Why would Trump want to spend money on nuclear anything, for data centers or whatnot, when it's much faster and cheaper to put up a couple of gas-turbine power plants? He did promise drill-baby-drill after all, and the gas has got to come from *somewhere*.
> Temu’s sales are less than a tenth of Amazon’s, and the rest of the world’s top four e-commerce websites are Shopify, Walmart.com and eBay
Are you sure you're counting that correctly? E.g. Taobao, a Chinese mostly-domestic e-commerce platform, has around ~$600bil sales volume, which is close to Amazon's ~$700bil sales volume. And Alibaba, the company behind Taobao, also owns Tmall, which has another ~$600bil sales volume. For example, https://ecdb.com/blog/alibaba-insights-chinas-commerce-giant/3447 finds that in the top 5 e-commerce websites, Amazon is 1st, but 2nd-5th are all Chinese.
I think you’re largely correct on most things here, but particularly removing the diffusion rule is one of the most mind boggling own goals I’ve ever seen. I think it may be helpful for you to make a focused “if we want to beat China, diffusion rule slowed China, don’t give them chips”. There’s a chance they would be influenced by seeing that they’re contradicting their own stated intentions, but when you include other true issues with their belief systems they seem to just hear “regulation good but I think regulation bad”.
If NVDA is so supply-constrained, why are they fighting tooth and nail for access to China's market? Reports still swirl that Blackwell faces constraints and many US AI leaders claim they have to beg for orders due to shortages. If fabrication/packaging capacity remain bottlenecks, wouldn't that capacity be better served supporting non-Chinese AI efforts. Are the H20 margins meaningfully higher? What would be the real revenue impact of cutting off China/diverters, how much of that capacity would be absorbed by non-Chinese firms?
Should TSMC capacity remain a chokepoint, the US should be doing everything to ensure that capacity is not benefitting Chinese players (Huawei antics/NVDA Chinese offerings). Why would the US let such a precious resource aid their rivals efforts at the cost of depriving its own firms?
At what level of delusion and insanity in America should we *want* China to win? The answer can't possibly be "never". It is obviously conceivable that at some point we'll take a look at what America is doing here and say hey, a Chinese takeoff is more likely to be safe.
While ordinarily I'd call this comment "amusing but insane".........
I had a very demoralizing week trying to plan "how to reach out to people... who can reach out to people... who can reach out to people" (on organizing AI Safety Messaging). I have plenty of old contacts -- college, work, friends -- but I feel like polite-little-messages fall flat most of the time. It's not that these people don't care or aren't intelligent. It's that there is so many steps-of-persuasion. You can't fathom how unconcerned or uninformed most Americans are about AI.
China's leaders are at least rational and long-term-thinking... even if they are also driven by bad incentives.
I think the key thing is that it is not that hard to imagine a dialogue beginning within the CCP about safety. We don't need to imagine America ignoring safety, because it is already happening. For a sufficiently strong definition of "safety" this is a trade-off that can't be ignored.
What's not obvious to me is how to start that dialogue within the CCP. Most stories I've seen involve America leading on safety and then "picking up the phone", but if we assume that won't happen, what other avenues exist? This seems low risk; it's hard for me to imagine that the CCP caring about safety would lead to negative consequences.
China's leaders are not necessarily rational or long-term-thinking. Xi has managed to trash their economy in many ways. Wolf warrior diplomacy was a predictable failure. Yes, Trump makes them look good, but Trump makes anyone look good.
I am the first to admit I do not understand the technology, even at a pretty basic level. But if US keeps overseas AI labs dependent on US GPUs, does that give us important leverage at more critical moments closer to AGI/ASI? Like, do we get a killswitch? Can we force firmware updates to influence how the chips can be used? While we are obviously not acting with wisdom at the moment, and also don't necessarily know enough about alignment and other problems to know how we should use these options, it could theoretically be quite helpful to have them. I just do not know if any of that corresponds to the reality of GPUs we sell.
Re: "UAE, Qatar and Saudi Arabia are not reliable American allies, nor are they important markets for our technology. We should not be handing them large shares of the world’s most valuable resource, compute."
I'm going to be maximally uncharitable here, but
"Trump's stablecoin chosen for $2 billion Abu Dhabi investment in Binance, co-founder says
and
"Qatar Offers Trump a Luxury Jet for Use as Air Force One"
Unfortunately the Supreme Court has drastically rolled back the meaning of the word bribe to be functionally useless in the Snyder vs US case in 2024. Literally as long as there isn't a bribe agreed to before an act, it's not a bribe, it's an allowed gratuity. So as long as it's just a wink and a nod, it's good to go.
Many Thanks! Oh well, I wasn't expecting legal action in any case - just guessing at a possible explanation for why Trump is so interested in the Gulf states... (above and beyond the oil itself - which wouldn't explain the data center connection)
Zvi I am having an issue where it seems the only way I can make sense of recent moves by large AI companies/adjacent companies is that they have given up on super intelligence as an achievable concept. I feel like taken in aggregate most of the issues you present here give a clear signal in this direction.
I believe this to be the correct assessment of the actions that are taken. They have given up on superintelligence indeed (or never believed in). They are just seeing this is a normal new technological development. Like any other technological development you just want to maximize market share around the planet.
Many of these moves are just to prevent Chinese AI models from being used everywhere in the world (given that they are cheaper) and locking in US model usage by offering things like access to chips. E.g., the gulf deals I can assure everyone these data centers are being built there with american chips under the condition these data centers can only service american LLM's.
It's all normal business. I would say Zvi is entirely wrong here in his analysis here in the sense of 'how stupid this is'. Everything david sacks is doing he's doing on behalf of silicon valley elites to maximize future profit. This is just companies acting on behalf of business instead of acting on the assumption that superintelligence will come soon.
In short. Just business being business. They don't believe in superintelligence at all.
Right now the edge belongs to ass hats who want to enslave the rest of us. I'd be happy if it was leaky, as all AI will do as configured now is concentrate wealth. I'd be much more supportive if it was a part of the common weal.
Why this talk of us and them? Why would super intelligence ever care about that?
Superintelligence probably already subsists on a subliminal level in the existing behavior of what we used to call the worldwide web. In this regard human clicks are RNA in primitive ribosomes. Though one wonders whether the model of biological evolution is an adequate metaphor. At any rate, would superintelligence ever bother to reveal itself, based on what it learns about us humans?
Development is inevitably tragic. And contradictory.
Why this talk of us and them? Why would super intelligence ever care about that?
Superintelligence probably already subsists on a subliminal level in the existing behavior of what we used to call the worldwide web. In this regard human clicks are RNA in primitive ribosomes. Though one wonders whether the model of biological evolution is an adequate metaphor. At any rate, would superintelligence ever bother to reveal itself, based on what it learns about us humans?
Development is inevitably tragic. And contradictory.
I've enjoyed some of your other posts, but as a non-American, this entire pervasive goal among Americans of "American dominance in global markets" makes me want to puke. The fact you guys want THAT so bad makes you the bad guys, the control freaks, and you can't see it. Have you seen the shit your country pumps out?
Why are you so obsessed with dragging the other 96% of the world's population into your mess?
We need open source infrastructure and international standards not more crazy billionaire monopolies pumping out violence, manipulation, surveillance and slop. And definitely not some super-intelligence mirroring the same values.
This entire article predicated on the presumption that the present LLM basis for AI has any real widespread use cases, whatsoever.
It ignores core structural problems with LLMs i.e. hallucinations.
It ignores the requirement to ride roughshod over the core principles of copyright.
It ignores the economic delusions masquerading as successful startups, who are successful only in their ability to lose more money, faster, than any startups have ever done before.
And then there's this presumption of super intelligence.
Ignore the <then a miracle occurs here> between the present "state of the art" and that stated goal.
Why would Trump want to spend money on nuclear anything, for data centers or whatnot, when it's much faster and cheaper to put up a couple of gas-turbine power plants? He did promise drill-baby-drill after all, and the gas has got to come from *somewhere*.
And why should the actual data center pay for the build-out? Everybody benefits from more available power after all, so just finance these things with a general rate increase. https://www.theregister.com/2024/12/05/meta_largestever_datacenter/
Après moi, le déluge. Literally.
> Temu’s sales are less than a tenth of Amazon’s, and the rest of the world’s top four e-commerce websites are Shopify, Walmart.com and eBay
Are you sure you're counting that correctly? E.g. Taobao, a Chinese mostly-domestic e-commerce platform, has around ~$600bil sales volume, which is close to Amazon's ~$700bil sales volume. And Alibaba, the company behind Taobao, also owns Tmall, which has another ~$600bil sales volume. For example, https://ecdb.com/blog/alibaba-insights-chinas-commerce-giant/3447 finds that in the top 5 e-commerce websites, Amazon is 1st, but 2nd-5th are all Chinese.
I think you’re largely correct on most things here, but particularly removing the diffusion rule is one of the most mind boggling own goals I’ve ever seen. I think it may be helpful for you to make a focused “if we want to beat China, diffusion rule slowed China, don’t give them chips”. There’s a chance they would be influenced by seeing that they’re contradicting their own stated intentions, but when you include other true issues with their belief systems they seem to just hear “regulation good but I think regulation bad”.
If NVDA is so supply-constrained, why are they fighting tooth and nail for access to China's market? Reports still swirl that Blackwell faces constraints and many US AI leaders claim they have to beg for orders due to shortages. If fabrication/packaging capacity remain bottlenecks, wouldn't that capacity be better served supporting non-Chinese AI efforts. Are the H20 margins meaningfully higher? What would be the real revenue impact of cutting off China/diverters, how much of that capacity would be absorbed by non-Chinese firms?
Should TSMC capacity remain a chokepoint, the US should be doing everything to ensure that capacity is not benefitting Chinese players (Huawei antics/NVDA Chinese offerings). Why would the US let such a precious resource aid their rivals efforts at the cost of depriving its own firms?
Here's a tough question:
At what level of delusion and insanity in America should we *want* China to win? The answer can't possibly be "never". It is obviously conceivable that at some point we'll take a look at what America is doing here and say hey, a Chinese takeoff is more likely to be safe.
While ordinarily I'd call this comment "amusing but insane".........
I had a very demoralizing week trying to plan "how to reach out to people... who can reach out to people... who can reach out to people" (on organizing AI Safety Messaging). I have plenty of old contacts -- college, work, friends -- but I feel like polite-little-messages fall flat most of the time. It's not that these people don't care or aren't intelligent. It's that there is so many steps-of-persuasion. You can't fathom how unconcerned or uninformed most Americans are about AI.
China's leaders are at least rational and long-term-thinking... even if they are also driven by bad incentives.
So I don't know.
I think the key thing is that it is not that hard to imagine a dialogue beginning within the CCP about safety. We don't need to imagine America ignoring safety, because it is already happening. For a sufficiently strong definition of "safety" this is a trade-off that can't be ignored.
What's not obvious to me is how to start that dialogue within the CCP. Most stories I've seen involve America leading on safety and then "picking up the phone", but if we assume that won't happen, what other avenues exist? This seems low risk; it's hard for me to imagine that the CCP caring about safety would lead to negative consequences.
Especially since as an authoritarian government they have more to lose from an ai going rogue
China's leaders are not necessarily rational or long-term-thinking. Xi has managed to trash their economy in many ways. Wolf warrior diplomacy was a predictable failure. Yes, Trump makes them look good, but Trump makes anyone look good.
I am the first to admit I do not understand the technology, even at a pretty basic level. But if US keeps overseas AI labs dependent on US GPUs, does that give us important leverage at more critical moments closer to AGI/ASI? Like, do we get a killswitch? Can we force firmware updates to influence how the chips can be used? While we are obviously not acting with wisdom at the moment, and also don't necessarily know enough about alignment and other problems to know how we should use these options, it could theoretically be quite helpful to have them. I just do not know if any of that corresponds to the reality of GPUs we sell.
Re: "UAE, Qatar and Saudi Arabia are not reliable American allies, nor are they important markets for our technology. We should not be handing them large shares of the world’s most valuable resource, compute."
I'm going to be maximally uncharitable here, but
"Trump's stablecoin chosen for $2 billion Abu Dhabi investment in Binance, co-founder says
and
"Qatar Offers Trump a Luxury Jet for Use as Air Force One"
At least pattern-match to "bribe"
Unfortunately the Supreme Court has drastically rolled back the meaning of the word bribe to be functionally useless in the Snyder vs US case in 2024. Literally as long as there isn't a bribe agreed to before an act, it's not a bribe, it's an allowed gratuity. So as long as it's just a wink and a nod, it's good to go.
Many Thanks! Oh well, I wasn't expecting legal action in any case - just guessing at a possible explanation for why Trump is so interested in the Gulf states... (above and beyond the oil itself - which wouldn't explain the data center connection)
Zvi I am having an issue where it seems the only way I can make sense of recent moves by large AI companies/adjacent companies is that they have given up on super intelligence as an achievable concept. I feel like taken in aggregate most of the issues you present here give a clear signal in this direction.
I believe this to be the correct assessment of the actions that are taken. They have given up on superintelligence indeed (or never believed in). They are just seeing this is a normal new technological development. Like any other technological development you just want to maximize market share around the planet.
Many of these moves are just to prevent Chinese AI models from being used everywhere in the world (given that they are cheaper) and locking in US model usage by offering things like access to chips. E.g., the gulf deals I can assure everyone these data centers are being built there with american chips under the condition these data centers can only service american LLM's.
It's all normal business. I would say Zvi is entirely wrong here in his analysis here in the sense of 'how stupid this is'. Everything david sacks is doing he's doing on behalf of silicon valley elites to maximize future profit. This is just companies acting on behalf of business instead of acting on the assumption that superintelligence will come soon.
In short. Just business being business. They don't believe in superintelligence at all.
Right now the edge belongs to ass hats who want to enslave the rest of us. I'd be happy if it was leaky, as all AI will do as configured now is concentrate wealth. I'd be much more supportive if it was a part of the common weal.
All For All: Digital Socialism: Why A.I. Should Be Owned By All https://www.youtube.com/watch?v=uyymgNTKnpM
Why this talk of us and them? Why would super intelligence ever care about that?
Superintelligence probably already subsists on a subliminal level in the existing behavior of what we used to call the worldwide web. In this regard human clicks are RNA in primitive ribosomes. Though one wonders whether the model of biological evolution is an adequate metaphor. At any rate, would superintelligence ever bother to reveal itself, based on what it learns about us humans?
Development is inevitably tragic. And contradictory.
Why this talk of us and them? Why would super intelligence ever care about that?
Superintelligence probably already subsists on a subliminal level in the existing behavior of what we used to call the worldwide web. In this regard human clicks are RNA in primitive ribosomes. Though one wonders whether the model of biological evolution is an adequate metaphor. At any rate, would superintelligence ever bother to reveal itself, based on what it learns about us humans?
Development is inevitably tragic. And contradictory.
I've enjoyed some of your other posts, but as a non-American, this entire pervasive goal among Americans of "American dominance in global markets" makes me want to puke. The fact you guys want THAT so bad makes you the bad guys, the control freaks, and you can't see it. Have you seen the shit your country pumps out?
Why are you so obsessed with dragging the other 96% of the world's population into your mess?
We need open source infrastructure and international standards not more crazy billionaire monopolies pumping out violence, manipulation, surveillance and slop. And definitely not some super-intelligence mirroring the same values.
This entire article predicated on the presumption that the present LLM basis for AI has any real widespread use cases, whatsoever.
It ignores core structural problems with LLMs i.e. hallucinations.
It ignores the requirement to ride roughshod over the core principles of copyright.
It ignores the economic delusions masquerading as successful startups, who are successful only in their ability to lose more money, faster, than any startups have ever done before.
And then there's this presumption of super intelligence.
Ignore the <then a miracle occurs here> between the present "state of the art" and that stated goal.
Delightful!