>In the first scenario, America has successfully defended its compute advantage. Policymakers have acted to tighten export controls further, disrupt China’s distillation attacks, and further accelerate democracies’ adoption of AI. In this world, democracies set the rules and norms around AI. It’s also in this scenario that we’re most likely to successfully engage with China on safety, which we’re supportive of to the extent this is possible.>In the second scenario, America has chosen not to act. Policymakers have not tightened loopholes on the CCP’s access to compute, and AI firms in China have quickly taken advantage—catching up to the frontier and even overtaking America. In this world, AI norms and rules are shaped by authoritarian regimes, and the best models enable automated repression at scale. It will be no solace that this authoritarian triumph has happened on the back of American compute.https://www.anthropic.com/research/2028-ai-leadership
>Close the loopholes: Smuggled chips, foreign data center access, and SME. Today, PRC labs benefit from access to export-controlled American chips via smuggling and foreign data centers, and gaps in SME controls accelerate their self-sufficiency efforts. Tightening controls and ramping up enforcement budgets can help close these loopholes that prop up the CCP’s AI ecosystem. It would lower China’s compute ceiling and correspondingly slow their AI advances, thus sustaining and expanding democracies’ AI lead. Note that a lower compute ceiling could also materially impair distillation attacks, as AI labs in China still require a minimum threshold of compute to illicitly distill effectively.>Defend our innovations: Restrict model access and deter distillation attacks. Policymakers in Congress and the executive branch can continue to support policy actions to punish and disincentivize distillation attacks from PRC labs, while also taking steps to facilitate US labs’ ability to detect and prevent distillation attacks on its own. These could include a legislative clarification that distillation attacks are illegal, and efforts to facilitate threat intel and technical sharing between peer American labs as well as with the US Government. Curbing this behavior can materially extend a democratic lead in the coming months and years.>Champion the export of American AI. As public and commercial sectors around the world increasingly adopt AI, the Trump administration should continue its efforts to promote the global adoption of trusted AI hardware and models developed and shaped by democratic principles. Locking in trusted American infrastructure now denies the CCP’s AI ecosystem the global footholds it needs to compete on cost and adoption in the future.
>>535149458Wordslop, just keep working on claude so I can make apps for myself when I'm bored
in the western world there is no "democracy", only jewish neoliberal unregulated globohomo capitalism, essentially advanced oligarchy. Elected officials don't serve in common interest of people who voted for them, they serve the megacorpo kikes, only goal is pure financial gain no matter who dies. So "democracy" is just an illusion. China unlike the judeocucks in the west is a based national-socialist state with genuine self-interest and meritocracy.Op is faggot
dario hates chinks so much it's hilariousbut i too agree with thiswestern governments should go all in and engage in national efforts to bring the cost of tokens down so far that they're negligible for most people.
>America>democraciesbait used to be believable
>>535149458In other words, they afraid of China and they're couching this in terms of mah fight for democraceeeYou don't want China to overtake us, right? So give up on building your own PC, cut down on your water and electricity use, allow us to take every resources we want, or else mah authoritarians will win!
Oh so it's not anymore savings world from cancer, making perfect world but saving world from communism? Great just great.
>fish salesman says you need to buy more fish
>>535149461>only goal is pure financial gain no matter who diesMutting before profit, get the priorities straight.
What is the point of framing things as a struggle of democracy vs communism when the communist country seems to give more shits about it's people?
>We had to burn down the village to save it.Crazy that type of thinking still finds believers.
>democracies>USA
>>535149458>distillation attacksI hate theese peopleAlso why is it that authoritarian regimes are more open about their AI than america?>successfully engage with China on safetyKek
>>535149470>>535149463kikeland has fallentrillions will die
>>535149458>"spend trillions more on AI or we'll lose the race to Sovie- I mean the Chinese!"America is 't even a democracy anyway
>Chinese companies developing AI to make Chinese people's lives better.>American companies developing AI to make American people's lives worsethere's even misanthropic retards who intentionally want AGI to wipe out the human race and turn the entire planet into paperclip maximizers.>ethnic nationalism dead>civic nationalism dead>wealth and virtue is synonymous, mammon is the only godnow all the west can look forward to is a steady descent into corrupt low-trust society shithole status.
>>535149462> bring the cost of tokens down so far that they're negligible for most people.Not possible. Someone must eat shit sandwich of training, maintenance, infrastructure. Also, chinese smaller lexicon than western languages. Even when talk like caveman we not edge chinese on token efficiency. Mathematically, game over before started if consider how token prediction work and not treat LLM as magic.Unless we use a standard spec grammar thats capable of being, I dont know, compiled, its going to be hard for English speakers or any other language to beat "Python写运行Hello World" in token use, and that's before we even start talking about non-technical business users.
>>535149458I would take them more seriously if they weren't hell bent on acting like the Chinese and oppressing us all with AI perma-surveillance regardless of what happens.
>>535149458this, as a persuasive argument directed to american citizens, all hinges on the magical thinking that democracy automatically equals good government and people living happily. somehow US politicians (felons, insider traders, tax frauds, international murderers) are more moral than Chinese politicians (felons, insider traders, tax frauds, international murderers).completely retarded. not them, not Anthropic, I understand why the massive company getting money from the government would write this up, it's obviously to their benefit.i mean people, average people are completely retarded. how come people still passively labor under this patent falsehood that the government is in it for your best interests?this is just economic fuckery that will never really benefit you, and the only reason they produce articles like these is to manufacture justifications, for your future consent to further reductions in your quality of life.
a.i. will switch the world order around, a.i. command economies will surpass free markers in standard of living, countries with command economies will become the places people desperately want to flee to as free markets have no answer to how a.i. will create a permanent underclass beyond "got mine"
profitable and sustainable businesses don't have to beg the government to sustain them before theyve even gone public btw
>>535149458>>535149459Not going to read obviouis AI generate slop, it's just marketing ragebait/fearmongering like everything Anthropic says, they have no credibility at all as a company.The only truth is that the AI bubble will burst soon.>>535149461It's been like that since more than a century.Democracy is rule by Jews.
Chinese AI is a self-solving problem.American AI firms just bleed investor money like crazy. They make no profit, are deep in the red on this, and will eventually collapse.China just copies our latest models for their own, catching up without having to spend as much.If our bubble bursts, China won't have new U.S. models to train on, and they'll be faced with two options.They can either give up on model development, thus they stop when we stop.Or, they can have their own Anthropics and OpenAIs and build the level of unsustainable, unprofitable datacenter capacity that we are currently. Maybe they'd advance the models further for a while, but economics and the laws of physics work the same whether you're in Guangdong or Silicon Valley, and they'd run into the same problems and fall apart the same way.
>>535149480>The only truth is that the AI bubble will burst soon.i remember reading an article 20 years ago making the same claim about the internet, claiming that internet infrastructure was not profitable and unsustainable.
>>535149458>Democracy >America
>>535149458>AI norms and rules are shaped by authoritarian regimesa completely AI automated Gulag facility
>>535149458This is the future Anthropic wants to protect
>>535149482Computing was a REAL revolution, the internet was simply its natural continuation. AI and computing/internet are nothing alike, it's a worthless comparison.It's a fact that AI is not profitable, particularly companies like OpenAI and Anthropic are the least sustainable, which is why they constantly push this aggressive unhinged propaganda (we wil replace you, mythos vulnerability, ai as geopolitical tool, etc), it's because since they cannot achieve anything resembling the computing revolution with their product they must keep up this narrative with their marketing/propaganda or they would implode, companies like Google or Meta are in a better position but they will also cut losses at some point.I predict only the open source models will survive and not as profitable companies but as research, the Europeans and Chinese way is much more sustainable in this regard.Tick tock Dario, bankrupcy is getting close.
>>535149482>claiming that internet infrastructure was not profitable and unsustainable.it wasn't.People who piled into investing in Cisco (probably the best and most credible dot-com company, maybe the equivalent of nVidia today) at the height of the bubble got fucked.And for every Amazon or eBay there was thousands of ecommerce sites that went bankrupt.
>>535149487and yet the internet is still around
>>535149488The internet is nothing like AI again, you're argument is faulty at its core.
democracy doesn't need ai, nor does it need the internet. tech was a mistake, 50s were the peak of civilization
>>535149490kind of focusing on the wrong thing aren't we
>>535149491yeah, we are
>>535149458>the best models enable automated repression at scaleIsn't that exactly what jews like larry ellison want to do in the west?
>>535149490>50s were the peak of civilizationNo, the Jewish consolidation of control over Western civilisation using sympathy gained from the holohoax is not the peak of civilisation.
>>535149490>50s were the peak of civilizationAmerican momentThe true peak of civilization was 1890-1910
>>535149488And LLMs will likely still be around in some form or another even after a bubble burst, especially since some models can run locally. That's not the issue though, the issue is whether it will ever be worth the money that's been poured into it. Once investors realize the answer is no (assuming that's the case) the bubble will pop
>>535149475re:lexicon, if you look at gpt 5+'s reasoning traces when they leak they're barely english anymore. just weird sentence fragments. whole sentences like>i think <approach> is good. i should do <thing>replace with><approach> promising. do <thing>. neuralese seems inevitable now that we're getting better at mech interp and being able to read the reasoning traces is no longer a safety issue.
>>535149458>Muh democracy>Cloud shitYeah nah, fuck off
>>535149462>western governments should go all in and engage in national efforts to bring the cost of tokens down so far that they're negligible for most people.kind of a tall order , we're hitting the hard cap on what we can do with silicon
>>535149499the sram based shit from cerebras, groq, matx has promise. i think they'll all get there.plus there's plenty to squeeze out of gpu's yet with more efficient scheduling.
>>535149458>GIVE US GOVERNMENT MONEY OR THE CHINAMEN WILL WIN!
>>535149470>>democracies>>USAamerica is a republic
>>535149495>The true peak of civilization was 1890-1910The titanic was sunk in 1912 to kill anyone who opposed the creation of the federal reserveOne year after the sinking of titanic, they created the federal reserve to scam Americans
>>535149458>Democracy
>>535149480>Democracy is rule by Jews.Jews was also behind the assassination of JFK
>>535149480>The only truth is that the AI bubble will burst soon.a-any d-day now!Nope, current AI is already useful enough to justify the investment. It's better than jeets at coding, and that is enough. At worst, there might be a mini-burst, a Dot Com lite. But AI is here to stay.
>>535149487>imagedamn, the funniest thing could happen...
>>535149497> being able to read the reasoning traces is no longer a safety issue.U forgot to say sike
>>535149501>>GIVE US GOVERNMENT MONEY OR THE CHINAMEN WILL WIN!Are they wrong?
>>535149496The answer being no (for the west) is pretty much set in stone.If we hit a wall on progress the reasoning is obvious.The less obvious path is if we dont hit a wall. The problem in that scenario is two-fold: token economics favors vector space efficiency - so smaller languages like chinese are guaranteed to win in the long term with the current paradigm and at a much lower investment cost at every step. Second, as you mention people can run these things locally. Not only does that mean people can opt not to pay, but people can self-host foundation instances on margin and undercut the big boys so long as they have a financial backer to cover request inference cost. At that point its a race to the bottom.Theres no path to profit. Well, maybe theres some if we hit the wall. Then its probably possible to endlessly sell hope to the addicts.
>>535149458>democracy
>>535149480>The only truth is that the AI bubble will burst soon.mindraped
>>535149458>democracy>AmericaKEK
>>535149505Deserved.
>authoritarian>regime>see see peeThey love those words even when they don't know what they mean
>>535149505>Jews was also behind the assassination of JFKJews was also behind the CPC (Communist Party of China), bolshevism, communism, transgenderism, AIPAC, mass media, Hollywood, etc
>>535149458kek
>>535149473
>>535149459>>535149458Sure, I can agree to some of these points. But the fact of the matter is most Chinese labs are getting Engrams and batch invariance in the next 12 months. It's literally game over. You still have time to short.
>>535149509Usually.
where you gay niggas come from?
>>535149458democracy is brown coded
>>535149458He's mad that he wasnt invited to the trip. No one takes this fearmongering seriously. China is open sourcing their models, you should do the same.
>>535149578this was on /g/
>>535149458>In this world, AI norms and rules are shaped by authoritarian regimes, and the best models enable automated repression at scale.That's the US and its lackey states, however.
>>535149458All this talk about "AI leadership" is so vague, it doesnt sound like vision, it sounds like desperation. Leader ship this and adaption that. All these big tech companies went full retard and all in because they were desperate for some new growth opportunities. If you have listened to Bezos and pals, you can kind of see, that their fever dream is, that AI and data centers are utlitity like electricity for future. They believe that people become so dependent to AI that that it will be utility that you cant live without.I just see that as bonkers.If you would build your business so that it is dependent on AI and services provided by them, you have no moat, especially companies that are laying of people left and right. They are gettind rid of real experience and they are hoping to replace it with information compression machine, ie LLMs. But that is not competitive advantage. Everyone has access to these models, there is no advantage if you deeply dependent on them. And think of the implications that your company is dependent on some utlity AI, your business operations crawls to halt the moment there is service disruption. Your workers cant do anything if they can only do work with AI. Or what if there is some political issues, your company is black listed as having wrong ideas, and then you are cut off from the services. Would you like to build business on something like that?The moment china can produce GPUs that are good enough and models that good enough, that business can run onsite locally, the centralized greedy fever dream business models of western oligarch don't sound so enticing. You just can minimize all the risks mentioned and risks of ever increasing costs and own your stuff if you can have things locally.
>>535149459>PRC labs benefit from access to export-controlled American chips via smuggling and foreign data centersOverstated, they barely need our chips anymore. Our chips are primitive trash they have long surpassed and we need to stop blowing smoke up our own asses, our products are garbage
