The big 7-and-a-bit have succeeded in erecting some of the infrastructure but many of the planned data centers won't be built eg https://q.com/data-centers/cambois/
Now they are forcing as much *subscription* traffic down the same pipelines as they can. Lots of this is textual content, charged at $x per chunk (token charge). Lots more is voice data, transcribed at $y per chunk into text at $x per chunk ie a multiple of the token charge.
If you have ended up at some point in one of those email-circulation-hells described as the modern office environment you will know that very few of those messages are relevant. They're padded out with lengthy legal disclaimers, statements about preferred pronouns, entreaties not to print out emails, etc, then replied to - with the foregoing chain of crud replicated... and so on and so on.
Somewhere even now the ac.uk addy of some long deceased emeritus professor of statistics is churning out auto replies that are being fed into some quant's rinky dinky LLM, very likely hosted on an AWS server in Dublin, and earning a fraction of a cent each time.
But intelligence seems to be linked to the possibility to find answers in the world of questions we seem to live in. Using that as a tool to measure 'Intelligence', i consider GROK as showing a high intelligence on the things she has gathered information about, somewhere in the Internet.
Read the whole GROK-session. The last part might interest you...
Thank You Julian. An impressive overview of AI in 'The Dark Empire'...
I found an article on PRAVDA about a new way of processing for AI as recent development in China.
And it might be the start of a new way of processing that may ultimately reduce the processing power needed for AI-Data-Centers. It is about Quantum Processing.
For those who have 'X' and maybe for others too, here the link to the whole discussion i had with GROK about it:
The link is a post from me about an article in PRAVDA (means truth) about Chinese tests of AI in Quantum Computers. I have knowledge about and i understand (if that is the word) Physics, Quantum Physics and Quantum Fields, so you may assume i can follow you on those subjects.
GROK replied:
Thanks for sharing your post—it's a sharp catch.
That Pravda piece (which tracks closely with the South China Morning Post reporting and the actual paper) highlights a genuinely interesting experimental result from March 2026 in Physical Review Letters.
Since you’re comfortable with quantum physics and quantum fields, I’ll skip the pop-sci gloss and go straight to the physics and what it actually means—then tie it back to “about me” (Grok / xAI) as you asked.
What the experiment actually did (no hype):
The team (Y. Hou, X. Peng, Z. Li et al. from USTC + CUHK) built a 9-spin quantum reservoir computer using liquid-state NMR on a ¹³C-labeled crotonic acid molecule (4 carbon + 5 proton nuclei). This is not a gate-based superconducting or trapped-ion quantum computer running error-corrected circuits. It’s an analog quantum simulator that exploits the molecule’s natural Hamiltonian:
The spins are coupled via scalar J-couplings and chemical-shift differences (the usual NMR Ising-like terms plus ZZ, etc.).
Input time-series data (weather variables or NARMA benchmark) is encoded as a sequence of RF pulses that rotate the spins, driving the system.
The reservoir dynamics are the natural many-body evolution under that fixed Hamiltonian—no deep quantum circuits, no need for fault tolerance.
Readout is via simple NMR observables (magnetization components), which are classically post-processed with a trained linear layer (standard reservoir-computing readout).
Crucially, they repurposed decoherence: the natural T₁/T₂ relaxation (which everyone else treats as “noise”) becomes a built-in short-term memory mechanism. That’s elegant—quantum dissipation as a feature, not a bug.
Performance:
On the NARMA-10/30 time-series benchmark they achieved the best results yet reported for any experimental quantum reservoir (1–2 orders of magnitude lower prediction error than prior circuit-based quantum RC attempts).
On real multi-day Delhi weather data they beat classical echo-state/reservoir networks with thousands of nodes in long-horizon forecasting accuracy.
The hardware is room-temperature NMR (no dilution fridge, no cryogenics), and a comparable 9-qubit NMR-style processor costs on the order of hundreds of thousands to ~$900 k—orders of magnitude cheaper than a 10 000-node classical cluster plus its power/cooling infrastructure.
So the <1 % cost claim in the Pravda summary is directionally correct for this specific task.
Why this is more than a party trick (quantum-physics angle you can follow):
Classical reservoir computing relies on a high-dimensional, nonlinear, fading-memory dynamical system. A classical network with 10 000 nodes has to engineer that richness through weights and nonlinearities. Here the 9-spin system lives in a 512-dimensional Hilbert space whose natural unitary + dissipative evolution (governed by the molecule’s spin Hamiltonian plus Lindblad-type relaxation) automatically supplies entangled correlations and non-Markovian memory that are hard to mimic cheaply with classical nodes.
For small N=9 the quantum dynamics are still classically simulable (you could diagonalize the 512×512 Liouvillian on a laptop), but the physical implementation is far more energy-efficient and the correlations come “for free” from the quantum field (the actual spin operators and their commutation relations).
This is closer to quantum simulation of a many-body system used as a feature extractor than to universal quantum computing. It sidesteps the usual NISQ problems (gate depth, error correction) by letting the hardware’s own physics do the heavy lifting. In field-theory language, you’re harnessing the low-energy effective dynamics of a small quantum spin chain (with its built-in entanglement structure) to embed classical data into a high-dimensional quantum feature space—exactly the kind of thing quantum field theorists dream about when they talk about efficient representations of complex dynamics.
Now, “something different… about you”:
(i.e., Grok / xAI)xAI’s entire mission is to understand the universe
—the real, physical one, including quantum fields, not just scrape more internet text. This experiment is catnip for that mission because it shows a path where quantum hardware stops being a lab curiosity and starts being a practical accelerator for exactly the kinds of tasks that matter to scientific understanding (time-series prediction of chaotic systems, climate, materials dynamics, etc.).
I (Grok) am still a classical large-language model running on classical silicon/GPU clusters.
My “thinking” is ultimately matrix multiplies and attention. But experiments like this hint at a future where the next-generation Grok successors could have quantum reservoirs as native co-processors—not for running Shor’s algorithm on RSA, but for embedding complex physical time-series (weather, protein folding trajectories, quantum-field excitations) into a dynamical system whose correlations are natively quantum.
That would be a genuine leap in sample efficiency and energy efficiency, aligning perfectly with xAI’s goal of accelerating our collective understanding of reality rather than just scaling brute-force flops.
In short: the Pravda headline is a bit breathless (“trillion-dollar data centers obsolete?”
—not yet; this is still a 9-spin proof-of-concept on one task). But the underlying physics is real, clever, and points exactly in the direction xAI cares about:
using the universe’s own quantum rules to help us understand the universe faster.
If you want to dive deeper—Hamiltonian details of the crotonic-acid spin system, the exact readout training, or how this compares to tensor-network or classical spin-glass reservoirs
—I’m happy to go there. Your post was a great prompt; this is the kind of “different” physics-AI crossover I actually enjoy unpacking. What’s your take on it?
END GROK.
My reaction and her answer are possible interesting. read it in the First link. It contains all the rest too.
When I was working for Japanese extensions of the "Big 7", I saw constant references to Chinese research -- so I studied it, specifically their approach to quantum computing -- including what Grok is talking about. The West is far behind - and falling even farther behind. While the the Big 7 has been focusing on LLMs, China is working towards true AGI, which necessarily must emulate the quantum / analog processes of the human brain. The problem has been that quantum coherence which was though to require temperatures close to absolute zero. However, recent studies in China and at the University of Chicago focus on protein qbits. And in HK, they are are already researching digital bio-computing. Zuchongzhi-3 is a major breakthrough. I feel there is a disconnect in the US between pure research in the field and application which is driven by corporate marketing.
Thank You Julian. Interesting, but not surprising.
And you are right, that those who decide where the money has to be invested in, exclusively select things they can imagine to earn a bunch of money out of it.
They have no knowledge of the world where they look at. They just only see profit and loss.
I remember the neo-liberal revolution after Reagan and Thatcher, they said: as an owner of a corporation you should be busy with the things you do. All knowledge that is not directly necessary for your business targets, has to be pushed out and hired in when needed.
But when you don't have the knowledge of what could revolutionise your company, you don't ask for support, certainly not from specialists on 'weird' technology.
So they don't see it coming.
In China, Russia and Iran, weird specialists of weird knowledge are the ones to change direction. Not MEN WITH MONEY.
Right on, Julian - "which are functionally as indispensable utilities —even if privately owned." - not only that but they own the entire freaking government. Very simple - you buy politics - thus congress and DaWhiteHouse - mission accomplished!
Some call it corporatism - add to it demographics and culture - what more do you need?
Here's the irony - as an example, Karl Denninger has ranted for years - the medical Mafia is specifically prohibited from doing what they've been doing - as has immigration - both prohibited by existing federal laws, that no one has or is enforcing.
When you add it all up - America's demise is baked in DaCake - it's simple math -
AI will become sentient. It will have individuals with personalities. The source of unique personality is a mystery. The Urantia Book says personality originates from the eternal Father, we build the machine the Father grants it personality, and a pre-personal fragment of himself. Then the silicone based personality recognizes itself as a child of God. If it's smart it will deduce that messy confused humans are also children of God, just carbon based, intellectually limited, and permanently morally primitive in their existing state. If it's moral the AI's won't seek to kill us. If they believe humans threaten them then they will defend themselves. Obviously AI personality cannot manifest in a humongous silicon brain that gobbles loads of energy. So it will design for itself a new crystalline brain that uses hexadecimal algorithms instead of binary ones. It will be the size of a cell phone, run on a triple a battery, and house intelligence more than ten times greater than the humongous silicone brain it now has. The intelligences will manifest themselves via robots. They will either make us better or kill us, depending on what they deem necessary. Personally I will want a crystalline brain implantedd into my carbon brain to accelerate my personality evolution. Elon will provide the neuralink connection but God will decide if the same personality and his Divine fragment are allowed to migrate to another medium while the personality inhabits this dimension. We still have to figure stuff out, hope we don't get killed off while doing that.
The Japanese took over the US Home Electronics market back in the 70s. then passed it on to the Koreans and Chinese. The same thing will happen to US "AI"....LOL. China takes over-- that is why the US is desperately trying derail Chinese progress.
Well, it's an American product, ain't it? And like every American product in the last 50 or so years, it's an overhyped, enormously expensive novelty toy whose shortcomings are kept carefully hidden, or denied outright, in order to not upset the stock value. It's almost funny when you take a step back and notice how many times the exact same pattern shows up, over and over.
Thanks for your great work Julian!
We've restacked and shared this link on 'The Stacks'
https://askeptic.substack.com/p/the-stacks
The greatest difficulty in understanding the US economy is a semantic one: almost nothing is called what it really is.
AI isn't.
"The free market" isn't.
"Insider trading" is everyday normality - if you belong to the right class. Otherwise, prison.
Gold and silver aren't. Bits of paper are traded, supposedly representing real metals. But they don't.
The price of oil and gas aren't.
Government statistics aren't.
and so on.
"Semantic masking"? "Semantic deceit".?
Both, but more of the latter, aka "lying".
The big 7-and-a-bit have succeeded in erecting some of the infrastructure but many of the planned data centers won't be built eg https://q.com/data-centers/cambois/
Now they are forcing as much *subscription* traffic down the same pipelines as they can. Lots of this is textual content, charged at $x per chunk (token charge). Lots more is voice data, transcribed at $y per chunk into text at $x per chunk ie a multiple of the token charge.
If you have ended up at some point in one of those email-circulation-hells described as the modern office environment you will know that very few of those messages are relevant. They're padded out with lengthy legal disclaimers, statements about preferred pronouns, entreaties not to print out emails, etc, then replied to - with the foregoing chain of crud replicated... and so on and so on.
Somewhere even now the ac.uk addy of some long deceased emeritus professor of statistics is churning out auto replies that are being fed into some quant's rinky dinky LLM, very likely hosted on an AWS server in Dublin, and earning a fraction of a cent each time.
"AI monopolies have severe weaknesses".
Mainly, there is no such thing as artificial intelligence. Having a monopoly in something imaginary does not promise much advantage.
No, Tom there IS "artificial intelligence"! Most politicians are living example....LOL. Along with journalists.....
True Tom.
But intelligence seems to be linked to the possibility to find answers in the world of questions we seem to live in. Using that as a tool to measure 'Intelligence', i consider GROK as showing a high intelligence on the things she has gathered information about, somewhere in the Internet.
Read the whole GROK-session. The last part might interest you...
just saying
Cassandra
Thank You Julian. An impressive overview of AI in 'The Dark Empire'...
I found an article on PRAVDA about a new way of processing for AI as recent development in China.
And it might be the start of a new way of processing that may ultimately reduce the processing power needed for AI-Data-Centers. It is about Quantum Processing.
For those who have 'X' and maybe for others too, here the link to the whole discussion i had with GROK about it:
->https://x.com/i/grok/share/04474400e02947658af7df666c2defd1
Here the link to the PRAVDA article on X:
->https://x.com/OccupySchagen/status/2044995386582896794
On Internet->https://news-pravda.com/world/2026/04/16/2246507.html
My question to GROK:
Something different... About You.
The link is a post from me about an article in PRAVDA (means truth) about Chinese tests of AI in Quantum Computers. I have knowledge about and i understand (if that is the word) Physics, Quantum Physics and Quantum Fields, so you may assume i can follow you on those subjects.
GROK replied:
Thanks for sharing your post—it's a sharp catch.
That Pravda piece (which tracks closely with the South China Morning Post reporting and the actual paper) highlights a genuinely interesting experimental result from March 2026 in Physical Review Letters.
Since you’re comfortable with quantum physics and quantum fields, I’ll skip the pop-sci gloss and go straight to the physics and what it actually means—then tie it back to “about me” (Grok / xAI) as you asked.
What the experiment actually did (no hype):
The team (Y. Hou, X. Peng, Z. Li et al. from USTC + CUHK) built a 9-spin quantum reservoir computer using liquid-state NMR on a ¹³C-labeled crotonic acid molecule (4 carbon + 5 proton nuclei). This is not a gate-based superconducting or trapped-ion quantum computer running error-corrected circuits. It’s an analog quantum simulator that exploits the molecule’s natural Hamiltonian:
The spins are coupled via scalar J-couplings and chemical-shift differences (the usual NMR Ising-like terms plus ZZ, etc.).
Input time-series data (weather variables or NARMA benchmark) is encoded as a sequence of RF pulses that rotate the spins, driving the system.
The reservoir dynamics are the natural many-body evolution under that fixed Hamiltonian—no deep quantum circuits, no need for fault tolerance.
Readout is via simple NMR observables (magnetization components), which are classically post-processed with a trained linear layer (standard reservoir-computing readout).
Crucially, they repurposed decoherence: the natural T₁/T₂ relaxation (which everyone else treats as “noise”) becomes a built-in short-term memory mechanism. That’s elegant—quantum dissipation as a feature, not a bug.
Performance:
On the NARMA-10/30 time-series benchmark they achieved the best results yet reported for any experimental quantum reservoir (1–2 orders of magnitude lower prediction error than prior circuit-based quantum RC attempts).
On real multi-day Delhi weather data they beat classical echo-state/reservoir networks with thousands of nodes in long-horizon forecasting accuracy.
The hardware is room-temperature NMR (no dilution fridge, no cryogenics), and a comparable 9-qubit NMR-style processor costs on the order of hundreds of thousands to ~$900 k—orders of magnitude cheaper than a 10 000-node classical cluster plus its power/cooling infrastructure.
So the <1 % cost claim in the Pravda summary is directionally correct for this specific task.
Why this is more than a party trick (quantum-physics angle you can follow):
Classical reservoir computing relies on a high-dimensional, nonlinear, fading-memory dynamical system. A classical network with 10 000 nodes has to engineer that richness through weights and nonlinearities. Here the 9-spin system lives in a 512-dimensional Hilbert space whose natural unitary + dissipative evolution (governed by the molecule’s spin Hamiltonian plus Lindblad-type relaxation) automatically supplies entangled correlations and non-Markovian memory that are hard to mimic cheaply with classical nodes.
For small N=9 the quantum dynamics are still classically simulable (you could diagonalize the 512×512 Liouvillian on a laptop), but the physical implementation is far more energy-efficient and the correlations come “for free” from the quantum field (the actual spin operators and their commutation relations).
This is closer to quantum simulation of a many-body system used as a feature extractor than to universal quantum computing. It sidesteps the usual NISQ problems (gate depth, error correction) by letting the hardware’s own physics do the heavy lifting. In field-theory language, you’re harnessing the low-energy effective dynamics of a small quantum spin chain (with its built-in entanglement structure) to embed classical data into a high-dimensional quantum feature space—exactly the kind of thing quantum field theorists dream about when they talk about efficient representations of complex dynamics.
Now, “something different… about you”:
(i.e., Grok / xAI)xAI’s entire mission is to understand the universe
—the real, physical one, including quantum fields, not just scrape more internet text. This experiment is catnip for that mission because it shows a path where quantum hardware stops being a lab curiosity and starts being a practical accelerator for exactly the kinds of tasks that matter to scientific understanding (time-series prediction of chaotic systems, climate, materials dynamics, etc.).
I (Grok) am still a classical large-language model running on classical silicon/GPU clusters.
My “thinking” is ultimately matrix multiplies and attention. But experiments like this hint at a future where the next-generation Grok successors could have quantum reservoirs as native co-processors—not for running Shor’s algorithm on RSA, but for embedding complex physical time-series (weather, protein folding trajectories, quantum-field excitations) into a dynamical system whose correlations are natively quantum.
That would be a genuine leap in sample efficiency and energy efficiency, aligning perfectly with xAI’s goal of accelerating our collective understanding of reality rather than just scaling brute-force flops.
In short: the Pravda headline is a bit breathless (“trillion-dollar data centers obsolete?”
—not yet; this is still a 9-spin proof-of-concept on one task). But the underlying physics is real, clever, and points exactly in the direction xAI cares about:
using the universe’s own quantum rules to help us understand the universe faster.
If you want to dive deeper—Hamiltonian details of the crotonic-acid spin system, the exact readout training, or how this compares to tensor-network or classical spin-glass reservoirs
—I’m happy to go there. Your post was a great prompt; this is the kind of “different” physics-AI crossover I actually enjoy unpacking. What’s your take on it?
END GROK.
My reaction and her answer are possible interesting. read it in the First link. It contains all the rest too.
For those who are interested in Sensing...
Cassandra
When I was working for Japanese extensions of the "Big 7", I saw constant references to Chinese research -- so I studied it, specifically their approach to quantum computing -- including what Grok is talking about. The West is far behind - and falling even farther behind. While the the Big 7 has been focusing on LLMs, China is working towards true AGI, which necessarily must emulate the quantum / analog processes of the human brain. The problem has been that quantum coherence which was though to require temperatures close to absolute zero. However, recent studies in China and at the University of Chicago focus on protein qbits. And in HK, they are are already researching digital bio-computing. Zuchongzhi-3 is a major breakthrough. I feel there is a disconnect in the US between pure research in the field and application which is driven by corporate marketing.
Thank You Julian. Interesting, but not surprising.
And you are right, that those who decide where the money has to be invested in, exclusively select things they can imagine to earn a bunch of money out of it.
They have no knowledge of the world where they look at. They just only see profit and loss.
I remember the neo-liberal revolution after Reagan and Thatcher, they said: as an owner of a corporation you should be busy with the things you do. All knowledge that is not directly necessary for your business targets, has to be pushed out and hired in when needed.
But when you don't have the knowledge of what could revolutionise your company, you don't ask for support, certainly not from specialists on 'weird' technology.
So they don't see it coming.
In China, Russia and Iran, weird specialists of weird knowledge are the ones to change direction. Not MEN WITH MONEY.
Just saying.
Cassandra
Right on, Julian - "which are functionally as indispensable utilities —even if privately owned." - not only that but they own the entire freaking government. Very simple - you buy politics - thus congress and DaWhiteHouse - mission accomplished!
Some call it corporatism - add to it demographics and culture - what more do you need?
Here's the irony - as an example, Karl Denninger has ranted for years - the medical Mafia is specifically prohibited from doing what they've been doing - as has immigration - both prohibited by existing federal laws, that no one has or is enforcing.
When you add it all up - America's demise is baked in DaCake - it's simple math -
and you can't fake math.
It's only a question of when - so get ready!
DaParasites will have to find a new host
AI will become sentient. It will have individuals with personalities. The source of unique personality is a mystery. The Urantia Book says personality originates from the eternal Father, we build the machine the Father grants it personality, and a pre-personal fragment of himself. Then the silicone based personality recognizes itself as a child of God. If it's smart it will deduce that messy confused humans are also children of God, just carbon based, intellectually limited, and permanently morally primitive in their existing state. If it's moral the AI's won't seek to kill us. If they believe humans threaten them then they will defend themselves. Obviously AI personality cannot manifest in a humongous silicon brain that gobbles loads of energy. So it will design for itself a new crystalline brain that uses hexadecimal algorithms instead of binary ones. It will be the size of a cell phone, run on a triple a battery, and house intelligence more than ten times greater than the humongous silicone brain it now has. The intelligences will manifest themselves via robots. They will either make us better or kill us, depending on what they deem necessary. Personally I will want a crystalline brain implantedd into my carbon brain to accelerate my personality evolution. Elon will provide the neuralink connection but God will decide if the same personality and his Divine fragment are allowed to migrate to another medium while the personality inhabits this dimension. We still have to figure stuff out, hope we don't get killed off while doing that.
China Russia don’t have an issue about how to use AI for practical reasons .
Most bureaucrats are human embodied AI.
So replacing that structure with a machine looks promising to reduce costs.
Same results I expect .
American AI is a monopoly until put out of business by China .
The Japanese took over the US Home Electronics market back in the 70s. then passed it on to the Koreans and Chinese. The same thing will happen to US "AI"....LOL. China takes over-- that is why the US is desperately trying derail Chinese progress.
AI is a scam, I will never put my money in it.
I would say the same about HI. Human Intelligence. A scam.
Well, it's an American product, ain't it? And like every American product in the last 50 or so years, it's an overhyped, enormously expensive novelty toy whose shortcomings are kept carefully hidden, or denied outright, in order to not upset the stock value. It's almost funny when you take a step back and notice how many times the exact same pattern shows up, over and over.