did it take so long for OpenAi to offer their "memory" feature in the EU, considering the only relevant point for them was that they would need to store the memory-data on EU-servers rather than USA-servers?
I mean really... this is an American company fail, not an EU fail.
It entirely depends on how anal the regulators are. Technically, anyone funneling their Apache logs to a SIEM are probably in violation of GDPR in practice.
Ah, moving the goalposts, are we? The fact is, Flux operates out of Germany—regardless of where their legal entity is registered. But thanks for the bureaucratic lesson!
The fuck are you talking about. If the company has to have their legal imprint in the USA even though they operate out of Germany they obviously have this setup for a specific reason.
You're missing the point. I was simply stating that Flux operates out of Germany. Debating their legal setup isn't relevant—especially since we don't know the full details.
No, you are the one missing the point. This entire thread is about EU AI regulation. In that context you can't say "Nuh uh, we have flux" when flux is legally a US company...
Flux operates and is based in Germany. The fact that they have a legal entity in the US doesn't change where they conduct their business. It's still possible to start and run AI companies in the EU despite the regulations. Never said I am a fan of the regulations.
>where they conduct their business
This is simply irrelevant in a discussion about regulation.
You just can not make the point that regulations aren't killing startups by providing flux as an example since it literally is an example for the complete opposite.
You're really trying to twist my words, huh? I never said the regulations *aren't* killing startups. The original point was that someone incorrectly stated Flux is based in HK, and I simply clarified they operate from Germany. The whole "legally US" thing doesn't change that fact. The meme says, "we have no tech companies," and guess what? Flux is one. Whether or not you like the regulations is another discussion entirely—one I never argued about in the first place. You're basically arguing with yourself at this point.
The point is that we DONT have flux if it is legally a US company.
Do we have Apple, Meta, NVIDIA or Google because they have remote workers or even offices that are in the EU? Fuck no!
Flux is totally irrelevant to the discussion because legally it is a US company and that is all that matters in the context of this discussion.
At this point, you're just being intentionally obtuse. Flux operates from Germany—that’s a fact. The legal setup is irrelevant to my original statement. The thread claimed there are no AI companies in the EU, and Flux, operating from Germany, proves that wrong.
Also, comparing Flux to giants like Apple or Meta is laughable. They're *based* in the US, with EU branches. Flux? It’s based in Germany with a legal entity in the US. Totally different. You're nitpicking legal technicalities to prove a point no one was even arguing about.
Ok lets go back to the very beginning where you say:
"the official site says germany, no?
https://blackforestlabs.ai/impressum/"
and I said
"No, the DE stands for Delaware..."
the official site, which you used to "prove your point" literally doesn't mention Germany at all
It's wild how you're still hung up on that, considering I *immediately* clarified that the legal imprint is Delaware and the company operates out of Germany. I acknowledged the Delaware registration right away, so your point about "not mentioning Germany at all" is moot. You're dragging this out by fixating on a side detail that I already addressed, instead of the main point: Flux *operates* in Germany, no matter where their legal entity is registered.
And Flux is a US company, no matter where they operate.
Flux is not a win for the EU but a highlight as to how terrible it has already gotten to the point where it doesn't even make sense for companies to register here anymore even if they operate out of the EU.
All that aside though, your initial comment trying to correct someones false comment was just plain false itself.
It’s funny how we’ve come full circle. Yes, I used the wrong wording initially by referencing the legal imprint, and I corrected myself right after. The *entire point* was correcting the "based in Hong Kong" claim, which is *factually* wrong—Flux operates from Germany. Whether they have a legal entity in Delaware doesn’t magically change where the company is based or operates.
The regulations argument? I never said I liked them or that they don’t make things harder. That was never my point.
To both of you: companies can do business in multiple countries. You're both right, you're both wrong. Look up tax law before you start arguing about it.
Panama must be the best place in the world to set up a business then, anyone who's anyone [is registering there](https://en.wikipedia.org/wiki/Panama_Papers).
You're missing the point. I was simply stating that Flux operates out of Germany. Debating their legal setup isn't relevant—especially since we don't know the full details.
Dear american citizens that love these memes, you dont have tech companies. The tech companies are owned by the rich people raping your rights to the point of using your private conversations(Meta, ClosedAI, Google, twitter) to train models to manipulate you and your society into making the choices the owners of those tech company want. Dear american citizens, you dont have companies, you are flock.
Dear american citizens, in this cotton movie you arent the planters, you are the slaves.
And in europe we are trying to prevent that, we dont want to be you. We want AI laws that protect our privacy. And what you see is tech companies attacking EU because those companies cant do in europe what they did in usa. And because those companies are afraid that the rest of the world will follow EU example on data and privacy protection. Including usa, where some states are approving laws protecting people, [like illinois](https://pirg.org/updates/the-illinois-data-protection-and-privacy-act-is-one-of-the-strongest-privacy-bills-nationally/) .
Dear islamified welfare queens in the EU. Get a fucking clue = your sorry counties can't even fight your own battles . Most of the welfare socialists would be speaking Cyrillic if it wasn't for Americans pulling your repressive asses out of the fire all the time.
Nah, we own stock in these tech companies. They've already made me a bunch of money in my retirement accounts and will help me retire early.
One of them also pays my salary, which is quite a bit better than I'd find in Europe...
As minority shareholder you have no voice, no power and no vote in those companies. And you are happy because you gave them money to help them control society, so people has no voice, no power and no vote.
You are no planter, you are powerless, but you trick yourself into thinking you are special, and this is true, there is something special in a slave protecting its patrons.
The "vibrant high-tech economy" of usa doesnt translate in usa citizens living better each year, it translates in usa citizens living worse each year while the biggest companies have record profits. The AI companies are a tool that those big companies will use to make their profits even bigger by making the life of the average american citizen even worse.
Thats why is important to restrict AI companies to control society or the economy. And open source models, and data and privacy laws are the key for that.
Meta share the same despicable goals than closedAI(microsoft), anthropic(amazon),google... but we are temporal allies, because both Meta and us (open source people) want to kill the business of closed source models. But make no mistake , when Meta "aligns" with closedAI or anthropic against data and privacy laws is only showing its true colors, because the goal of Meta is achieving those goals using open source software, the same way that google does with android on phones.
I'm not american, but I am curious. What's the correlation between quality of life and high tech-economy? You point out Europe's quality of life is better even if their tech industry is weaker, but that correlation only works between the US and EU, if you take any other region the correlation won't hold up.
Lol. How am I a slave, exactly? I'm making bank!
> The "vibrant high-tech economy" of usa doesnt translate in usa citizens living better each year
Of course it does, silly. The quality of life gap is only growing between the US and Europe. European welfare systems will soon collapse, and the median American is already retiring years earlier than the median European.
[April 13, 2023 – U.S. life expectancy has declined to 76.4 years, the shortest it’s been in nearly two decades,](https://www.hsph.harvard.edu/news/hsph-in-the-news/whats-behind-shocking-u-s-life-expectancy-decline-and-what-to-do-about-it/)
the life expectancy of EU is [81.5 years](https://ec.europa.eu/eurostat/web/products-eurostat-news/w/DDN-20240503-2)
---------------
The retirement age in usa is 65.2 years for men and 65.3 for women.
In EU is 62.6 and 62.3.
[page 2 OECD ](https://www.oecd-ilibrary.org/docserver/e4d8d9b3-en.pdf?expires=1727378440&id=id&accname=guest&checksum=393C8736EFABCE1E172EBA72B4D50E44)
-----------------
Im going to explain you the american "welfare" system, my best friend is american and had a heart attack last year. After the treatment he got many bills for a total of $250k, if he is lucky, the insurance company will pay half.
In my country that costs 0 euros.
When i asked him how he will pay the bills, he said me he will add them to the debt he still have from the college. He is close to the age retirement, and his future is living in poverty because he had a college education and a heart attack. Thats America for many americans.
lol, it doesn’t cost you zero euros. You pay out the ass in taxes! Americans have a significantly lower tax burden than Europeans.
As for the life expectancy, most of the difference comes from deaths of young people from drug overdoses. If they don’t do drugs or join a gang an American will love just as long. And if they get cancer, for instance, they’ll actually have a higher survival rate than a European.
We’re looking at the median. The quality of life I have here is indisputably higher than it would be in Europe because I earn about 4x more than what I would earn there. The median American makes much more and retires earlier.
Furthermore, the US economy is surging while Europe’s is stagnating. You guys won’t be able to pay for your welfare systems in the near future, and then you’ll be absolutely fucked because everyone there is utterly reliant on the government to get by.
That day is going to be very difficult for you guys.
As a dev i want to say many EU regulations have very good intentions and solid reasoning. In practice a lot of stuff needs refinement of course.
Also there is plenty of room in Ai world. It's good when does it's thing while other do theirs ...
I think you are looking at it the wrong way. Instead the EU wants it's citizens to have privacy and choice, rather than big tech giants turning citizens into a product.
Are we talking about privacy after what just happened with Telegram? Privacy had been eroded
If they allowed choice then Apple would be able to launch Apple Intelligence in the EU and GPT voice mode would be available as well. The choice is made for the citizens
But that would allow apple to dominate that segment and push any competition out, as well as removing consumer choice.
I'd personally take EU laws and regulations any day over the US ones.
Can someone explain why eu regulations are so bad? The goal is to help people not corporations. Corporations aren't your friend. I truly don't understand Americans:my job exploits me like slave and I enjoy it.
People are homeless because they're addicted to drugs, not because Bezos has billions. The reason why the EU is in decline is because they are too used to being ruled by authoritarians, as they have throughout their history. The US will always be on top because it values freedom over authoritarianism, allowing people to become billionaires by objectively making the world a better place.
> People are homeless because they're addicted to drugs, not because Bezos has billions
Tell me you've never been outside the US without telling me you've never been outside the US.
Because the average US citizen considers themselves a temporarily embarrassed multinational CEO, and thinks that regulations prevent him from fully realizing his destiny, while the megacorps keep squeezing more and more value out of his minimum wage pittance.
Yep pretty much...
The upside is, of course, that "magical American ingenuity", which does, in fact, also lead to more great companies. But, this doesn't really benefit the average American, considering that people in the EU can use Google just like Americans can, while the fact that Google is an American company doesn't really benefit Americans, since Google is able to use all of its lobbying to rewrite American laws in its favor.
And sometimes, it's even "worse" than that: During the recent tech layoffs, Google wanted to layoff various European developers - but then they found they couldn't really do that very much, due to European regulations. So, they fired more American developers instead...
> The upside is, of course, that "magical American ingenuity", which does, in fact, also lead to more great companies. But, this doesn't really benefit the average American
Well, this is the best summary of the situation I've read recently. Spot on.
I don't know about EU regulations in particular, but regulatory capture is a thing that can happen. Basically, regulations are written in a way to reduce competition in a field by making it too expensive for competitors to operate in, and/or making the barrier to entry too high for newcomers. The end result is fewer players in the field, then competition and innovation goes down.
[https://en.wikipedia.org/wiki/Regulatory\_capture](https://en.wikipedia.org/wiki/Regulatory_capture)
Honestly, I have *much* more faith in the EU do stay clear of regulatory capture, than any American law makers... the latter are much more likely to be simply bought by powerful American companies.
I also used to be somewhat Anti-EU in the past, but they have done a great job in the context of the Ukraine war, and it also got me to more generally look at the type of regulations they are doing - and the regulations are generally actually pretty good.
Also, there are many powerful actors out there who want to paint the EU in a bad light: Major American companies, China, Russia, the American government to some degree, and even various European national governments frequently blame the EU for stuff which is really their own fault. So, pretty much any information you might ever read about the EU will be distorted towards the negative, and as such you should always pay close attention to whether the specific Anti-EU arguments really make sense (because usually they don't make sense).
Because if corporations decide they need AI, they won't be coming to EU - > less jobs for Europeans. All jobs exploit is, good thing you are so self-aware
That's a very vague question, regulation can be good or bad. GDPR is mostly very good, while the AI regulations made absolutely no sense. Feels like you're trying to rile people up with this comment.
GDPR is a great regulation. If USA has same regulation a lot of scumbags would be rotting in prison right now, while been bankrupt (Microsoft, Amazon, Google, insurance companies, even your pizza shop etc) because they scoop and sell your data to each other for profit.
Problem is GDPR was made in a period that LLMs didn't exist. So now we have the problem where Llama 3.2 Vision (not the text version) is banned in the EU because during training, images from Instagram were used without those images been included actually in the LLM.
Trying to fix this problem could take years if not decade. And the MEPs (Members of EU Parliament) majority are dumber than rocks and only are there to make money. Such complex stuff are way over their head. They are so dumb that they voted for the re-writing of European History earlier this year, and when call out the local MEP what he voted for, they look at you like Zeus hit them with lightning bolt. They don't even read what they vote for. I do hope there will be some tech savvy German or Dutch MEPs trying to fix this. Alternative never will.
GDPR is great because it has severe penalties that large tech companies may actually take seriously. It's great specifically because it's one of the first laws that includes enforcement provisions that go beyond a meaningless slap on the wrist.
It is, however, still largely ritualistic bureaucracy. It hasn't done anything to mitigate the enshittification of online services because the driving force there is venture capitalism, not the lack of "designated data protection officers" in small businesses or whatever.
EU data privacy regulations make it basically impossible to have a "real" AI; one with a body that can see the world and live-update its memories like a human. Because the AI seeing somebody's face (or a picture of it) and memorising it would be considered a privacy violation. In future this would severely limit the kinds of AI Europeans are allowed to access; only AIs with no vision or no ability to memorise new things would be permitted.
No, it is not.
Every regulation in recent history was written with some corporation to criple their competition. When you codify what your favorable competitor is doing into the law, competition is foced to compete on broken market and cannot possibly inovate.
mistral-large-2407
The concept of "the logic of political survival" is often associated with the work of political scientists Bruce Bueno de Mesquita, Alastair Smith, James D. Morrow, and Randolph M. Siverson. Their book **"The Logic of Political Survival"** presents a theory about how political leaders make decisions based on their primary goal of staying in power.
The central argument of the book is that political leaders prioritize their own survival in office above other considerations, such as the well-being of their citizens or the long-term interests of their country. The authors develop a model to explain how different political institutions—such as democracies, autocracies, and mixed regimes—influence the strategies leaders use to maintain their power.
Key points from **"The Logic of Political Survival"** include:
1. **Selectorate Theory** : The authors introduce the concept of the "selectorate," which refers to the group of people who have a say in choosing the leader, and the "winning coalition," which is the subset of the selectorate whose support is crucial for the leader to remain in power.
2. **Institutional Constraints** : The type of political system (democratic, autocratic, etc.) affects the size of the selectorate and the winning coalition, which in turn influences the leader's behavior and policies.
3. **Policy Choices** : Leaders make policy decisions based on what will keep their winning coalition satisfied and loyal, rather than what is best for the broader population.
4. **Public Goods vs. Private Goods** : In democracies, leaders tend to provide more public goods (e.g., infrastructure, education) to satisfy a larger winning coalition. In autocracies, leaders often provide private goods (e.g., patronage, favors) to a smaller, more influential group.
5. **War and Peace** : The authors also explore how the logic of political survival affects decisions about war and peace, arguing that leaders are more likely to engage in conflict when it serves their political interests.
The book provides a framework for understanding the strategic behavior of political leaders and the institutional factors that shape their decision-making processes. It has been influential in the field of political science and offers insights into the dynamics of power and governance across different types of political systems.
A random academic paper doesn't necessarily reflect the reality in Parliament.
I get the feeling some people think we sit around the table cackling with demonic laughter while sacrificing European businesses and destroying the economy. it's utter delusion.
I know almost nothing about the EU parliament to be fair.
I've seen some empassioned clips from people who appear to be dressed in nothing more than their best pair of pyjamas. You can probably trust people like that to advocate on your behalf.
I only posted this as your post reminded me about a book I once read so thought I'd share.
That is completely incorrect. MEPs can block or amend any legislation, the only limitation they have is that they can't directly \*propose\* legislation.
Recently MEPs have:
- Introduced an exemption for Open Source in the AI act
- Ended data roaming charges within Europe
- Banned harmful chemicals in childrens toys
to name just a few things.
Those regulations are not bad - that's just the Meta narrative (or people who don't know what they're talking about). Meta probably wanted to train (or even trained) on people private and/or personal data without having their consent - and being f..ked like that is not legal in the EU. I've read both GDPR (1) and AI Act (2), and I see nothing in those acts that would prevent releasing AI models trained on public and legally obtained data. All the other big techs vision models can be used in the EU, so it seems it's only Meta that did something shady with this release.
1. [https://eur-lex.europa.eu/eli/reg/2016/679/oj](https://eur-lex.europa.eu/eli/reg/2016/679/oj)
2. [https://eur-lex.europa.eu/eli/reg/2024/1689/oj](https://eur-lex.europa.eu/eli/reg/2024/1689/oj)
They aren't. In fact the AI Act is extremely thoughtful. It's all about consumer protection. It doesn't really restrict research and development. It categorises the various risks (pretty reasonably too) and then expresses what private companies may do when it comes to users and provides mechanisms for assessment of what corporate power is doing.
The GDPR forces corporate power to delete user data on request, under severe penalties. That's a very good thing. The EU dismantles monopoly crap, like forcing Apple to allow other wallets or RCS support.
Regulations ultimately help large corporations over small ones.
Large corporations can afford to find a way to comply with regulations or to lobby to change laws. Also, the regulations are slow to change with technology progress, so slow down adoption of new technologies.
A.good recent example is diesel cars
The EU put in a raft of regulations to favour diesel over petrol and hybrids. This protected their large car manufacturers from hybrids from Japanese companies.
What happened? The large manufacturers found ways around the regulations, and the EU were sloppy on enforcement on their own companies, hence deiselgate was discovered in America.
And due to the protection the EU car companies had, they didn't need to develop hybrids and EVs, and are now behind the global competition.
Also good points. Large corporations can spend a lot of money on their lawyers to circumvent the law and lobbying is a huge problem near everywhere. But I see also a trend that the EU looks more to how companies use their regulations. Like Apple with their app store, EU is not happy how they tried to get around it and let that Apple know. But we need more such strength against such huge corporations who think they can do anything.
It isn't, in many ways they have done more for Americans user rights that the so called free market have ever done.
But huge billion companies hate to do what others say, they're used to control everything including the politicians they bought. So they spin a story and useless lemmings run with a even creates memes with them.
Just this week the same free market no regulations country had the ceo of the biggest medical company lined up for a senator grilling. His company was asking for to much money for his European product.
Let's get a meme with that..
Keep in mind that until P2P AI training tech becomes a thing OR enterprise level GPUs become affordable to the masses, all LLMs are open source according to the whims of those corporations.
If the goal is to make AI accessible to anyone, we have to keep open source models alive either through developing P2P training technology or reliance on corporations (🤮)
Well they are the clever minds which decided that we europeans, had too much time available so why not spend some time clicking stupid cookie warning banners on every goddamn page under the sun, forever. (What about embedding that functionality in the freacking brower so it could even automated a globally configurable!!).
Too much Brussels croissants in the morning I guess.
If it is embedded in the browser and automatically agreed to, it defeats the whole point of it. you’re supposed to be able to see what data the website is collecting and either agree or disagree.
Why I'm supposed to do that? That has to be my choice. So first time the browser runs allow user to decide what they want do to about it, and stablish a policy about it. In the end is annoying and absurd, is much more effective to keep a white list of sites permanently storing cookies and just clean everything else whenever the browser closes.
No that can’t be your choice because allowing that would mean that the legal system will have to change to allow people to consent to things they don’t know unknowingly. That would be an insane standard that would lead to massive issues which i hope are obvious.
So inform people about what's going on, but don't hassle them on every single web page FGS. Inform people when the browser is opened for the first time and then establish a policy, which may be reflected in some icon in the URL bar for example. We are doing that already with SSL certificates. There has to be a better way to keep people informed, than imposing everywhere terrible banners which no one reads neither pays attention.
The main reason the EU has poor entrepreneurship has to do with their bankruptcy laws. Most founders there only get one shot, because when their first startup fails, they can never get out from under the debts again. America's relatively forgiving bankruptcy laws incentivize entrepreneurs to try multiple times (and hint: most don't succeed until multiple tries and they're in their 40s). It's the main factor that disincentivizes entrepreneurship in the EU. Tech laws like GDPR don't hurt EU startups, they help them - giving them a market segment with a degree of external protection when foreign competitors decide to focus on other markets rather than comply.
That's consumer facing. GDPR causes companies to adopt a lot standards and certifications to fit the regulation, which in turn ensures your data is safe and not exploited.
As I've heard European banks tend to not give credits to newly founded LLC without founders having personal liability. And rules for personal bancrupcy are stricter in Europe.
You won't be disqualified from being a director of a company that goes into insolvency. Misconduct, fraud ... sure. You can check the relevant Act: [https://www.legislation.gov.uk/ukpga/1986/46/contents](https://www.legislation.gov.uk/ukpga/1986/46/contents)
The problem is you have to put up collateral at the start, which is quite significant in some countries. Most people can't just throw away 25k to try a random business idea.
A business needs capital to operate, which is what the initial investment is meant to provide. You're also not throwing away the money, as all the invested money goes into running the business (except for a registry fee of around $100)
The minimum investment is also nowhere around $25k, but instead somewhere around $2.5k depending on the country. More often than not, this initial investment is used as equity capital for the business to loan up to $20k, usually from a bank who views your business idea as one with promise
Well for Slovenia it's 7.5k for a d.o.o., for Germany it's 25k for a GmbH. I'm pretty certain about those two figures.
Yes you do get it onto your company account, but will your expenses be that much? The average tech startup is just some guy with a laptop without any real expenses or ways to pay for them anyhow.
Damn Austria, what the fuck.
But interesting I had the notion that it was just France, Netherlands and Belgium that were on the low end, maybe something's changed in recent years. I sure hope this is abolished entirely EU-wide one day or at least reduced to a few hundred, but at the moment this is the core reason why the sole proprietor entity is so popular in the half of countries that have this stupendously high cost of entry... and with that one you unfortunately remain liable.
I can't speak for other countries, but in Norway we raised the price from USD $750 to USD $2500 beacause because A LOT of shady people started estabilishing companies with the intent of either scamming people or by taking too much of a risk with other people's money (loans), intentionally running the companies into the ground after having cashed out.
USD $2500 became the level where it was too expensive for scammers to take that risk, while it was still not too expensive for a regular citizen to start their own company if they saved for a few months.
Depending on the economy of the country, I say $2k-3k is the perfect level of entry for most western countries, while also avoiding scammers running rampant
Yeah idk, people still manage to do exactly that here in Slovenia despite the 7.5k barrier, I guess the scams are quite profitable or enforcement is too lax. Probably both.
I'd much rather have stricter laws in case of fraud than pricing everybody out.
hmm idk about bankruptcy laws as but lack of investment capital and also a fractured market (language, regulations, etc) are definitely a reason. At least those are the things that impact me personally
The investment capital comes directly and indirectly from debt. And the US one, both public and private, is ominous. So.... gotta larp as entrepreneurs and VCs while the circus is going on.
I think it's more of a lack of any VC firms to support those startups and accelerators are kinda shit. LLCs do generally absolve you from debt, but making one in say, Germany costs like 25k EUR (iirc) for starting capital as collateral so you lose at least that much. In most other countries it's less but still in the 5-15k range typically, except a few.
https://preview.redd.it/z9cyoawfg4rd1.png?width=600&format=png&auto=webp&s=a08c688bf6dd7442f0815ee80cb4029cefc536df
It's not the bankruptcy but capital.
No one wants to invest in highly regulated countries. Startups should be able to fire poor performers so that they can pivot or accelerate. One bad hire can bring down startups (because they are usually > 10% of the firm).
Startups would typically prefer hungrier workers. Not entitled, WLB seekers
I was thinking of moving to the EU for a job, but between the lack of first amendment protection and this kind of shit I don't want to be there when things change. They're going to struggle hard and be playing catch up in a bad way.
Actually all the globe is going into AI regulation. Each region with its own degree.
I recently attended a lesson where this was the topic. At one point professor said something which fits a lot:
One invent, One copy, One regulate.
Guess who's who..
GDPR is still horrible for small website owners who have no profit in mind. They need to put their private address and phone number on their imprint so everyone at the whole internet could see where your house lives. So much for private data protection, what a joke!
Yeah, we have people like you to thank for this crap. As if there was no other way to hold the website owner responsible without directly wanting his private address etc. Why not my bank account number etc.?
Even before the GDPR, there was an imprint obligation and anyone who adhered to it and took care of their website was always reachable if something should happen. I had my first website back in 1998 and have never had any problems with accessibility from my site since then. Apart from the fact that in over 25 years I have never had a case where someone had to reach me urgently or had something wrong with my website. But in the unlikely event that something might happen, you have to publish your private address 24/7/365 for everyone to see, which anyone who wants to can misuse. I don't even want to know which data traders now have this address where I've lived for over 20 years. And there are absolutely no weirdos who would think of “visiting” someone.
There are other ways as that for a solution and that is my point. On one side "safe our data" on the other side "put your private address out to the whole world".
Yeah, well piece after centureies of wars is not that bad either. Shengen is great, right of working everywhere as well. I love euro also. So there is that.
Euro is problematic. Shengen and Right of Work yes I agree as have used them on my move to UK 27 years ago.
My only gripe is the cars. Car registered in Germany needs to be reregistered in Greece (and pay heavy taxes) and if you move to Netherlands, have to pay taxes there to re-register it and back again to Germany.
Not having a unified car registration and pay once all the fees is just there for the money grabbing. I have no idea how USA works in that part, but cannot have only 1/3 of common stuff and everything else been exploited for money.
I will give you another example. Look at BWM. EU should have regulated to force the company build factories in the periphery of the EU. But they have done? Allowed them to be built in Turkey, a country not in the EU. Like USA car manufacturers went to Mexico.
Define "integration"?
What is the point of Right to Work when you cannot move your car with you?
What is the point of Euro when you have countries completely outright banning all foreign EU banks so the few oligarchs can control all the flow of money?
Wrong. GDPR is great tool to force big corps to show what they have of me, and wipe it out of their system. And if they sold my data, lawsuits and compensations are in place.
I thought GDPR would be a good thing (UK). The 'right to forget' and all that. Felt empowering, should I ever need to use it.
I did a credit check on myself the other day, via Experian, to find I have a CCJ that belongs to someone else on my fucking credit record.
Three emails to Experian and long story short, they absolutely do not give a fuck.
GDPR does not appear to be a useful stick to beat them with.
Report them to the authorities and say you expect answers within the set time periods the law stipulates. Continue a written record everytime they breach it and make sure to write that a non-reply on your messages contistitute another breach.
Document and send with your report to the authorities
Funny that OP believes regulation is blocking progress - when large tech companies like Facebook essentially want to get your data entirely for free to build products with which they make tons of money.
What have you been smoking and where can I get it?
Or do you really find such conspiracy theories somehow logical? It doesn't even make technical sense what you're saying.
Before you assume it’s wrong look it up chat control. https://www.patrick-breyer.de/en/chat-control-2-0-eu-governments-set-to-approve-the-end-of-private-messaging-and-secure-encryption/
I know about the chat control. And I know that this has been going on for years without an agreement and dozens of mitigations. It hasn't been decided, nor is there a majority for it, nor does the eu want to control your thoughts, wtf.
People show use models as they please and they want control what does or doesn’t say. Also was being hyperbolic about thought control. But the decryption part is only a matter of time IMO.
Laugh in MistralAI, ESA, Medical research, CERN, AirBus, Dassault System...
Just because we don't let US companies do whatever they want doesn't mean we don't have tech companies.
We don't sell computers. That's it.
Companies sell my medication to the pharmacy at a higher price in Canada than they do in France (I checked the prices). I'm pretty sure they sell them at even higher prices in the US.
Most countries have national bodies that negotiate with pharmaceutical companies and agree on prices for important medicines not just the ones you get through healthcare, but what anyone in a pharmacy will pay. Not everywhere and not for all medicines, but generally they have predictable and regulated prices, introducing risks like medical re-export from a country that negotiated lower prices to another with higher ones. None of the producers went bankrupt, so regulation works for both consumers and vendors, with some challenges, which are insignificant compared to the US problems in this regard.
Cat has been out of the bag when it comes to the information services and it’s not going back into the bag. more legislation around it will only make it harder for common folk.
Crazy how the EU just hands its best and brightest minds to the West/Asia and is proud of it... in the name of "regulation" or "safety" or "equality" or whatever it is.
Other companies can release multimodal models perfectly fine in the EU. Why can’t Meta? I haven’t found the reason they withhold the models apart from vague speculation so I‘m guessing there‘s more to it than they want to admit.
Without Llama, it's not unlikely that there would be no large open-weight models at all. No Qwen, no Mistral, no Gemma even, as everything that's come out since Llama has been more or less a response to Meta deciding to invest so heavily in open AI (not to be confused with OpenAI, which is somehow the opposite). But this was only possible at the time because politicians weren't paying attention. The moral panic hadn't set in yet. There weren't easy points to score by banging your fist against the table and shouting, "something's got to be done!"
And so here we are now, looking anywhere but Europe (and apparently California) for the next big development. Which is coming, make no mistake. It just won't come from Europe. China is surging ahead. Hell, I wouldn't be surprised if this is how Russia ends up becoming economically relevant again.
As an EU citizen, you just fell for Meta's bullshit of playing politics with such messages without any real consequences. Everyone in the EU will still be able to download legally. Meta just wants to create a mood against regulations that would otherwise affect them elsewhere.
Don't fall for such nonsense. There is no EU Ai regulation that would ban LLama3.2 per se.
the regulation could have been clearer and more effective: no doubt. but not aiming at regulation is not a clever idea either.
once society gets the hang of AI business protecting IPs will be all the rage. all that matters then is to be able to control of access rights and regulation that can enforce those. at a certain point there will be no future AI-business / commercial AI use without proper proof of copyright access and established terms of usage. also you will have to proof: a bill of software, a bill of data, and a bill of AI, and a rigorous public record of emissions and other related compliance.
Its not just regulation on AI - its the fucking dark ages guild system being enforced by most countries in the EU that makes it almost impossible to open a company, especially if it isnt easily classifiable by god damn medieval parameters.
Fundamentally I think we should be skeptical about the tech companies who do not launch models in the EU citing "regulatory issues", because none of them are being precise about what their issues are with the regulations as-is. One of the things that could be critiqued about the regulations I think is that certain functions are prohibited not only for deployers (the people using a general purpose model), but also on the level of the model itself. Basically any LLM \*could\* be used for social scoring for example, but that doesn't/shouldn't make all LLMs prohibited... however I am skeptical that that's the real reason why tech companies are holding back on some but not all of their models.
As Right wing anarchist I am against gov all together.
However that doesn't mean a company can use my personal data that harvested or bought to train LLMs, there has to be regulation to prohibit that.
If we de-propagandized "EU is anti AI", it actually means "EU is the only place that puts it's inhabitants rights to their personal data before the profits of giga-corporations". Meta/Facebook is spearheading this desinformation campaign.
I see no loss for EU . These companies should be penalised to the end for the user data they have harvested and misused inspite of whatever zuck's redemption arc is...phew
Llama 3.2 might not be available yet, but it’s wild how the EU’s regulations are shaping things! For those who need access to models like Llama 3.1, Qwen, or other open-source LLMs without the hassle, I’ve been using Hyperbolic to get the best tools at a fraction of the cost.
You can check out what they offer here: https://app.hyperbolic.xyz/models. Worth a look!
That’s is the problem of not being full capitalistic in a capitalist dominated world, where you always have to be compétitive and get more money, or you lose the (rate)race. I prefer the regulation tho
Meta: we love open source.
Proceed to ban 27 countries in the license of the vision models because I imagine they regulate the usage of user data in the training dataset, Meta doesn't like that.
I thought the main reason we’re all here is to regulate the AI ourselves by running it locally? And yes it is a bit harder for data selling big tech to thrive in regulated environments.
grow up you moron. ai is cool, but all potentially wildly dangerous, it absolutely NEEDS to be regulated in some way. it's not about control, get a fucking grip.
531 Comments
fazkan@reddit
mpasila@reddit
Fabulous-Ad-6061@reddit
AirconWater@reddit
Fabulous-Ad-6061@reddit
satireplusplus@reddit
Chongo4684@reddit
Atupis@reddit
jman6495@reddit
hanjh@reddit
jman6495@reddit
AirconWater@reddit
jman6495@reddit
Jamais_Vu206@reddit
Koalateka@reddit
AirconWater@reddit
jman6495@reddit
Jamais_Vu206@reddit
jman6495@reddit
Jamais_Vu206@reddit
jman6495@reddit
Jamais_Vu206@reddit
jman6495@reddit
AirconWater@reddit
wildebeest3e@reddit
jman6495@reddit
Useful44723@reddit
jman6495@reddit
Useful44723@reddit
jman6495@reddit
TikiTDO@reddit
jman6495@reddit
TikiTDO@reddit
appenz@reddit
jman6495@reddit
appenz@reddit
jman6495@reddit
appenz@reddit
jman6495@reddit
appenz@reddit
wildebeest3e@reddit
jman6495@reddit
wildebeest3e@reddit
jman6495@reddit
Koalateka@reddit
MoffKalast@reddit
jman6495@reddit
sdmat@reddit
jman6495@reddit
sdmat@reddit
jman6495@reddit
sdmat@reddit
jman6495@reddit
sdmat@reddit
jman6495@reddit
sdmat@reddit
FullOf_Bad_Ideas@reddit
jman6495@reddit
FullOf_Bad_Ideas@reddit
jman6495@reddit
FullOf_Bad_Ideas@reddit
jman6495@reddit
FullOf_Bad_Ideas@reddit
jman6495@reddit
jman6495@reddit
spokale@reddit
jman6495@reddit
spokale@reddit
PoliteCanadian@reddit
jman6495@reddit
appenz@reddit
spokale@reddit
Atupis@reddit
Ptipiak@reddit
AirconWater@reddit
IlIllIlllIlllIllll@reddit
NekoHikari@reddit
fazkan@reddit
Atupis@reddit
AirconWater@reddit
cyan2k@reddit
AirconWater@reddit
Nomad_Red@reddit
jman6495@reddit
DuplexEspresso@reddit
Character-Refuse-255@reddit
molbal@reddit
AirconWater@reddit
phenotype001@reddit
satireplusplus@reddit
moncallikta@reddit
phenotype001@reddit
satireplusplus@reddit
molbal@reddit
bguberfain@reddit
AggressiveDick2233@reddit
molbal@reddit
JohnMcPineapple@reddit
deliadam11@reddit
molbal@reddit
Wonderful-Wasabi-224@reddit
molbal@reddit
BadUsername_Numbers@reddit
GreatBritishHedgehog@reddit
Chongo4684@reddit
physalisx@reddit
Jamais_Vu206@reddit
jman6495@reddit
shroddy@reddit
jman6495@reddit
AirconWater@reddit
shroddy@reddit
jman6495@reddit
novexion@reddit
jman6495@reddit
novexion@reddit
jman6495@reddit
novexion@reddit
jman6495@reddit
novexion@reddit
jman6495@reddit
novexion@reddit
Radiant_Dog1937@reddit
jman6495@reddit
jugalator@reddit
jman6495@reddit
procgen@reddit
jman6495@reddit
procgen@reddit
AirconWater@reddit
jman6495@reddit
---AI---@reddit
jman6495@reddit
---AI---@reddit
jman6495@reddit
antihero-itsme@reddit
AirconWater@reddit
---AI---@reddit
jman6495@reddit
---AI---@reddit
jman6495@reddit
---AI---@reddit
jman6495@reddit
---AI---@reddit
---AI---@reddit
ReturningTarzan@reddit
HighDefinist@reddit
jman6495@reddit
---AI---@reddit
jman6495@reddit
cac2573@reddit
jman6495@reddit
ReturningTarzan@reddit
HighDefinist@reddit
tyoma@reddit
jman6495@reddit
ReturningTarzan@reddit
Meesy-Ice@reddit
jman6495@reddit
Chongo4684@reddit
jman6495@reddit
Chongo4684@reddit
jman6495@reddit
Chongo4684@reddit
GaggiX@reddit
spiritusastrum@reddit
EDLLT@reddit
AirconWater@reddit
Meesy-Ice@reddit
EDLLT@reddit
spiritusastrum@reddit
Bite_It_You_Scum@reddit
williamwalker@reddit
Meesy-Ice@reddit
williamwalker@reddit
pijuskri@reddit
williamwalker@reddit
spiritusastrum@reddit
williamwalker@reddit
alongated@reddit
emprahsFury@reddit
Cuidads@reddit
spiritusastrum@reddit
CortaCircuit@reddit
Chongo4684@reddit
mistergoodfellow78@reddit
ILikeBubblyWater@reddit
IlIllIlllIlllIllll@reddit
ILikeBubblyWater@reddit
IlIllIlllIlllIllll@reddit
NobleKale@reddit
ThomasBudd93@reddit
mrdevlar@reddit
AirconWater@reddit
HighDefinist@reddit
AirconWater@reddit
matteogeniaccio@reddit
redballooon@reddit
matteogeniaccio@reddit
redballooon@reddit
Blizado@reddit
HighDefinist@reddit
Blizado@reddit
HighDefinist@reddit
RedditDiedLongAgo@reddit
Rich_Repeat_22@reddit
goqsane@reddit
Rich_Repeat_22@reddit
Hugi_R@reddit
um-xpto@reddit
Hugi_R@reddit
physalisx@reddit
ThomasBudd93@reddit
matteogeniaccio@reddit
Uhlo@reddit
Brilliant-Elk2404@reddit
keepthepace@reddit
AutomaticDriver5882@reddit
Xauder@reddit
FrermitTheKog@reddit
jman6495@reddit
FrermitTheKog@reddit
jman6495@reddit
FrermitTheKog@reddit
AirconWater@reddit
moncallikta@reddit
FrermitTheKog@reddit
Atupis@reddit
JohnMcPineapple@reddit
Additional_Test_758@reddit
This_Is_The_End@reddit
AirconWater@reddit
PoliteCanadian@reddit
This_Is_The_End@reddit
218-69@reddit
Minute_Attempt3063@reddit
PoliteCanadian@reddit
AirconWater@reddit
fiddly_foodle_bird@reddit
Kiwi_In_Europe@reddit
Xauder@reddit
Minute_Attempt3063@reddit
redoubt515@reddit
Minute_Attempt3063@reddit
Embrace-Mania@reddit
Glad-Way-637@reddit
Minute_Attempt3063@reddit
Snoo44080@reddit
jrcapablanca@reddit
antihero-itsme@reddit
sha256md5@reddit
jman6495@reddit
eposnix@reddit
jman6495@reddit
eposnix@reddit
jman6495@reddit
moncallikta@reddit
jman6495@reddit
moncallikta@reddit
hansfellangelino@reddit
shaman-warrior@reddit
sofixa11@reddit
doorMock@reddit
sofixa11@reddit
Xauder@reddit
qroshan@reddit
shaman-warrior@reddit
Xauder@reddit
VectorD@reddit
Blizado@reddit
SirRece@reddit
Xauder@reddit
SirRece@reddit
Honey_Badger_Actua1@reddit
BalorNG@reddit
ServeAlone7622@reddit
CheatCodesOfLife@reddit
emprahsFury@reddit
HighDefinist@reddit
AirconWater@reddit
keepthepace@reddit
spokale@reddit
AndroidePsicokiller@reddit
AIPornCollector@reddit
_wallSnake_@reddit
AirconWater@reddit
danigoncalves@reddit
Atom_101@reddit
_lindt_@reddit
AIPornCollector@reddit
Mart-McUH@reddit
_lindt_@reddit
1Soundwave3@reddit
Upstairs_Tie_7855@reddit
gus_the_polar_bear@reddit
monerobull@reddit
Upstairs_Tie_7855@reddit
monerobull@reddit
Upstairs_Tie_7855@reddit
monerobull@reddit
Upstairs_Tie_7855@reddit
monerobull@reddit
Upstairs_Tie_7855@reddit
monerobull@reddit
Upstairs_Tie_7855@reddit
monerobull@reddit
Upstairs_Tie_7855@reddit
monerobull@reddit
Upstairs_Tie_7855@reddit
monerobull@reddit
Upstairs_Tie_7855@reddit
Responsible-Leg-9205@reddit
pohui@reddit
Upstairs_Tie_7855@reddit
AndroidePsicokiller@reddit
cyberdork@reddit
HighDefinist@reddit
Neither_Service_3821@reddit
cov_id19@reddit
AirconWater@reddit
Different_Cake@reddit
Spirited_Example_341@reddit
brahh85@reddit
Data_drifting@reddit
procgen@reddit
brahh85@reddit
Math_31416@reddit
procgen@reddit
brahh85@reddit
procgen@reddit
Rich_Repeat_22@reddit
Deep-Ad-4991@reddit
PopPsychological4106@reddit
thewayofthewu@reddit
Soggy-Camera1270@reddit
thewayofthewu@reddit
Soggy-Camera1270@reddit
nikitastaf1996@reddit
marknutter@reddit
Poromenos@reddit
marknutter@reddit
Poromenos@reddit
marknutter@reddit
Poromenos@reddit
marknutter@reddit
Poromenos@reddit
marknutter@reddit
Poromenos@reddit
Poromenos@reddit
HighDefinist@reddit
Poromenos@reddit
Chongo4684@reddit
MrZoraman@reddit
HighDefinist@reddit
HighDefinist@reddit
neat_shinobi@reddit
cac2573@reddit
CondiMesmer@reddit
Chongo4684@reddit
Rich_Repeat_22@reddit
ReturningTarzan@reddit
logicchains@reddit
AssistBorn4589@reddit
jman6495@reddit
Additional_Test_758@reddit
jman6495@reddit
Additional_Test_758@reddit
AssistBorn4589@reddit
jman6495@reddit
MoonRide303@reddit
TitularClergy@reddit
Spursdy@reddit
Blizado@reddit
jman6495@reddit
FuckKarmeWhores@reddit
TheSilverSmith47@reddit
Exotic_Illustrator95@reddit
Meesy-Ice@reddit
Exotic_Illustrator95@reddit
Meesy-Ice@reddit
Exotic_Illustrator95@reddit
Hmz_786@reddit
robogame_dev@reddit
roofgram@reddit
dethorin@reddit
genshiryoku@reddit
Any_Pressure4251@reddit
dethorin@reddit
OYTIS_OYTINWN@reddit
_supert_@reddit
Amblyopius@reddit
_supert_@reddit
314kabinet@reddit
_supert_@reddit
Severin_Suveren@reddit
MoffKalast@reddit
Severin_Suveren@reddit
MoffKalast@reddit
Severin_Suveren@reddit
MoffKalast@reddit
Severin_Suveren@reddit
MoffKalast@reddit
KingGongzilla@reddit
rini17@reddit
And-Bee@reddit
dydhaw@reddit
MoffKalast@reddit
I_AM_BUDE@reddit
qroshan@reddit
FarVision5@reddit
Many-Addendum-4263@reddit
lazywiing@reddit
AwesomeDragon97@reddit
Tellesus@reddit
soboa2@reddit
LuganBlan@reddit
MahmoudAI@reddit
ziphnor@reddit
Gunner_McCloud@reddit
ambient_temp_xeno@reddit
ziphnor@reddit
ambient_temp_xeno@reddit
ziphnor@reddit
ambient_temp_xeno@reddit
ziphnor@reddit
Blizado@reddit
Meesy-Ice@reddit
Blizado@reddit
TitularClergy@reddit
AssistBorn4589@reddit
virtualghost@reddit
Rich_Repeat_22@reddit
cha_ppmn@reddit
Rich_Repeat_22@reddit
cha_ppmn@reddit
Rich_Repeat_22@reddit
Rich_Repeat_22@reddit
poetry-linesman@reddit
bigmanbananas@reddit
Rich_Repeat_22@reddit
AssistBorn4589@reddit
ziphnor@reddit
AssistBorn4589@reddit
Additional_Test_758@reddit
mloDK@reddit
CheatCodesOfLife@reddit
AnyAsparagus988@reddit
Mplus479@reddit
AnyAsparagus988@reddit
Mplus479@reddit
Meesy-Ice@reddit
Nikifemboy18@reddit
astralkoi@reddit
fabkosta@reddit
AutomaticDriver5882@reddit
Low-Boysenberry1173@reddit
AutomaticDriver5882@reddit
Low-Boysenberry1173@reddit
AutomaticDriver5882@reddit
Low-Boysenberry1173@reddit
AutomaticDriver5882@reddit
vwildest@reddit
SrData@reddit
Hopeful-Site1162@reddit
oneharmlesskitty@reddit
__some__guy@reddit
TechnoHenry@reddit
oneharmlesskitty@reddit
Optimal_Leg638@reddit
Captain_Pumpkinhead@reddit
Robswc@reddit
Puzzleheaded-Lab-635@reddit
Adrian_F@reddit
goodatburningtoast@reddit
Queasy-Board390@reddit
ReturningTarzan@reddit
Low-Boysenberry1173@reddit
Revolutionary_Ad6574@reddit
Low-Boysenberry1173@reddit
B3lthazar@reddit
dontpushbutpull@reddit
On-The-Red-Team@reddit
v202099@reddit
ThrowAwayAlyro@reddit
EastSignificance9744@reddit
Rich_Repeat_22@reddit
Due-Listen2632@reddit
Pkittens@reddit
hansfellangelino@reddit
sigiel@reddit
Sure_Display9664@reddit
sigiel@reddit
Sure_Display9664@reddit
dydhaw@reddit
TitularClergy@reddit
BroomBroomMmmmm@reddit
Prestigious-Two8928@reddit
AlarmingAffect0@reddit
herozorro@reddit
Wrong_User_Logged@reddit (OP)
herozorro@reddit
doyouevenliff@reddit
Shnuksy@reddit
JellyFluffGames@reddit
fets-12345c@reddit
ZmeuraPi@reddit
Defiant_Magazine_138@reddit
Lost_County_3790@reddit
dahara111@reddit
snozburger@reddit
Deluded-1b-gguf@reddit
MrWeirdoFace@reddit
ServeAlone7622@reddit
aLong2016@reddit
Massive_Robot_Cactus@reddit
GaggiX@reddit
Comfortable-Slip4716@reddit
fixtwin@reddit
Inaeipathy@reddit
Rich_Repeat_22@reddit
bick_nyers@reddit
Rich_Repeat_22@reddit
NancyPelosisRedCoat@reddit
Zealousideal_Map3806@reddit
when_the_soda-dry@reddit
goj1ra@reddit
RandumbRedditor1000@reddit
NancyPelosisRedCoat@reddit
Nikhil_Vajra@reddit
ObjectiveBrief6838@reddit
AlanzhuLy@reddit