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.
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).
Solving millennium problems doesn’t pay the bills.
Their best models might be quite good when directed at extremely difficult and very focused problems, but most business use cases are nothing like that.
If you believe Yann LeCun and David Silver (and others), there's an architecture wall; maybe labs are starting to see this on the horizon. Maybe the DRAM supply constraints are forcing it.
Are these real step changes - big picture wise, or refinements in RL/agentic orchestration/"taste" and advancements due to bigger models and hardware technology/capacity scaling? If they not, does this tactic - and hardware improvements - continue to scale non-linearly like they need to?
It is clear that whatever does change in each model increment has resulted in meaningfully better end user capabilities (as well as regressions in some areas, honestly), but that doesn't prove anything. I'm not sure what I personally believe, but stating with your full chest that a stall is ridiculous ignores a lot of potential evidence to the contrary.
Stupid example: Astra. Its main improvements are: much much better computer use and 3d modeling capabilities; better subagent orchestration; better and more reliable tool use; slightly worse coding.
This looks to me, from a feature perspective, to be an incremental improvement across several functional areas, plus new features which are unquestionably excellent but are probably the result of RL focus, not magic.
Step change? Ehhhh depends on how you squint. But how many more iterations of this do we have? Are we going to squeeze quintillion parameter transformers into GPUs?
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.
huge problem in discourse is that there are real interesting wedges in debate like p(doom) and RSI. but zitronion thought is not the real interesting wedge and almost everyone is on this side.
- AI can't be profitable (lol)
- everything is hype
- HF incident was a conspiracy between HF, METR, OpenAI, Anthropic to drive up market share
- Navier Stokes was solved because they literally looked at one of the codex chats and presented as their own (lol)
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)
"it's just physically impossible for this to go on for long"
I mean, my uneducated guess is that it's probably within the laws of physics for us to have at least 100x as much compute in the world as there is now. There would be a lot of engineering challenges, but I think it's physically possible. If so, OAI could double for at least a few more years.
At the current rate of 3.4x per year in the US, you get 100x the current capacity in just 3 more years and 9 months, that's not physically impossible yet, but just looking at the impact today makes it hard to imagine the impact 100x would have, and I find it very plausible for top labs to prefer making some sort of truce instead of keeping up this pace.
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.
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).
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.
I dunno, why is 90% the limit? Maybe they can set something up in the chips to use 1/10th the power and set something up in the model design to use 1/10th the power and so we use 99% less power than today.
After all GPUs were not really designed with AI efficiency in mind, and current AI architecture was not really designed with efficiency in mind, so there's probably some things that can be done. Over 5-10 years, who knows how much more efficient things could get?
Without pressure, the release valves will never be invented. Electricity, datacenter, and DRAM demand outstripping supply sounds like an opportunity - if we can only find a way to divert funding from the current gpu/ai funnel. Maybe jack up corporate tax rates like the 1970s, then every corporate office park will have a research lab :).
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.
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.
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.
LMFTFY: If all the LLM pure plays who are stuck between impossible valuations and the inability to IPO are agreement that they should colludes to conserve capital, they might not flame out before they figure out how to fix their numbers, and they might not be seen as admitting that more training data plus more GPUs will never lead to AGI.
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.
Mainstream narratives for the first year or two of the war were:
- Russia was a paper tiger and weak
- Ukraine was going to win against a corrupt/unmotivated Russian military so it should keep fighting
- Putin had made an error and he would get overthrown
- The Russian economy was going to collapse due to sanctions
- Russia oil pipelines would freeze and never come back online without western help (this was a Peter Zeihan favorite)
- Europe would come to Ukraine's defense
- Etc.
He threw cold water on every single one of those narratives from the beginning, meanwhile western media was busy spinning false rah-rah propaganda narratives like "the ghost of Kiev" etc.
Basically he was correct that we should not have cheered on this war. If a settlement and concessions were reached much earlier it would have been better for Ukraine (they likely would have kept more territory) and we wouldn't have 1+ million casualties and EU economies like Germany death spiraling and turning far right due to cutting themselves off from Russian energy.
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
So to clarify...your position is that: it's better that Ukraine has been destroyed and half a million have died fighting an un-winnable war...instead of taking lesser concessions upfront than they will ultimately have to make for this thing to end?
Russia was clearly in the wrong here, and I wish Ukraine could win, but reality exists, and we kind of have to anchor our decisions based on it or suffer the consequences.
I get it, realism is emotionally upsetting. But if you actually care about Ukraine, the only rational approach is realism.
Because while Russia was in a way stronger position than western media was portraying, they don't have an infinite capacity for pain.
They would absolutely love to not be sending their kids to die and selling natural resources to whomever again and traveling around the globe without sanctions.
The concessions they would have accepted at the beginning were less. However, now that they've lost so many men and risked their regime on it, they can't simply accept the same deal.
This is why the incentives that wars create are so terrible, especially for authoritarian regimes, even if they're at fault.
If Russia does not have "an infinite capacity for pain", then it does not sound like an unwinnable war. Ukraine estimates that they need to inflict at least 50 000 casualties a month to break the backbone of Russian military. The latest surge has brought it to 42 000 a month, up from 30 000 earlier this year. This already exceeds recruitment and training capacity.
Who are those who would love not to send their kids to die, sell resources and travel the world? Putin? His children already live a pampered life in palaces and he himself is probably the richest man in the world, currently on a trip to India. The rest don't matter. Even the head of security services Patrushev, historically the second most powerful man in the country, is scared of Putin like a schoolboy and fumbles his words before him. That has been one of the most shocking revelations since the war began: that oligarchs, security services, nor anyone else can put out even the slightest opposition. The whole country dances to Putin's whims. His only real opponent is time.
And Putin doesn't give a shit about Donbas. The aim is to restore the world that birthed him, the USSR at its peak. Russia has not backed one one inch from the 2021 ultimatum to return to the European political order of the 1980s. Any concession only confirms that attacking Ukraine was a viable strategy, that piece by piece he can claw his way closer to that goal.
This is the realistic take. Bomb Moscow before it is too late.
Wow, clueless. You entirely miss the point of the war.
Putin sees the mere existence of a successful Ukraine's as an existential threat. Putin initially got elected on promises that included becoming an advanced Western-style economy. He abandoned that to crony oligarchism, stripping the country for parts. Now, he and the Russian people see former Soviet state Ukraine actually becoming successful, well-off, and aspiring to EU membership. THAT can not be tolerated, because it would give Russian people the idea that they should have the same, and create a likely revolution at home.
So Putin's goal is to wipe Ukraine not only off the map, but off the consciousness of everyone in the world as Ukrainians as a people
Putin will stop until either 1) he achieves this goal, or 2) he is forced to stop.
THAT is the reality.
Thinking that Putin would somehow have accepted lesser sanctions earlier, and that he would have actually respected your new peace plan when he has never respected any other is to be played for a fool.
Allowing 1) to happen is unacceptable. Note the Ukrainians who are directly suffering the daily assaults can think as well as you can of the "realpolitik solution", and they reject it, because they know it is absolutely stupid, and leads only to rape, torture, and death — theirs, personally.
Making 2) happen is the only course of action. Ukrainians will follow it until they cannot.
If you want to stop the suffering, then accelerate Russia's exit. Garry Kasparov articulated a fine plan days ago [0]:
1: Provide complete protection of Ukrainian airspace
2: Immediately send Ukraine all the weapons it needs; no delay
3: Enforce existing sanctions. To this I add including seizing Russian and Iranian shadow fleet vessels
4: Cessation of diplomatic affairs with Russia
5: Seize Russian sovereign funds immediately and use them to power the war effort in Ukraine
This would be a start. Ukraine continuing its "long-range sanctions" is already bringing the Russian state coffers to rapid decline and looming budget crisis. Continuing it longer will force Putin's hand.
I agree the EU actually stepping up and entering the war in a serious way would tilt everything in Ukraine's favor.
The problem is they just won't. The most important members of the EU (Germany and France) are in shambles domestically and turning far right. Both have little desire to stick their necks out to support Ukraine any further and are likely turning the opposite direction.
Also, Russia has nukes. Nobody wants to provoke that any more than necessary. If Europe weren't a feckless, fragmented military vacuum things might be different, but nothing about that appears to be changing any time soon.
Your "realism" is [unprintable here]. Worse, it is not reality at all, but believing a false promise from Putin.
The ACTUAL realism is to recognize Putin's Russia for what it is: a relentless aggressor who will respect no peace. Putin's Russia has violated EVERY "peace agreement it has ever made. "Peace" in their mind is merely a strategic pause in the conflict to regroup and rearm.
The list is extremely long, Chechnya, Georgia, Crimea, Donbas, and more. Exactly ZERO concessions brought peace.
The REALITY is Ukraine submitting to such a peace would merely be buying another worse war a few months or years down the road, after Russia has had a breather and a chance to rearm. It'll just be more deaths at a faster pace in Ukraine.
If you think otherwise, I'd have both a bridge and some oceanfront property in Kansas to sell you.
Meanwhile, your view of Ukraine is years out of date and was never true.
Ukraine has been anything but "completely decimated" — they have emerged as the leading global power in drones and air defense.
As to your 'half-million who have died', they are overwhelmingly Russian fighters — the absolute f'kn aggressors, and documented war criminals, torturers, rapists, thieves and worse. Zero sympathy. If they didn't want to die they could have just stayed home.
Also recognize the absolute goal of Russia, and Putin has NOT ONCE backed off from the stated goal: The entire concept of Ukraine as a people and as a state must be eliminated.
There is not even a surrender that can be acceptable to the people of Ukraine.
There is no compromise with someone who has the goal of murdering you or genociding your people.
The ONLY realistic solution is to boot Russia from every square inch of Ukrainian land, and force it back i it's box.
You obviously haven't noticed that Ukraine is on its way to accomplishing that, both decimating Russia's economic engine, and now taking back more land than RU has taken in 2026.
The shame here is the Western countries and Russia convinced Ukraine to give up nukes in return for inviolate borders, and when Russia violated that treaty back in 2014 taking Crimea, not one Western country pushed back.
Allowing Ukraine to get wiped off the map, and Ukrainians each, individually, to be forced to abandon their identity and language, and suffer rape, torture, and death may be a "realistic" option, but it is not an option to anyone who has an ounce of morality. Perhaps you should learn about it.
The only outcome that actually works is to force despots like Putin back into their box. Preferably six feet underground.
It sounds like your whole position is again dependent on the wishful thinking narrative of this war being correct (which it hasn't been thus far).
The numbers I cited were purposely in between the Russian numbers of Ukrainian casualties (1.5M, obviously false) and the Ukrainian numbers (100k, obviously undercounted). The Netherlands intelligence service and the CSIS put the number around 500K, which could also be wrong, but intelligence services tend to have better methods than media orgs like the BBC: https://en.wikipedia.org/wiki/Casualties_of_the_Russo-Ukrain...
As far as your assessment of 'decimated', the UN projects Ukraine will require over half a $Trillion to rebuild. For a country with a GDP per capita of $7,000...I would call that decimated unfortunately. And that's not something that makes me happy: https://www.worldbank.org/en/news/press-release/2026/02/23/u...
The idea that Ukraine is winning or Russia is winning on any given week is entirely dependent on which news source you're reading, and given the fog of war, unless you have actual intelligence service briefs we don't actually know how its going other than the fact the frontlines have basically stayed +/- some kilometers for a super long time at this point.
Also, the idea that Russia would not accept concessions and also that Putin being dead would be desirable (hard truth: there's people who are even more hardliners than Putin waiting in the wings), are unfortunately incomplete assessments of the reality.
Again, I don't disagree with your emotional take on this. But on the reality of the situation I don't think your assessment is accurate as far as what's better for Ukraine long term.
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.
Source? on HN you can say whatever you want. Your own opinion, without even facts to back it up. The more confident you sound the more convincing you are.
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
When it stops with the compacted isomorphic plagiarism using other peoples work like an intelligence campaign. LLM are not "AI" in my opinion, but are good at domain context search. Neuromorphic computing may change that one day, but it will unlikely arise from the LLM cults. =3
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.
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.
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.
> RSI with misaligned models, where interpretability is worse every 6 months, is incredibly dangerous.
This. So very much this. Misalignment (among other things) means that the parent in the RSI cycle isn't going to be working as hard on the alignment of the children as we need.
Very true. Or further, access to capital is the moat. It is the very reason that we don't have a real open-source community that trains frontier models - individuals simply can't afford the training infrastructure, nor sufficient high-quality training data.
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.
Yeah it's either that or the scientists are right about their field of study. Absolutely shocking and depressing how many people are willing to trust this Capitalist Realism hunch over the world in front of their eyes.
I get your point. There’s a lot hysteria right now. However I think people are willing to trust a capitalist realism hunch because over the last >20 years, that hunch is usually going to be right.
> 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.
If it's true, it makes them evil. Especially Dario, who speaks about moral ground and fate of humanity all day, yet it's not that different from a cult leader does.
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.
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.
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.
> 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.
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.
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.
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
This is the same social group that executed one of the biggest financial frauds in history. They just want money, like everyone else. Everything else is a way to justify how they're going to go get it. These are all smart people who know exactly what they're doing. They absolutely love the idea that AI is more dangerous than nukes, because they're the ones building it, and that means they control something more dangerous than nukes, which they all know will make them tons of money. Stating any of this directly reduces credibility with those outside the group, which reduces the amount of control the group has, which reduces the amount of money they can make. Therefore, they select for people who can infer the strategy and push it forward without ever stating any of it. The most valuable members of the group are those who can convince those outside the group of the group's strategy without ever stating it, and that's what Dario's essay is an example of.
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.
They might need regulations, if they are for-profit companies with a responsibility to shareholders to maximize profit while sacrificing all other values.
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.
> 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.
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.
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.
Once they observe real safety risks, they're going to slow down on their own, because a major safety incident would destroy their valuations and businesses. The beauty of this principle is that it applies to everyone all over the world. Nobody is exempt from it. There's no coordination required. No country would benefit from developing an AI system that wreaks havoc on its own infrastructure, so there's a powerful built-in global incentive to build safe, steerable systems.
Right now the risks are hypothetical. That doesn't mean we shouldn't take them seriously, but crippling an industry through government intervention before we have any real understanding of it is clearly unnecessary. The fearmongering is clouding basic judgment and thinking. It's possible to take the risks seriously without immediately introducing poorly understood regulation which will have its own dangerous side effects.
OpenAI and Anthropic are doing extraordinary amounts of alignment testing internally. There is no scenario where they release a dangerous model without realizing what they are doing.
The only thing the government needs to do is hold AI companies legally accountable if their products cause harm. That's it. That's not what the labs want. They don't want to be held legally accountable for harm, and they do want there to be a huge compliance burden that only they can afford to pass.
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.
Hacking the internet includes things like "hacking banks", "hacking power plants", "hacking water treatment plants", and "hacking cars". How about "hacking vote counting machines"?
Hardening everything is in fact part of the answer. Getting stuff disconnected from the internet that shouldn't be connected is also part.
> you will face criminal and financial penalties for damages caused
Even with where we're at right now, if people involved with the HF or Ruby Gems attacks faced legal consequences, the industry would get a lot more conservative very quickly. And unlike social media, section 230 doesn't apply, and there aren't any free speech concerns.
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
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.
Here's the context for the "abundance" reference, for those curious:
> I believe that AI could cure most major diseases in the next 5–10 years, greatly accelerate economic growth rates, create a world of abundance and empowerment, and usher in a renaissance of democracy and freedom.
Whether that matches your description of it, I'll leave for folks to decide for themselves.
I have high hopes for AI yet am hard pressed to envision said abundance trickling down to the common man. The diseases cured will be locked behind a paywall and a renaissance of democracy and freedom is laughable in 2026, let alone beyond.
Dario and his team are on top of the world. He is a good salesman, no doubt. The shareholders will eat well.
I don't agree with Sacks on a lot 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 autistic intern.
Fable and Astra-class models were the first ones trained on Blackwell GPUs (the latest and greatest) and from what I've heard from folks on the inside, 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.
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 don't really have much understanding of how the world works outside of their extremely narrow specialization.
When I hear them throwing out silly 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, and we're still without driverless cars in 99% of places.
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.
140 comments
[ 0.18 ms ] story [ 9.4 ms ] threadI 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?
So if the new models they can develop right now are less frontier, it could be a net loss on their balance sheets.
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).
Their best models might be quite good when directed at extremely difficult and very focused problems, but most business use cases are nothing like that.
Are these real step changes - big picture wise, or refinements in RL/agentic orchestration/"taste" and advancements due to bigger models and hardware technology/capacity scaling? If they not, does this tactic - and hardware improvements - continue to scale non-linearly like they need to?
It is clear that whatever does change in each model increment has resulted in meaningfully better end user capabilities (as well as regressions in some areas, honestly), but that doesn't prove anything. I'm not sure what I personally believe, but stating with your full chest that a stall is ridiculous ignores a lot of potential evidence to the contrary.
Stupid example: Astra. Its main improvements are: much much better computer use and 3d modeling capabilities; better subagent orchestration; better and more reliable tool use; slightly worse coding.
This looks to me, from a feature perspective, to be an incremental improvement across several functional areas, plus new features which are unquestionably excellent but are probably the result of RL focus, not magic.
Step change? Ehhhh depends on how you squint. But how many more iterations of this do we have? Are we going to squeeze quintillion parameter transformers into GPUs?
This is a ploy for regulatory capture.
- AI can't be profitable (lol)
- everything is hype
- HF incident was a conspiracy between HF, METR, OpenAI, Anthropic to drive up market share
- Navier Stokes was solved because they literally looked at one of the codex chats and presented as their own (lol)
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
I mean, my uneducated guess is that it's probably within the laws of physics for us to have at least 100x as much compute in the world as there is now. There would be a lot of engineering challenges, but I think it's physically possible. If so, OAI could double for at least a few more years.
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).
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.
After all GPUs were not really designed with AI efficiency in mind, and current AI architecture was not really designed with efficiency in mind, so there's probably some things that can be done. Over 5-10 years, who knows how much more efficient things could get?
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.
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.
It doesn't need to based in reality or even reasonable.
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.
What's specifically?
- Russia was a paper tiger and weak
- Ukraine was going to win against a corrupt/unmotivated Russian military so it should keep fighting
- Putin had made an error and he would get overthrown
- The Russian economy was going to collapse due to sanctions
- Russia oil pipelines would freeze and never come back online without western help (this was a Peter Zeihan favorite)
- Europe would come to Ukraine's defense
- Etc.
He threw cold water on every single one of those narratives from the beginning, meanwhile western media was busy spinning false rah-rah propaganda narratives like "the ghost of Kiev" etc.
Basically he was correct that we should not have cheered on this war. If a settlement and concessions were reached much earlier it would have been better for Ukraine (they likely would have kept more territory) and we wouldn't have 1+ million casualties and EU economies like Germany death spiraling and turning far right due to cutting themselves off from Russian energy.
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
Russia was clearly in the wrong here, and I wish Ukraine could win, but reality exists, and we kind of have to anchor our decisions based on it or suffer the consequences.
I get it, realism is emotionally upsetting. But if you actually care about Ukraine, the only rational approach is realism.
There is surprisingly little realism among the crowd calling themselves realists.
They would absolutely love to not be sending their kids to die and selling natural resources to whomever again and traveling around the globe without sanctions.
The concessions they would have accepted at the beginning were less. However, now that they've lost so many men and risked their regime on it, they can't simply accept the same deal.
This is why the incentives that wars create are so terrible, especially for authoritarian regimes, even if they're at fault.
Who are those who would love not to send their kids to die, sell resources and travel the world? Putin? His children already live a pampered life in palaces and he himself is probably the richest man in the world, currently on a trip to India. The rest don't matter. Even the head of security services Patrushev, historically the second most powerful man in the country, is scared of Putin like a schoolboy and fumbles his words before him. That has been one of the most shocking revelations since the war began: that oligarchs, security services, nor anyone else can put out even the slightest opposition. The whole country dances to Putin's whims. His only real opponent is time.
And Putin doesn't give a shit about Donbas. The aim is to restore the world that birthed him, the USSR at its peak. Russia has not backed one one inch from the 2021 ultimatum to return to the European political order of the 1980s. Any concession only confirms that attacking Ukraine was a viable strategy, that piece by piece he can claw his way closer to that goal.
This is the realistic take. Bomb Moscow before it is too late.
Putin sees the mere existence of a successful Ukraine's as an existential threat. Putin initially got elected on promises that included becoming an advanced Western-style economy. He abandoned that to crony oligarchism, stripping the country for parts. Now, he and the Russian people see former Soviet state Ukraine actually becoming successful, well-off, and aspiring to EU membership. THAT can not be tolerated, because it would give Russian people the idea that they should have the same, and create a likely revolution at home.
So Putin's goal is to wipe Ukraine not only off the map, but off the consciousness of everyone in the world as Ukrainians as a people
Putin will stop until either 1) he achieves this goal, or 2) he is forced to stop.
THAT is the reality.
Thinking that Putin would somehow have accepted lesser sanctions earlier, and that he would have actually respected your new peace plan when he has never respected any other is to be played for a fool.
Allowing 1) to happen is unacceptable. Note the Ukrainians who are directly suffering the daily assaults can think as well as you can of the "realpolitik solution", and they reject it, because they know it is absolutely stupid, and leads only to rape, torture, and death — theirs, personally.
Making 2) happen is the only course of action. Ukrainians will follow it until they cannot.
If you want to stop the suffering, then accelerate Russia's exit. Garry Kasparov articulated a fine plan days ago [0]:
1: Provide complete protection of Ukrainian airspace
2: Immediately send Ukraine all the weapons it needs; no delay
3: Enforce existing sanctions. To this I add including seizing Russian and Iranian shadow fleet vessels
4: Cessation of diplomatic affairs with Russia
5: Seize Russian sovereign funds immediately and use them to power the war effort in Ukraine
This would be a start. Ukraine continuing its "long-range sanctions" is already bringing the Russian state coffers to rapid decline and looming budget crisis. Continuing it longer will force Putin's hand.
[0] https://ukrainetoday.org/five-hammer-blows-that-could-finall...
The problem is they just won't. The most important members of the EU (Germany and France) are in shambles domestically and turning far right. Both have little desire to stick their necks out to support Ukraine any further and are likely turning the opposite direction.
Also, Russia has nukes. Nobody wants to provoke that any more than necessary. If Europe weren't a feckless, fragmented military vacuum things might be different, but nothing about that appears to be changing any time soon.
The ACTUAL realism is to recognize Putin's Russia for what it is: a relentless aggressor who will respect no peace. Putin's Russia has violated EVERY "peace agreement it has ever made. "Peace" in their mind is merely a strategic pause in the conflict to regroup and rearm.
The list is extremely long, Chechnya, Georgia, Crimea, Donbas, and more. Exactly ZERO concessions brought peace.
The REALITY is Ukraine submitting to such a peace would merely be buying another worse war a few months or years down the road, after Russia has had a breather and a chance to rearm. It'll just be more deaths at a faster pace in Ukraine.
If you think otherwise, I'd have both a bridge and some oceanfront property in Kansas to sell you.
Meanwhile, your view of Ukraine is years out of date and was never true.
Ukraine has been anything but "completely decimated" — they have emerged as the leading global power in drones and air defense.
As to your 'half-million who have died', they are overwhelmingly Russian fighters — the absolute f'kn aggressors, and documented war criminals, torturers, rapists, thieves and worse. Zero sympathy. If they didn't want to die they could have just stayed home.
Also recognize the absolute goal of Russia, and Putin has NOT ONCE backed off from the stated goal: The entire concept of Ukraine as a people and as a state must be eliminated.
There is not even a surrender that can be acceptable to the people of Ukraine.
There is no compromise with someone who has the goal of murdering you or genociding your people.
The ONLY realistic solution is to boot Russia from every square inch of Ukrainian land, and force it back i it's box.
You obviously haven't noticed that Ukraine is on its way to accomplishing that, both decimating Russia's economic engine, and now taking back more land than RU has taken in 2026.
The shame here is the Western countries and Russia convinced Ukraine to give up nukes in return for inviolate borders, and when Russia violated that treaty back in 2014 taking Crimea, not one Western country pushed back.
Allowing Ukraine to get wiped off the map, and Ukrainians each, individually, to be forced to abandon their identity and language, and suffer rape, torture, and death may be a "realistic" option, but it is not an option to anyone who has an ounce of morality. Perhaps you should learn about it.
The only outcome that actually works is to force despots like Putin back into their box. Preferably six feet underground.
The numbers I cited were purposely in between the Russian numbers of Ukrainian casualties (1.5M, obviously false) and the Ukrainian numbers (100k, obviously undercounted). The Netherlands intelligence service and the CSIS put the number around 500K, which could also be wrong, but intelligence services tend to have better methods than media orgs like the BBC: https://en.wikipedia.org/wiki/Casualties_of_the_Russo-Ukrain...
As far as your assessment of 'decimated', the UN projects Ukraine will require over half a $Trillion to rebuild. For a country with a GDP per capita of $7,000...I would call that decimated unfortunately. And that's not something that makes me happy: https://www.worldbank.org/en/news/press-release/2026/02/23/u...
The idea that Ukraine is winning or Russia is winning on any given week is entirely dependent on which news source you're reading, and given the fog of war, unless you have actual intelligence service briefs we don't actually know how its going other than the fact the frontlines have basically stayed +/- some kilometers for a super long time at this point.
Also, the idea that Russia would not accept concessions and also that Putin being dead would be desirable (hard truth: there's people who are even more hardliners than Putin waiting in the wings), are unfortunately incomplete assessments of the reality.
Again, I don't disagree with your emotional take on this. But on the reality of the situation I don't think your assessment is accurate as far as what's better for Ukraine long term.
His posture about all this nonsense has been clear for months already
https://en.wikipedia.org/wiki/Regulatory_capture
Did you know that an LLM solved a millennium problem last week?
Do you have any personal threshold past witch you will acknowledge that this technology is real?
https://en.wikipedia.org/wiki/The_Subservient_Chicken
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.
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.
This. So very much this. Misalignment (among other things) means that the parent in the RSI cycle isn't going to be working as hard on the alignment of the children as we need.
Very true. Or further, access to capital is the moat. It is the very reason that we don't have a real open-source community that trains frontier models - individuals simply can't afford the training infrastructure, nor sufficient high-quality training data.
If it's true, it makes them evil. Especially Dario, who speaks about moral ground and fate of humanity all day, yet it's not that different from a cult leader does.
Hitler level evil.
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.
> "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 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.
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.
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...
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
The real problem is the governments acquiesce instead of putting them in their place.
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...
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.
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.
Right now the risks are hypothetical. That doesn't mean we shouldn't take them seriously, but crippling an industry through government intervention before we have any real understanding of it is clearly unnecessary. The fearmongering is clouding basic judgment and thinking. It's possible to take the risks seriously without immediately introducing poorly understood regulation which will have its own dangerous side effects.
OpenAI and Anthropic are doing extraordinary amounts of alignment testing internally. There is no scenario where they release a dangerous model without realizing what they are doing.
The only thing the government needs to do is hold AI companies legally accountable if their products cause harm. That's it. That's not what the labs want. They don't want to be held legally accountable for harm, and they do want there to be a huge compliance burden that only they can afford to pass.
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 ....
How many figures of damage did Aaron Swartz cause?
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.
Hardening everything is in fact part of the answer. Getting stuff disconnected from the internet that shouldn't be connected is also part.
Even with where we're at right now, if people involved with the HF or Ruby Gems attacks faced legal consequences, the industry would get a lot more conservative very quickly. And unlike social media, section 230 doesn't apply, and there aren't any free speech concerns.
FDA is just way too much red tape.
> I believe that AI could cure most major diseases in the next 5–10 years, greatly accelerate economic growth rates, create a world of abundance and empowerment, and usher in a renaissance of democracy and freedom.
Whether that matches your description of it, I'll leave for folks to decide for themselves.
Dario and his team are on top of the world. He is a good salesman, no doubt. The shareholders will eat well.
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 autistic intern.
Fable and Astra-class models were the first ones trained on Blackwell GPUs (the latest and greatest) and from what I've heard from folks on the inside, 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.
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 don't really have much understanding of how the world works outside of their extremely narrow specialization.
When I hear them throwing out silly 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, and we're still without driverless cars in 99% of places.
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.
What if we get the govt to regulate the industry and create an oligopoly?! Genius...