Is this any surprise? Of course OpenAI now wants to maintain the status quo, so while they do want regulation to make it more difficult for new players to enter the market, they don’t want regulation to hinder their own business model. So they’re lobbying for exactly that.
They’re a for-profit business doing what for-profit businesses do. All I hope is that the politicians make a sensible choice, and find a balance between innovation and regulation.
They've been saying "hmm, this could be dangerous" since GPT-2.
They've also said that if they can't do business under some new regulation, they will stop doing business.
These do not seem like the actions of a sneaky sneaky double-bluff, but who knows — perhaps Sam Altman has automated himself away with an unannounced version of GPT-5.
It's not a surprise, but the hypocrisy is unfathomable. The same company who has kept nearly everything about GPT closed source since v2 because they said "it would be too dangerous to release" now says that an even more powerful version of the same model is not supposed to be classified as dangerous, because... reasons. And the EU legislators bought it. Just shows again that money is the only real policymaker in the world.
Not only that, but they also kept saying that governments around the world need to be quick at regulating AI usage. Many people pointed out how that could have been just a way to have some free advertising, this is the proof.
I have no reason to think 200 nations nationalising AI are going to be any safer than 200 companies, even if the companies are the hubristic archetype beloved by Hollywood scripts.
PopeGPT debating آیت الله-GPT about theology could be interesting; but do you want to wake up to the news that советский-GPT is running a successful propaganda war in the USA and now the American Communist Party is likely to win the next presidential election? That ChatKimJongUn is successfully boosting the industrial capacity of the North despite sanctions?
That's why international agreements are also relevant, even if you don't share Yudkowsky's fear of alignment failures paper-clipping everyone.
This article is about how they convinced EU that GPT is not "high risk AI".
But the very idea of "high risk AI" in general is great for ClosedAI business because any of their competitors would have to convince EU their LLM or whatever is not "high risk" too. And no other competitor has billions of Microsoft's USD to spend on lobbying lawmakers.
You are downvoted but I agree with you. HN was pretty convinced about that and, as it turns out, wrong. (Or, maybe partially wrong because OpenAI probably wants more regulation that helps them and less regulation that hinders them.)
They submitted a paper to congress talking about how they should regulate AI.
HN is not wrong about the fact that there were trying to push regulation in the United States that would help their business. It just turns out the European laws will actually hurt their business.
They're not mutually exclusive. Ideally OpenAI wants regulation that will put them ahead of others, which means:
* set the bar high enough that competition needs to put more effort (or maybe is entirely prevented from entering the market)
* set the bar low enough that OpenAI's business model is not affected (e.g. does not require them to spend significant amount of time and resources on complying with the regulation)
So in US they can lobby for more rules while in EU for fewer restrictions.
Not mutually exclusive. Europe can get pretty crazy with their takes. Outright bans on things because they disrupt the status quo is the norm. After the rest of the world adapts and outcompetes Europe, then they remove the ban, and international companies flood the market to mop up any potential profit.
In fact, corruption is greatly reduced by techniques such as democracy, a free press, transparency, etc. There's plenty of research to show it.
It's like saying we'll never be rid of cancer; that's true, but the number of cancer deaths is greatly reduced.
The choice isn't binary, 0% corrupt or 100% corrupt. The issue is where we fall on that scale. You are an unwitting agent of the corrupt by repeating the trope that we are powerless.
We need regulation of AI because something on the Internet could confuse or mislead someone? Not every potential problem needs a law to protect us. What things is AI doing that existing laws against fraud aren't protecting us from?
Most things become a "different" problem (from legal enforcement perspective) when the scale changes. Without entering in the merit of whether laws in question are good/bad, it's understandable to have a different legal regime at larger scale, such as harsher punishment above certain thresholds.
And AI-generated content changes the scale of many many things you can do.
> ask people involved in the field, and risk conflicts of interests
True, but in this case you can greatly mitigate the risk of conflict of interest by asking researchers that aren't directly involved with companies making money with AI. There are many of those researchers working in universities, make sure to filter out those that also work for some AI company and you still have a long list of possible candidates for expert consultancy.
Do you think researchers that really have 0 conflicts or interest with other researchers working in those companies really are up-to-date as to what the real capabilities of NN will be in the coming years ? We all know there's a huge gap between what researchers think is possible in theory and what's actually released as a working product in the industry.
My intuition is nobody saw chatGPT coming, and only if you were inside openAI a few months before the release could you have anticipated the impact.
I think it's exactly the opposite: researchers are currently studying models that are better than ChatGPT. The reason why when we use one of the open source models those tend to feel worse than ChatGPT is twofold:
1. Those models are usually optimized to run on an average PC, at the cost of accuracy
2. The researchers behind it don't have the kind of money that OpenAI has for training, but this doesn't mean that the model itself is less advanced than ChatGPT.
I also suggest to read a somewhat famous Google memo that leaked a few months ago [1].
that's why i was mentioning the gap between better in theory and in practice. My feeling is that the transition between something that looks fantastic in a lab, and something that can actually be used as a product, like chatGPT is pretty hard.
But this has nothing to do with the quality of a consultancy from university researchers. They know very well how the models work and the direction the research is taking, they not having the same amount of money available for training is irrelevant w.r.t. their ability to provide advice and guidance in policy making.
And even if I was wrong about this, the link I provided in the previous comment explains how the money required to effectively train those models is rapidly dropping, making the advantage of companies like OpenAI smaller and smaller.
Reading through the white paper leaves me feeling that some commenters here may have had an axe to grind already. The positions raised by OpenAI seemed to be focused on tailoring overly-broad language, and curtailing phrasing which confused possibility of misuse for intent to misuse.
The ai regulations have the possibility of becoming a GDPR-like boondoggle in the absence of domain-expertise from people in the field.
Researchers without a conflict of interest should also be heavily be engaged in this dialog as well, and the law specifically made flexible enough to adapt in the face of rapid change.
It makes the case for local models even more urgent.
Between ClosedAI not giving me GPT4 API access, and how effective it is at medical diagnosis(The AMA is going to ban it after it kills 1 person, just watch); we need local models.
So Sam Altman gets in front of governments around the world, alarms everyone with science fiction about how he’s created the technology that might be the doom of the human race, asks for regulation, now doesn’t want regulation even though he’s stirred up a hornets nest of government panic. There’s leadership for you.
(Edit remove criticism of Altman too negative)
I suppose I should be careful saying stuff about zuckerberg like that. He might want to cage fight me.
That's not how companies usually lobby for regulations. Much more often, they get regulation that is full of fangs, laws and rules which made specifically for them and that they can navigate with their huge legal departments, but which would deter any smaller competitors.
So all of this "end of civilisation" talk from OpenAI and other AI company leaders is nothing more than a competitive play to secure their position in the space.
Yes, ignore the fact that many of these people have been talking about this risk since long before they had the companies! Since sometimes companies lobby for regulation to create barriers to entry and carveouts, any public statements about AI risk must simply be lies to hobble competition!
Self-interest is important to note, but self-interest doesn't automatically mean they're disingenuous. Having been worried about alignment for many years prior to working on an AI startup is an example of great evidence that they're actually worried rather than merely incentivized.
What I've realised is at a certain point companies grow to the size where for senior management to comptently do their job to the best of their abilities for what will benefit their company the most they need to reduce the amount they care about others.
This isn't meant to be a dig at them or say they're evil or whatnot. It just seems to be true. For example, a company I worked forced through a migration project that was just constantly going on and on. It was paralyzing the company so to speak. They couldn't do marketing drives, improve the product, make new deals, fullfill requests from clients, etc. Forcing through a half done migration obiviously screws pretty much all the folk who have to work on things. However, the company needed it done. The parent company wanted to sell and couldn't sell a company where the product is basically still in development with a big 18-month overdue project (it was scheduled for 3-months, whoever told anyone that was possible should have been fired). For the benefit of the company forcing through the migration made sense. They could start fullfilling client requests, sell the company, etc. But for the average worker, it meant a large amount of extra work with lots of things no longer working at all. One of the product managers told an account manager she would have to send invoices manually to 500 companies. Invoicing was broken. That's how broken this application was. But for the company, caring that you're overworking employees and increasing stress doesn't make sense in the short term. So from the employers point of view, forcing it through is what was needed. If they didn't they would have been fired. Fundamentally, they did the job they were paid for. To the detriment of the well being of others.
And I think the same is true here. For Altman to do what benefits OpenAI, he has to do things that are to the detriment of everyone else. The Regulations that the EU want to put in place will almost certainly in the future be seen as too weak. And they probably all make sense and Altman probably knows they do. But it's harmful for the company and he has a job to do. For him to be good at his job, he has to ingore what is benefical to everyone else and only care about what is benefical to the company.
I generally agree (although I'm not sure if the example really illustrates it enough to justify it's length). However, this is an incomplete or misleading view on three counts.
First, it sounds like you're justifyimg immortal behavior, which I think wasn't your intention, but the wording isn't so clear on that. The same argument can be given (and would apply) to saying a concentration camp guard has to do what they do in order to "be good at the job". That is a relevant explanation of an important dynamic at play. It doesn't make it OK.
Second, being a scumbag has a cost (e.g. reputation) even from the cold-hearted and immoral poit of view of disembodied money. Society controls this calculation and is not powerless against it.
Third, disembodied money has no agency. Concretely, the mechanism is a manifestation of diffusion of (moral) responsibility. The board and shareholders are not absolved from moral responsibility by some divine right to make maximum profit on their investment. Neither are they justified in ignoring the responsibility they have by diffusing it, neither by choice nor carelessness and willful ignorance.
> the company pushed back against a proposed amendment to the AI Act that would have classified generative AI systems such as ChatGPT and Dall-E as “high risk” if they generated text or imagery that could “falsely appear to a person to be human generated and authentic.” [...] The company argued that it would be sufficient to instead rely on another part of the Act, that mandates AI providers sufficiently label AI-generated content and be clear to users that they are interacting with an AI system.
This sounds pretty reasonable? I don't think it's hypocritical to be talking about the doom of humanity and also arguing that GPT-3, a 3-year old model, should not be classified as "high-risk" in that sense.
Even if you disagree, questioning Altman's leadership and calling him an "empty soul" over this kind of regulatory detail is not adding substance to the discussion imo.
Given his past and present moves ( not what he says publicly ), calling him an "empty soul" are pretty kind words.
Also, the playing dumb card and the "I'm just a tech bro full of innocent dreams" story has already been done to death by the bros from the previous cycle ( think Zuckerberg and his peers ).
It's a scumbag move that most of his peers actually do too and it should at least be publicized and called out.
When Altman says high risk, he means only Microsoft should be allowed to run an AI in case the plot of Terminator happens.
When the EU say high risk, they mean that an AI pacemaker should be explainable enough that you can guarantee it won't randomly kill people. They also mean that low risk applications such as AI holiday recommendation or fiction writing should be more or less unregulated.
It’s an open secret that Elon is running around trying to get a foundational shop off the ground and it doesn’t take rocket scientist to understand that the chaos monkey won’t be paying any alignment tax
Given those dynamics it’s entirely obvious why Sam may want to force a level playing field - they can only lose against both open source and WhiteSupremacistDogwhistle GPT
I don't know if any criticism can be "too negative". All of these SV "luminaries" are on an ego trip and willing to use functioning global civilization as collateral.
Threads like these make me realize how little foresight there is into the genuine creation of Frankenstein's Monster Situation.
Because you never learn the Monsters Name, nor was the Monster a Monster. Just Poof Here he is and there is a Mob with Torches. Rather than being seen in wonder and possibilities.
Not to mention at what Scale can we see one of these models in this Light? As Life enough for a Mob and Torches. Or how soon that will be. And who wants to be Frankenstein? As they remember his Name as the Monster's.
SV leaders often aim for disruption. One hypothesis is that Altman is maximizing OpenAI's disruption with the publicity campaign.
There is also the megalomania trend. As Meredith Whittaker says: "It's a significant thing to cast yourself as the creator of an entity that could be more powerful than human beings,"
I don't think "disruption" in the sense of "disruptive innovation" actually is synonymous with just causing chaos, and the latter doesn't seem like a very coherent theory of Altman's motives.
Wikipedia: "disruptive innovation is innovation that creates a new market and value network or enters at the bottom of an existing market and eventually displaces established market-leading firms, products, and alliances"
ChatGPT is at least arguably disruptive innovation, but this sense of disruption is not something you can maximize or minimize with a publicity campaign.
And by "coherent", I mean a theory of Altman's motives that is consistent with at least some of his incentives. I don't see what he would get out of generally causing chaos.
76 comments
[ 6.2 ms ] story [ 83.9 ms ] threadThey’re a for-profit business doing what for-profit businesses do. All I hope is that the politicians make a sensible choice, and find a balance between innovation and regulation.
They've also said that if they can't do business under some new regulation, they will stop doing business.
These do not seem like the actions of a sneaky sneaky double-bluff, but who knows — perhaps Sam Altman has automated himself away with an unannounced version of GPT-5.
PopeGPT debating آیت الله-GPT about theology could be interesting; but do you want to wake up to the news that советский-GPT is running a successful propaganda war in the USA and now the American Communist Party is likely to win the next presidential election? That ChatKimJongUn is successfully boosting the industrial capacity of the North despite sanctions?
That's why international agreements are also relevant, even if you don't share Yudkowsky's fear of alignment failures paper-clipping everyone.
If we’re following that theory and playing devils advocate
But the very idea of "high risk AI" in general is great for ClosedAI business because any of their competitors would have to convince EU their LLM or whatever is not "high risk" too. And no other competitor has billions of Microsoft's USD to spend on lobbying lawmakers.
HN is not wrong about the fact that there were trying to push regulation in the United States that would help their business. It just turns out the European laws will actually hurt their business.
* set the bar high enough that competition needs to put more effort (or maybe is entirely prevented from entering the market)
* set the bar low enough that OpenAI's business model is not affected (e.g. does not require them to spend significant amount of time and resources on complying with the regulation)
So in US they can lobby for more rules while in EU for fewer restrictions.
Nothing has ever nor will ever remove the corruption from big government. It has been demonstrably present for 100s of years.
It's like saying we'll never be rid of cancer; that's true, but the number of cancer deaths is greatly reduced.
The choice isn't binary, 0% corrupt or 100% corrupt. The issue is where we fall on that scale. You are an unwitting agent of the corrupt by repeating the trope that we are powerless.
Ancient Greece had democracy and ended up with an oligarchy, that's what happens.
Just because something happened thousands of years ago doesn't mean it will happen now.
Have some voice generation, phone data from millions of people, data on who's related, a scam trained AI and some crypto wallets and let it run wild.
- ask people not involved in the field, and risk not understanding enough
- ask people involved in the field, and risk conflicts of interests
- don’t do it, and be blamed by future generations when a disaster happen.
True, but in this case you can greatly mitigate the risk of conflict of interest by asking researchers that aren't directly involved with companies making money with AI. There are many of those researchers working in universities, make sure to filter out those that also work for some AI company and you still have a long list of possible candidates for expert consultancy.
My intuition is nobody saw chatGPT coming, and only if you were inside openAI a few months before the release could you have anticipated the impact.
1. Those models are usually optimized to run on an average PC, at the cost of accuracy
2. The researchers behind it don't have the kind of money that OpenAI has for training, but this doesn't mean that the model itself is less advanced than ChatGPT.
I also suggest to read a somewhat famous Google memo that leaked a few months ago [1].
[1] https://www.semianalysis.com/p/google-we-have-no-moat-and-ne...
And even if I was wrong about this, the link I provided in the previous comment explains how the money required to effectively train those models is rapidly dropping, making the advantage of companies like OpenAI smaller and smaller.
The ai regulations have the possibility of becoming a GDPR-like boondoggle in the absence of domain-expertise from people in the field.
Researchers without a conflict of interest should also be heavily be engaged in this dialog as well, and the law specifically made flexible enough to adapt in the face of rapid change.
Sure they have a great product which is better than most competitors but now they try and use their new money to prohibit others of making progress.
I really hope regulators doesn't get fooled by this old trick. Fuck OpenAI.
Between ClosedAI not giving me GPT4 API access, and how effective it is at medical diagnosis(The AMA is going to ban it after it kills 1 person, just watch); we need local models.
(Edit remove criticism of Altman too negative)
I suppose I should be careful saying stuff about zuckerberg like that. He might want to cage fight me.
Yup, and this was of course the whole point of getting up on a soap box and "calling for" regulation in the first place.
To make sure that at the end of the day, this "regulation" was thoroughly defanged, declawed, and ultimately -- pliable to the company's interest.
So all of this "end of civilisation" talk from OpenAI and other AI company leaders is nothing more than a competitive play to secure their position in the space.
This isn't meant to be a dig at them or say they're evil or whatnot. It just seems to be true. For example, a company I worked forced through a migration project that was just constantly going on and on. It was paralyzing the company so to speak. They couldn't do marketing drives, improve the product, make new deals, fullfill requests from clients, etc. Forcing through a half done migration obiviously screws pretty much all the folk who have to work on things. However, the company needed it done. The parent company wanted to sell and couldn't sell a company where the product is basically still in development with a big 18-month overdue project (it was scheduled for 3-months, whoever told anyone that was possible should have been fired). For the benefit of the company forcing through the migration made sense. They could start fullfilling client requests, sell the company, etc. But for the average worker, it meant a large amount of extra work with lots of things no longer working at all. One of the product managers told an account manager she would have to send invoices manually to 500 companies. Invoicing was broken. That's how broken this application was. But for the company, caring that you're overworking employees and increasing stress doesn't make sense in the short term. So from the employers point of view, forcing it through is what was needed. If they didn't they would have been fired. Fundamentally, they did the job they were paid for. To the detriment of the well being of others.
And I think the same is true here. For Altman to do what benefits OpenAI, he has to do things that are to the detriment of everyone else. The Regulations that the EU want to put in place will almost certainly in the future be seen as too weak. And they probably all make sense and Altman probably knows they do. But it's harmful for the company and he has a job to do. For him to be good at his job, he has to ingore what is benefical to everyone else and only care about what is benefical to the company.
First, it sounds like you're justifyimg immortal behavior, which I think wasn't your intention, but the wording isn't so clear on that. The same argument can be given (and would apply) to saying a concentration camp guard has to do what they do in order to "be good at the job". That is a relevant explanation of an important dynamic at play. It doesn't make it OK.
Second, being a scumbag has a cost (e.g. reputation) even from the cold-hearted and immoral poit of view of disembodied money. Society controls this calculation and is not powerless against it.
Third, disembodied money has no agency. Concretely, the mechanism is a manifestation of diffusion of (moral) responsibility. The board and shareholders are not absolved from moral responsibility by some divine right to make maximum profit on their investment. Neither are they justified in ignoring the responsibility they have by diffusing it, neither by choice nor carelessness and willful ignorance.
This sounds pretty reasonable? I don't think it's hypocritical to be talking about the doom of humanity and also arguing that GPT-3, a 3-year old model, should not be classified as "high-risk" in that sense.
Even if you disagree, questioning Altman's leadership and calling him an "empty soul" over this kind of regulatory detail is not adding substance to the discussion imo.
Scary to think how much power they have choosing winners and losers. I never got GPT4 API access.
Until the local models get up to speed, we are at the whim of this company deciding who wins and who loses.
Also, the playing dumb card and the "I'm just a tech bro full of innocent dreams" story has already been done to death by the bros from the previous cycle ( think Zuckerberg and his peers ).
It's a scumbag move that most of his peers actually do too and it should at least be publicized and called out.
Kind of like the usual "I'm not an asshole, I have Aspergers.."
Sam Altman is none of these things.
When the EU say high risk, they mean that an AI pacemaker should be explainable enough that you can guarantee it won't randomly kill people. They also mean that low risk applications such as AI holiday recommendation or fiction writing should be more or less unregulated.
Which one is reasonable?
They want the law and they want to promise safety while not being impacted by the overbearing regulation they've invited onto the rest of us.
I guess you just take news headlines at face value though.
Given those dynamics it’s entirely obvious why Sam may want to force a level playing field - they can only lose against both open source and WhiteSupremacistDogwhistle GPT
Because you never learn the Monsters Name, nor was the Monster a Monster. Just Poof Here he is and there is a Mob with Torches. Rather than being seen in wonder and possibilities.
Not to mention at what Scale can we see one of these models in this Light? As Life enough for a Mob and Torches. Or how soon that will be. And who wants to be Frankenstein? As they remember his Name as the Monster's.
"You have to buy protection from us" = "You have to lobby to influence our decisions"
Also, these are legislators who are being lobbied, not regulators.
There is also the megalomania trend. As Meredith Whittaker says: "It's a significant thing to cast yourself as the creator of an entity that could be more powerful than human beings,"
https://www.technologyreview.com/2023/06/19/1075140/how-exis...
Also, what do you mean by 'coherent' in this context?
ChatGPT is at least arguably disruptive innovation, but this sense of disruption is not something you can maximize or minimize with a publicity campaign.
And by "coherent", I mean a theory of Altman's motives that is consistent with at least some of his incentives. I don't see what he would get out of generally causing chaos.
https://s3.documentcloud.org/documents/23850240/ares20226851...