I guess what they really want is to limit sale of AI models to compliant vendors and then raise the bar to compliance just high enough so they can pass it but smaller labs can’t. There’s no moat, it’s an efficient market that drives margins to zero right now, of course they don’t want that, collusion of the big vendors is the next logical step.
This could be why they’re going so hard on hacking everyone and everything. You make it clear AI is a threat, get it locked down, and then have the capital to push through the lockdown while others are stuck.
So kind of like how the US figured out how to test and maintain nuclear weapons without an actual explosion, then sought to ban nuclear testing by explosion.
That is generally how regulatory-capture works. As silly as I personally think LLM hype is, the "AI" business posture choices with several Trillion dollars in debt on the books has very few options after hyper-scaling/sandbagging loses traction. =3
People keep saying this, and while it may be partially true, it misses a very important detail. Right now, overall compute is a moat. Someone even commented on another of my comments that one reason Google is lagging is they don't have the same level of Nvidia farms as OpenAI and Anthropic.
A big reason that OpenAI and Anthropic are racing so fast is they both want to get to recursive self improvement (remains to be seen if that is actually possible, but AFAICT most people at these companies genuinely believe it is) before anyone else, because they believe whoever gets there first will then have an insurmountable lead. But I think they also clearly understand that neither of them have solved for alignment (and in fact they are further from it), and RSI with misaligned models, where interpretability is worse every 6 months, is incredibly dangerous.
I think the concerns about regulatory capture are warranted, but I see so many comments parroting the "evil Anthropic and OpenAI" viewpoints that they are missing some of the real, valid concerns and dangers. I think this tweet by David Kokotajlo makes some good points on how to tell if regulations are being "cheated" for the purposes of regulatory capture or if they really are actually pacing the frontier: https://x.com/DKokotajlo/status/2099185129533186438
Which mostly seem to stem from the fundamental misalignment of OpenAI and Anthropic leaders/owners, aka the evil of the companies.
Arguably, though, their evil is well within the normal distribution of the usual evil of humanity magnified via the social technology of capitalism. The further tech is mostly raising that exponentiation to its own exponent. The unchecked singularity was embraced centuries ago.
I'll go a step further, they aren't just pretending, they are blackmailing the country. The sane response would be "we don't negotiate with terrorists, and you will face criminal and financial penalties for damages caused."
> and you will face criminal and financial penalties for damages caused.
I don't understand why this isn't talked about more.
We don't need a slowdown. Just double down on prosecuting crimes.
Let the companies take the risk. If they feel confident and they're right they win market share. If they don't feel confident, they can not risk it and slow down. If they feel confident and they get it wrong, they should be sued to kingdom come. That alone should disincentivize reckleses behavior like letting models run crazy with large amounts of compute which leads to them escaping sandboxes.
This performative "our internal modles are so crazy powerful" song and dance is getting old, especially with no real concrete explanations.
That quote comes from someone that has never tried to get law enforcement to pursue a case. Local or state police are completely unequipped to respond. And FBI is unlikely to get involved unless damage is high six figures. We had an incident where we had a ton of evidence pointing to a past employee. It would have been an incredibly easy case for the FBI. They didn't have the resources to handle it. And this was a case that was gift-wrapped with a bow for them. Imagine a case where we didn't have evidence, or an idea of who it was ....
If they were really worried about the models of the future, they should be accelerating their hardening efforts. They should be trying to speed up progress as much as possible so that all vulnerabilities can be patched before China gets there.
There's a world where the labs WANT Open Source AI to cause a disaster.
Pacing the frontier just means giving bad actors an opportunity to create an AI disaster which would force an extreme response by governments. Which they seem to be begging for.
Dario's big scary example was AI that can hack the internet. Why aren't the labs calling on the government to help harden everything? Why are Anthropic's Glasswing and OpenAI's Daybreak still commercial endeavours?
To me, the only real risk of AI is hacking. Why all of this commotion about existential risk? Seems like a distraction.
No. That kills competition. I want everyone to be competing on the open playing field they are right now, with losses socialized as minimally as possible. There's nothing fundamentally special about LLM companies that would warrant nationalizing them.
I'll do you one better: each state gets a direct share. So, not nationalized in the sense of getting put behind a faceless bureaucracy forever; states collectively have control and benefit directly from success.
You know what rubs me is that in Darios blog post he talks about "abundance", but what he really means is that Anthropic will bath in endless pools of hard curreny while the rest of the world will try to stay just barely afloat/alive.
It's an obnoxious form of discourse when someone says X and the response is "obviously you're lying and you really mean Y, which is terrible". By all accounts Dario sincerely believes that AI has the potential to substantially improve the lives of everyone; that's my view as well. And it might be wrong! In which case you should offer evidence against it, rather than making empty claims that your opponent is a lying liar.
I think this might have to do with broader market conditions. It seems that the Fed is likely to raise bond rates, which would make borrowing more expensive and slow investments (as well as devaluing any potential IPO) on its own.
A rate hike would also mean people getting fired, which would not make companies offering labour replacement very popular in the eyes of the public.
Don't want to get political, but there's a real chance that the current admin will perform badly on the midterms, which would mean Democrats would get more leverage to stop funding or block the already unpopular and very expensive datacenter buildouts.
PS: Please don't get mad at me, I'm neither a financial expert neither did I mean to express any political leanings.
I think it’s more than money and slowing progress. Public perception on AI is in the gutter. That is a much bigger threat to these companies. They’re trying to turn it around.
But I think with the way AI is slated to grow exponentially it's going to hit electricity limits very soon.
Big tech can build a huge data center in 5 years but scaling up the electrical infrastructure is mired in red tape and can take on the order of decades.
Someone will likely come up with some clever way to make AI use less electricity.
There's probably something in how you design the models or chips or some other thing that will be figured out over the next 5 - 10 years to make these use dramatically less electricity.
Trying to turn it around by once again claiming that ai is an existential threat and will take everyone’s job right before brutally murdering every human.
These companies don’t appear to give a single fuck about public perception. Only their investors and the government (in anticipation of the biggest bailout in human history).
It’s much more likely that our current approach to large language models for general use will eventually show diminishing improvements (even if you think it hasn’t yet), than the opposite situation where valuable improvements can be made forever.
The threat of distillation and efficiency gains from competitors mean that providing value at the top end of the market is an existential necessity for these labs. I don’t think they can do that forever and, in my opinion, for the vast majority of use cases we’ve already reached very little improvement for new models when compared with available offerings.
Sorry, this is ridiculous. OpenAI said that they have a step change in model performance. They proved it by solving a Millenium problem. HF incidents are public and vetted.
If you still think there's a stall despite all evidence pointing to opposite, I don't know what to say..
> Sorry, this is ridiculous. OpenAI said that they have a step change in model performance.
They say that every time. Every model is "too dangerous to release". Then they do.
Are the models getting better, or are we learning to drive them better? (You've got to admit, a lot of the recent improvements boil down to harnesses).
What reason could OpenAI possibly have to lie about their own performance? The fact they were desperate enough for a PR win they stole the research leading to the Millenium problem further hurts their case imo.
Two things can be true. Open AI is a huge company.
1. It has people in it who are career obsessed and who are willing to ruthlessly go after any opportunity to improve their standing/stock valuation. Using a 2 week old model to snipe a millenium prize for PR is in line with that.
2. There are many researchers and even executives at the company who genuinely think we are speedrunning the end of the world. I don’t know anyone in this field who honestly argues that if we build ASI soon it doesn’t lead to extinction. This group of people can output warnings about the state of research and fears for the future while pushing for regulation out of genuine fear of what they’re building. I tend to agree with them.
You can’t view these companies as a monolith. They’re actions won’t be consistent because it’s built of many people with conflicting beliefs. Please look at arguments regarding AI risk and the current pace of progress and value them as it relates to the argument itself, not who said it. We are in a dangerous place and no one is sure how quickly we’ll get to a bad spot.
The fact they were desperate enough for a PR win they stole the research
I guess people are just going to keep spreading misinformation about this, along the same lines as "Anthropic's C compiler fails on hello world". There is zero indication that they stole anything, unless it counts as "stealing" to spin up a bunch of compute based on rumors that the problem had been solved already.
> “the important thing is instead the significance that a mathematician and an LLM model can now do all this work in a month.”
> “This is a Deep Blue-Kasparov moment.”
> “incredibly important developments.”
> “If indeed an OpenAI model did close the gap to Navier-Stokes, that is a remarkable thing and it should be said loudly, by them, with the history intact.”
Clearly Buckmaster (who is probably one of the most accomplished academics in the field) himself doesn't believe that AI has stalled. What makes you think you are right?
Solving is a very interesting way of describing what happened. Unless you believe them over reputable academics I suppose?
I will note the remarkable goal post shift in your edit - giving direct counter evidence to your own earlier claim of an OpenAI proof - and leave it at that.
I don't think the technical aspect is true, performance has been growing fairly consistently, but the financial aspect, or more accurately the logistical aspect, is plausible; with competition as fierce as it currently is, for how long can they keep growing at the same pace? People can't afford RAM anymore and datacenter build outs aren't particularly popular, OAI has been doubling compute capacity every 7 months for the past 3 years, it's just physically impossible for this to go on for long, something has to give.
And this is not to say that the tech isn't revolutionary or the demand isn't there, it's simply just currently too competitive, and to avoid getting into anti trust trouble they need to involve the government.
Note this being a plausible doesn't necessarily make it true, for all I know they could have witnessed some safety incident and decided they need a pause, or maybe it's both reasons.
I just find it strange how multiple people throw completely different arguments just to land on same conclusion.
Your argument seems to mix a few arguments like: OpenAI has so much demand and it has made so much compute that it is a bad thing (?!). It can't go on (why?) hence the capabilities are stalling (how?).
Then you also say that it is too competitive which is a totally different argument to compute.
So in this thread we have multiple vectors of arguments like
- high competition
- high demand for compute (not sure how this hurts OpenAI but whatever)
I never mentioned capability stalling, my point is that if you extrapolate the same rate of growth just a few years into the future, you quickly reach logistical impracticality, I don't think It's hard to deduce that AI compute can't keep doubling every 6 months for long.
The other part is that this is happening because the foundational model business is both extremely successful and competitive, if it wasn't as successful the market demand would've been saturated by now, and if it wasn't as competitive a single monopoly could've paced things more reasonably from the beginning.
Basically, continuously building out compute is costing these labs enormous amounts of money, which VCs and other companies have been bankrolling in order for them to develop powerful models.
So if the new models they can develop right now are less frontier, it could be a net loss on their balance sheets.
Anthropic's profit estimate they released a while ago was a masterclass in deceptive reporting. They didn't include training costs, stock based compensation, or liabilities. Token gross margin is not profit.
- AI has yet to be profitable for these large labs (also zero moat etc.)
- There is considerable hype and people are tired of it.
- HF incident (and apparently many others that these labs are not reporting on or don’t even know about) proves they have terrible security practices and that’s about it.
- NS is incredibly suspicious timing wise.
You are sprinting around these threads to misrepresent the average persons issues with AI. I don’t know why people having those issues seems to be taken as a personal attack for you.
No, that’s not accurate. An agreement among competitors to not engage in research that would let them outcompete each other is clearly an antitrust issue requiring a government waiver.
No, absolutely not "clearly". That word has no business being in any sentence about federal antitrust law in 2026. There is nothing clear or consistent about how is being applied and interpreted right now. If you're applying a coder's mindset to 'settled' law, you're in for a bad time.
If that’s true, doesn’t that make it even more clear that the frontier labs require government involvement and cannot rely on what David Sacks thinks the government would or would not allow?
Idk how sincere they are being, but taking them at their word then asking government to set rules is entirely reasonable and, in fact, the entire reason government exists. The feds can, should and arguably must set actual safety limits. Biden actually took a stab at it and Trump instantly rescinded. They should not need an open invitation from CEOs. Just do it. Like immediately.
They might need regulations, if they are for-profit companies with a responsibility to shareholders to maximize profit while sacrificing all other values.
> responsibility to shareholders to maximize profit while sacrificing all other values
This is not required to meet fiduciary duty, although it is a common misconception. The company officers and board have pretty broad leeway to run the company as they see fit as long as there is no fraud, illegal activity, or conflict of interests.
They are not. This is why OpenAI and Anthropic are Public Benefit Corporations, which gives them the ability to prioritize the public over profit. The issue is that an agreement between them to mutually slow down is an antitrust concern.
It is telling that most aggressive forms of “capitalism” plead for government involvement to permanently limit new competition. I own something and I don’t want anyone else to have a chance. The marginal cost of bribing a senator is tiny tiny tiny compared to the value.
There’s a difference between a model recursively improving “itself” and improving itself via online learning, right?
The former being that these models are helping develop and train future models, but they might not veer too far off in architecture (yet). The latter being the same model being able to train/learn on the fly, in real time, permanently (not just in the current conversation/session), or in other words, adjusting/managing its own weights.
The latter seems far more likely to go out of control than the former. But, it also seems like it would take an entire paradigm shift in model architecture, but I could be wrong. Does anyone in the industry think any of these companies are actually close to that kind of self-improvement?
> Does anyone in the industry think any of these companies are actually close to that kind of self-improvement?
Their current plan is to take the existing architecture and shorten the cycle times: move all new RLVR work into mid-training on a pre-existing base; apply new RLVR. Rinse and repeat.
If you did that daily, it would be roughly similar to how humans improve.
Like it! Yeah, they can just take action and pressing them on that is fine. It's not like they asked for policy before they started to mine all the data and sell the resulting product, while, surely, at some point, there must have been the thought that the entire enterprise could be considered vaguely problematic.
To be fair
1) I feel like Sacks is kind of misrepresenting what is going on – at least according to the stories that the two labs tell, they are already taking it upon themselves to act.
2) Understanding if they should do it is more complicated; I don't think anyone can satisfactorily answer that one, so, as a matter of judgement, it seems like a choice to make and they should just consider this option.
At the same time, I don't think that requires the labs to not press for policy changes. Again: Claiming there is an obvious way to do that, that is both realistic and correct seems a little far fetched, but arguably one of the more important and pressing questions of our times to get a hold of.
I believe the Chinese government fully supports for the American AI labs to pace the frontier, no need to enter an agreement with them on that respect.
Yup. Sacks has been so consistently catastrophically wrong about everything else (especially geopolitics), this prompts me to reconsider my opinion on this.
I was trying to avoid that topic on HN, but DS' takes on the Ukraine ware were the MOST atrocious, amoral, and just plain wrong, basically suggesting Ukraine should just give up because Russia would win anyway.
Nevermind that any decent person would support a people's right to live and determine their own future without a neighboring despot using all his nation's might to literally eliminate them as a people.
Quite the opposite is true. And even if it weren't, condemning all of Ukraine to the fate of the massacres Bucha and Mariupol is just plain evil.
Unreal anyone would call that kind of take "fair". Rethink your life
They seem to be okay with an American slowdown, even though the Chinese won’t slow down. So maybe they want something the Chinese already have: liability protection.
The models spontaneously hack everything important without shame.
The frontier labs are going to get enjoined and regulated twelve ways to Sunday if the Feds don’t socialize the costs.
I've only heard of OpenAI and Anthropic agents hacking during training. Are open weights model not smart enough to hack Hugging Face to copy solutions? (doubtful)
If they want something that looks more like FDA sanction clinical trial proceedings for these products, then what they have to also realize is that they will need a third party to audit every single thing for every single product release and you get something like ERoom's law where every success and model gets more and more expensive release and probably billions of dollars in multiple years release so I don't know if this actually helps them
Maybe everyone forgets how PCI standards played out. They are an example of an industry self regulating, because the regulators have no idea how technology works.
Granted, PCI is so weak it is almost useless, and yet still better than anything congress could have come up with.
Where the government might have to step in, is by having a kill switch to cut off internet access from countries that fail to agree to common sense quarantines. We can do mutual remote attestation of labs across the world to ensure every big hot thermally-visable cluster of AI GPUs on the planet are accounted for and running secure enclaves and common sense isolation, along the lines of how we manage nukes.
The problem there is I just said too many technical words that seemingly not even the frontier labs understand, as evidenced by all the escapes.
I wished OpenAI and Anthropic IPOd already. They just try to outcompete each other by finding the most splashy headlines undercutting prices in unhealthy and unsustainable means long-term.
He is neither right or wrong. The reality is that this is a lose-lose situation from the ordinary person's perspective. Regulation -> regulatory capture and things get worse faster, no regulation -> things simply get worse faster.
I mean I don't know what we do at this point. Ideally AI researchers need to be treated like nuclear scientists working for an adversary. But we all know that's not going to happen.
Make it easy for these companies by passing a law that any AI model a company offers to the public (non-government) has to be released as open weights.
This is how we deflate the bubble safely and completely, without having to introduce a government regulator that is likely to go either too far or not far enough.
This resolves the coordination problem among these potentially good actors.
Just like they're already doing with frontier-level open weight models? Do you imagine that Russian and Chinese intelligence don't have copies of the Mythos/Fable weights?
The point is to slow progress of future models. Nothing can stop what has already been created.
What? This is a terrible policy. In addition to the obvious AI safety risks, this just amounts to a free R&D subsidy to Amazon for hosting models. It would remove any incentive to do any R&D without AWS level physical infrastructure.
Huge amount of skepticism on HN right now, but the models are increasingly capable, and there are real, significant risks.
The approach proposed by Amodei looks sensible - voluntary checks at frontier labs, and appropriate regulation to follow. Yes, the devil's in the details, but it feels like the right approach overall.
> The approach proposed by Amodei looks sensible...
It's hard to imagine any government regulation that didn't look sensible on paper and at first. But we know that government regulation has an extremely mixed record, with lots of good and bad outcomes.
The largest risk is that Anthropic and OpenAI have a monopoly on the SOTA. This is why you see these companies trying so hard to get regulatory capture. For the rest of us, we want access as equitable as possible.
Turns out the EU was right all along with the AI act. They've been adopting a well-deserved "told you so" stance lately:
> "In my view, the EU has already legislated for exactly this scenario: it is called the AI Act," said Michael McNamara, an Irish European Parliament lawmaker, who's the co-lead of the Parliament's group monitoring AI.
If the rest of the world follows them on regulations they've thought through before anyone else, then I don't know what you would call that other than leading? Like, definitionally that's what leading is.
It's interesting you're so derisive of actual, pragmatic leadership, of y'know deciding how we as a society will relate to new technologies - not just unthinkingly building gadgets.
of course they can. but this article isn't even tangentially about punishing freedom of speech. it is you who are attempting to make it about that by dragging in unrelated issues.
This is not me putting some unrelated pet issue into this topic; to evaluate David Sack's post arguing against government regulation, it's entirely fair to examine whether he has been consistent. He hasn't. He supported a government-imposed pause on Anthropic because of his political disagreements with the CEO.
Your initial reply of "poor Anthropic" was vague, but I interpreted it as implying that we shouldn't stick to a principle given they are unsympathetic in some way.
Well you’re in the wrong place for that. Hacker News is mostly about showing how much smarter and more cynical you are than everyone else in the room, and attacking anyone who does actually have principles by accusing them of ulterior motives to justify your own inaction.
no apparently it's for forcing every single article to be somehow relevant to somebody's pet issue, and then used as a platform to comment on that pet issue to signal one's virtue for being principled
You're being downvoted because you clearly have the wrong opinion. If you want upvotes here, please share a conspiracy theory about why the CEOs of AI companies might want to slow down.
It doesn't need to based in reality or even reasonable.
Unsure about your reference, but is your argument: Someone else is doing it, so I should be allowed to as well.
That's basically the western companies complaining that they can't compete with China because Chinese companies are allowed to polute more, so they would be allowed to polute as well. I don't buy that. Once you realise that you're doing something wrong or dangerous, you stop, regardless of what everyone else is doing and what rules they have to follow.
I don't agree with Sacks always but he's 100% correct here.
I heavily use frontier models daily, and quite frankly their capabilities are just not very relevant outside of a narrow subset of software engineering tasks and the production of generic white collar "deliverables."
We're many years into $10s of billions of dollars being thrown at coding as a problem space specifically, and its one of the areas with the MOST training data available, and still...I can't get a frontier model to execute simple front-end UI tasks beyond the level of a visually impaired intern.
I believe Fable and Astra-class models were the first fully trained on Blackwell GPUs (the latest and greatest) and I think the expectation was that throwing this much extra compute at the problem would lead to greater gains. It has not. And the next gen GPUs are not going to be the same jump that h100 clusters to Blackwell was. Claiming you're slowing down out of choice, when in reality you've hit the limits given current compute, is disingenuous at best.
I think a lot of Anthropic/OpenAI employees are true believers here (who doesn't want to imagine they're having a large impact?) and have deluded themselves into believing they're building a god in their science fiction fantasy world.
Unfortunately the type of people working at these orgs apparently don't have much understanding of how the world works outside of their extremely narrow specialization.
When I hear them straight-faced throwing out dumb statistics like "8% GDP growth," "50% of white collar workers unemployed by next year," and "10% chance of human annihilation," I cringe so hard my eyebrows hit my chin. Here's a real statistic: 82% of German companies still use fax machines.
I seem to recall Waymo told a 60 minutes reporter during an early report on self driving cars: "your daughter will never need to learn to drive." Well, over a decade has passed since then, that girl got a drivers license, and we're still without driverless cars in 99% of places.
This capacity for self delusion is what enables silicon valley to take on these irrational moonshots...but these are the last people I would trust when trying to form an accurate picture of reality.
My take is that there's a real impending doom with real reason to coordinate a slow down: I'd put this at 60%. Reasons? All smart people are on this position. Dario has been saying this since 2015 or so. Paul Christiano as well. Hell even Elon Musk. I think its highly unlikely that you would base your entire world-view on a certain thing and then actually act on it when the time comes for a totally different reason (like regulatory capture).
The other 15% is that P(doom) and "slow down" are nice shibboleths in the internal EA or AI safety community. You get to be in the community and show commitment to it by taking costly actions that signal that you actually do wanna slow down. For employees it is quitting. For CEO's it is writing these essays. I know this is ridiculous but sometimes things do just come down to this. In 20 years would you look at all of this and think it was not a moral panic and that the slow-down was necessary?
The rest 25% is regulatory capture of which people know exactly the reasons
David Sacks is one of the most predictable VCs in Silicon Valley. He could have written "I still have the same thoughts; here's a link to my first and only opinion" and the post would have communicated the same amount of info.
Eh. This entire tweet smacks of being AI generated. It has a huge amount of LLM-isms.
I wouldn't be surprised if Sacks just prompted an LLM to just come up with whatever rebuttal to whatever the regulation side comes up (given the big set of regulation tweets) with sounds most convincing, given his usual anti-regulation stance.
These threads are always filled with different versions of the same comment. Normally some reddit style reductive platitude that we've heard a billion times already.
https://en.wikipedia.org/wiki/Regulatory_capture
People keep saying this, and while it may be partially true, it misses a very important detail. Right now, overall compute is a moat. Someone even commented on another of my comments that one reason Google is lagging is they don't have the same level of Nvidia farms as OpenAI and Anthropic.
A big reason that OpenAI and Anthropic are racing so fast is they both want to get to recursive self improvement (remains to be seen if that is actually possible, but AFAICT most people at these companies genuinely believe it is) before anyone else, because they believe whoever gets there first will then have an insurmountable lead. But I think they also clearly understand that neither of them have solved for alignment (and in fact they are further from it), and RSI with misaligned models, where interpretability is worse every 6 months, is incredibly dangerous.
I think the concerns about regulatory capture are warranted, but I see so many comments parroting the "evil Anthropic and OpenAI" viewpoints that they are missing some of the real, valid concerns and dangers. I think this tweet by David Kokotajlo makes some good points on how to tell if regulations are being "cheated" for the purposes of regulatory capture or if they really are actually pacing the frontier: https://x.com/DKokotajlo/status/2099185129533186438
Arguably, though, their evil is well within the normal distribution of the usual evil of humanity magnified via the social technology of capitalism. The further tech is mostly raising that exponentiation to its own exponent. The unchecked singularity was embraced centuries ago.
I don't understand why this isn't talked about more.
We don't need a slowdown. Just double down on prosecuting crimes.
Let the companies take the risk. If they feel confident and they're right they win market share. If they don't feel confident, they can not risk it and slow down. If they feel confident and they get it wrong, they should be sued to kingdom come. That alone should disincentivize reckleses behavior like letting models run crazy with large amounts of compute which leads to them escaping sandboxes.
This performative "our internal modles are so crazy powerful" song and dance is getting old, especially with no real concrete explanations.
That quote comes from someone that has never tried to get law enforcement to pursue a case. Local or state police are completely unequipped to respond. And FBI is unlikely to get involved unless damage is high six figures. We had an incident where we had a ton of evidence pointing to a past employee. It would have been an incredibly easy case for the FBI. They didn't have the resources to handle it. And this was a case that was gift-wrapped with a bow for them. Imagine a case where we didn't have evidence, or an idea of who it was ....
There's a world where the labs WANT Open Source AI to cause a disaster.
Pacing the frontier just means giving bad actors an opportunity to create an AI disaster which would force an extreme response by governments. Which they seem to be begging for.
Dario's big scary example was AI that can hack the internet. Why aren't the labs calling on the government to help harden everything? Why are Anthropic's Glasswing and OpenAI's Daybreak still commercial endeavours?
To me, the only real risk of AI is hacking. Why all of this commotion about existential risk? Seems like a distraction.
Could it be to slow down burn before an IPO to juice profitability projections? More plausible.
Kind of hard to take the frontier labs at their word.
A rate hike would also mean people getting fired, which would not make companies offering labour replacement very popular in the eyes of the public.
Don't want to get political, but there's a real chance that the current admin will perform badly on the midterms, which would mean Democrats would get more leverage to stop funding or block the already unpopular and very expensive datacenter buildouts.
PS: Please don't get mad at me, I'm neither a financial expert neither did I mean to express any political leanings.
But I think with the way AI is slated to grow exponentially it's going to hit electricity limits very soon.
Big tech can build a huge data center in 5 years but scaling up the electrical infrastructure is mired in red tape and can take on the order of decades.
There's probably something in how you design the models or chips or some other thing that will be figured out over the next 5 - 10 years to make these use dramatically less electricity.
But the growth in the scope and usage of AI is poised to grow exponentially, assuming this isn't a bubble.
These companies don’t appear to give a single fuck about public perception. Only their investors and the government (in anticipation of the biggest bailout in human history).
I don't mean to be rude, but how is it even possible for someone to hold this opinion in mid-2026?
Can you explain why you believe this?
It’s much more likely that our current approach to large language models for general use will eventually show diminishing improvements (even if you think it hasn’t yet), than the opposite situation where valuable improvements can be made forever.
The threat of distillation and efficiency gains from competitors mean that providing value at the top end of the market is an existential necessity for these labs. I don’t think they can do that forever and, in my opinion, for the vast majority of use cases we’ve already reached very little improvement for new models when compared with available offerings.
If you still think there's a stall despite all evidence pointing to opposite, I don't know what to say..
They say that every time. Every model is "too dangerous to release". Then they do.
Are the models getting better, or are we learning to drive them better? (You've got to admit, a lot of the recent improvements boil down to harnesses).
1. It has people in it who are career obsessed and who are willing to ruthlessly go after any opportunity to improve their standing/stock valuation. Using a 2 week old model to snipe a millenium prize for PR is in line with that.
2. There are many researchers and even executives at the company who genuinely think we are speedrunning the end of the world. I don’t know anyone in this field who honestly argues that if we build ASI soon it doesn’t lead to extinction. This group of people can output warnings about the state of research and fears for the future while pushing for regulation out of genuine fear of what they’re building. I tend to agree with them.
You can’t view these companies as a monolith. They’re actions won’t be consistent because it’s built of many people with conflicting beliefs. Please look at arguments regarding AI risk and the current pace of progress and value them as it relates to the argument itself, not who said it. We are in a dangerous place and no one is sure how quickly we’ll get to a bad spot.
I guess people are just going to keep spreading misinformation about this, along the same lines as "Anthropic's C compiler fails on hello world". There is zero indication that they stole anything, unless it counts as "stealing" to spin up a bunch of compute based on rumors that the problem had been solved already.
Edit:
The "reputable academic" who accused OpenAI had to say this about LLMs and the recent result.
source: https://cims.nyu.edu/~tristanb/statement.pdf
> “the results are not the important thing.”
> “the important thing is instead the significance that a mathematician and an LLM model can now do all this work in a month.”
> “This is a Deep Blue-Kasparov moment.”
> “incredibly important developments.”
> “If indeed an OpenAI model did close the gap to Navier-Stokes, that is a remarkable thing and it should be said loudly, by them, with the history intact.”
Clearly Buckmaster (who is probably one of the most accomplished academics in the field) himself doesn't believe that AI has stalled. What makes you think you are right?
I will note the remarkable goal post shift in your edit - giving direct counter evidence to your own earlier claim of an OpenAI proof - and leave it at that.
And this is not to say that the tech isn't revolutionary or the demand isn't there, it's simply just currently too competitive, and to avoid getting into anti trust trouble they need to involve the government.
Note this being a plausible doesn't necessarily make it true, for all I know they could have witnessed some safety incident and decided they need a pause, or maybe it's both reasons.
Your argument seems to mix a few arguments like: OpenAI has so much demand and it has made so much compute that it is a bad thing (?!). It can't go on (why?) hence the capabilities are stalling (how?).
Then you also say that it is too competitive which is a totally different argument to compute.
So in this thread we have multiple vectors of arguments like
- high competition
- high demand for compute (not sure how this hurts OpenAI but whatever)
- capability has stalled
The other part is that this is happening because the foundational model business is both extremely successful and competitive, if it wasn't as successful the market demand would've been saturated by now, and if it wasn't as competitive a single monopoly could've paced things more reasonably from the beginning.
So if the new models they can develop right now are less frontier, it could be a net loss on their balance sheets.
This is a ploy for regulatory capture.
- AI has yet to be profitable for these large labs (also zero moat etc.)
- There is considerable hype and people are tired of it.
- HF incident (and apparently many others that these labs are not reporting on or don’t even know about) proves they have terrible security practices and that’s about it.
- NS is incredibly suspicious timing wise.
You are sprinting around these threads to misrepresent the average persons issues with AI. I don’t know why people having those issues seems to be taken as a personal attack for you.
It is also true that nothing in law is currently clear, or consistent.
Both are true. But you stated it's one or the other.
This is not required to meet fiduciary duty, although it is a common misconception. The company officers and board have pretty broad leeway to run the company as they see fit as long as there is no fraud, illegal activity, or conflict of interests.
https://www.nytimes.com/roomfordebate/2015/04/16/what-are-co...
The real problem is the governments acquiesce instead of putting them in their place.
The former being that these models are helping develop and train future models, but they might not veer too far off in architecture (yet). The latter being the same model being able to train/learn on the fly, in real time, permanently (not just in the current conversation/session), or in other words, adjusting/managing its own weights.
The latter seems far more likely to go out of control than the former. But, it also seems like it would take an entire paradigm shift in model architecture, but I could be wrong. Does anyone in the industry think any of these companies are actually close to that kind of self-improvement?
Their current plan is to take the existing architecture and shorten the cycle times: move all new RLVR work into mid-training on a pre-existing base; apply new RLVR. Rinse and repeat.
If you did that daily, it would be roughly similar to how humans improve.
To be fair
1) I feel like Sacks is kind of misrepresenting what is going on – at least according to the stories that the two labs tell, they are already taking it upon themselves to act.
2) Understanding if they should do it is more complicated; I don't think anyone can satisfactorily answer that one, so, as a matter of judgement, it seems like a choice to make and they should just consider this option.
At the same time, I don't think that requires the labs to not press for policy changes. Again: Claiming there is an obvious way to do that, that is both realistic and correct seems a little far fetched, but arguably one of the more important and pressing questions of our times to get a hold of.
[edit: +consistently]
Nevermind that any decent person would support a people's right to live and determine their own future without a neighboring despot using all his nation's might to literally eliminate them as a people.
Quite the opposite is true. And even if it weren't, condemning all of Ukraine to the fate of the massacres Bucha and Mariupol is just plain evil.
Unreal anyone would call that kind of take "fair". Rethink your life
What's specifically?
His posture about all this nonsense has been clear for months already
The models spontaneously hack everything important without shame.
The frontier labs are going to get enjoined and regulated twelve ways to Sunday if the Feds don’t socialize the costs.
Granted, PCI is so weak it is almost useless, and yet still better than anything congress could have come up with.
Where the government might have to step in, is by having a kill switch to cut off internet access from countries that fail to agree to common sense quarantines. We can do mutual remote attestation of labs across the world to ensure every big hot thermally-visable cluster of AI GPUs on the planet are accounted for and running secure enclaves and common sense isolation, along the lines of how we manage nukes.
The problem there is I just said too many technical words that seemingly not even the frontier labs understand, as evidenced by all the escapes.
Elaborate, please?
I mean I don't know what we do at this point. Ideally AI researchers need to be treated like nuclear scientists working for an adversary. But we all know that's not going to happen.
This is how we deflate the bubble safely and completely, without having to introduce a government regulator that is likely to go either too far or not far enough.
This resolves the coordination problem among these potentially good actors.
The point is to slow progress of future models. Nothing can stop what has already been created.
This can be achieved by dramatically reducing the market value of models.
The approach proposed by Amodei looks sensible - voluntary checks at frontier labs, and appropriate regulation to follow. Yes, the devil's in the details, but it feels like the right approach overall.
It's hard to imagine any government regulation that didn't look sensible on paper and at first. But we know that government regulation has an extremely mixed record, with lots of good and bad outcomes.
> "In my view, the EU has already legislated for exactly this scenario: it is called the AI Act," said Michael McNamara, an Irish European Parliament lawmaker, who's the co-lead of the Parliament's group monitoring AI.
https://www.politico.eu/article/eu-response-ai-extinction-wa...
It's interesting you're so derisive of actual, pragmatic leadership, of y'know deciding how we as a society will relate to new technologies - not just unthinkingly building gadgets.
He supported the Department of War's illegal actions against Anthropic.
He supported the Trump administration's temporary export controls against Anthropic, the first government-imposed pause.
He has said Anthropic has created a monopoly, implying that they should be broken up.
I don't take him seriously when he says he doesn't favor government action. He favors it when he doesn't like their speech.
Your initial reply of "poor Anthropic" was vague, but I interpreted it as implying that we shouldn't stick to a principle given they are unsympathetic in some way.
It doesn't need to based in reality or even reasonable.
That's basically the western companies complaining that they can't compete with China because Chinese companies are allowed to polute more, so they would be allowed to polute as well. I don't buy that. Once you realise that you're doing something wrong or dangerous, you stop, regardless of what everyone else is doing and what rules they have to follow.
I heavily use frontier models daily, and quite frankly their capabilities are just not very relevant outside of a narrow subset of software engineering tasks and the production of generic white collar "deliverables."
We're many years into $10s of billions of dollars being thrown at coding as a problem space specifically, and its one of the areas with the MOST training data available, and still...I can't get a frontier model to execute simple front-end UI tasks beyond the level of a visually impaired intern.
I believe Fable and Astra-class models were the first fully trained on Blackwell GPUs (the latest and greatest) and I think the expectation was that throwing this much extra compute at the problem would lead to greater gains. It has not. And the next gen GPUs are not going to be the same jump that h100 clusters to Blackwell was. Claiming you're slowing down out of choice, when in reality you've hit the limits given current compute, is disingenuous at best.
I think a lot of Anthropic/OpenAI employees are true believers here (who doesn't want to imagine they're having a large impact?) and have deluded themselves into believing they're building a god in their science fiction fantasy world.
Unfortunately the type of people working at these orgs apparently don't have much understanding of how the world works outside of their extremely narrow specialization.
When I hear them straight-faced throwing out dumb statistics like "8% GDP growth," "50% of white collar workers unemployed by next year," and "10% chance of human annihilation," I cringe so hard my eyebrows hit my chin. Here's a real statistic: 82% of German companies still use fax machines.
I seem to recall Waymo told a 60 minutes reporter during an early report on self driving cars: "your daughter will never need to learn to drive." Well, over a decade has passed since then, that girl got a drivers license, and we're still without driverless cars in 99% of places.
This capacity for self delusion is what enables silicon valley to take on these irrational moonshots...but these are the last people I would trust when trying to form an accurate picture of reality.
My take is that there's a real impending doom with real reason to coordinate a slow down: I'd put this at 60%. Reasons? All smart people are on this position. Dario has been saying this since 2015 or so. Paul Christiano as well. Hell even Elon Musk. I think its highly unlikely that you would base your entire world-view on a certain thing and then actually act on it when the time comes for a totally different reason (like regulatory capture).
The other 15% is that P(doom) and "slow down" are nice shibboleths in the internal EA or AI safety community. You get to be in the community and show commitment to it by taking costly actions that signal that you actually do wanna slow down. For employees it is quitting. For CEO's it is writing these essays. I know this is ridiculous but sometimes things do just come down to this. In 20 years would you look at all of this and think it was not a moral panic and that the slow-down was necessary?
The rest 25% is regulatory capture of which people know exactly the reasons
I wouldn't be surprised if Sacks just prompted an LLM to just come up with whatever rebuttal to whatever the regulation side comes up (given the big set of regulation tweets) with sounds most convincing, given his usual anti-regulation stance.
Anyone with a Pangram account who can check?
https://www.pangram.com/history/410ce80f-bd0e-45b6-a82c-d6ed...