Has China Won The AI Race Already?

[John Clark]

The problem is electricity, or rather the lack of it. 

Has China Won The AI Race Already?

John K Clark    See what's on my new list at  Extropolis

31=

[Brent Meeker]

Or what if just more training produces only marginal gains in intelligence.

Brent

--
You received this message because you are subscribed to the Google Groups "Everything List" group.
To unsubscribe from this group and stop receiving emails from it, send an email to everything-li...@googlegroups.com.
To view this discussion visit https://groups.google.com/d/msgid/everything-list/CAJPayv3XMmKuyiq6QgAci_mzjtSBHV630pDsnmjioqAP8U_9KA%40mail.gmail.com.

[John Clark]

On Thu, Aug 21, 2025 at 9:56 PM Brent Meeker <meeke...@gmail.com> wrote:

> what if just more training produces only marginal gains in intelligence.

Then you're obviously using the wrong AI model to educate, and maybe even the wrong basic AI architecture. There are plenty of new architectures that achieve impressive performance with very little training, I talked about one of them, Hierarchical Reasoning Models, in a previous post, and there are others.

John K Clark    See what's on my new list at  Extropolis

ihy

 

Brent

On 8/21/2025 7:35 AM, John Clark wrote:

The problem is electricity, or rather the lack of it. 

Has China Won The AI Race Already?

-

[Brent Meeker]

On 8/22/2025 3:35 AM, John Clark wrote:

On Thu, Aug 21, 2025 at 9:56 PM Brent Meeker <meeke...@gmail.com> wrote:

> what if just more training produces only marginal gains in intelligence.

Then you're obviously using the wrong AI model to educate, and maybe even the wrong basic AI architecture. 

Exactly my point.  You originally wrote, "The problem is electricity, or rather the lack of it."  Having a better AI model bypasses the need for more electricity which is based on making marginal gains by just bigger neural nets and more training.

Brent 

There are plenty of new architectures that achieve impressive performance with very little training, I talked about one of them, Hierarchical Reasoning Models, in a previous post, and there are others.

John K Clark    See what's on my new list at  Extropolis

ihy

 

Brent

On 8/21/2025 7:35 AM, John Clark wrote:

The problem is electricity, or rather the lack of it. 

Has China Won The AI Race Already?

-

--

You received this message because you are subscribed to the Google Groups "Everything List" group.
To unsubscribe from this group and stop receiving emails from it, send an email to everything-li...@googlegroups.com.

To view this discussion visit https://groups.google.com/d/msgid/everything-list/CAJPayv0arfO9E9UGcYrnDGdRsD9szPZPxLoK8fL4w4MOM6b2uQ%40mail.gmail.com.

[John Clark]

On Fri, Aug 22, 2025 at 3:33 PM Brent Meeker <meeke...@gmail.com> wrote:

>>> what if just more training produces only marginal gains in intelligence.

>> Then you're obviously using the wrong AI model to educate, and maybe even the wrong basic AI architecture. 

> Exactly my point.  You originally wrote, "The problem is electricity, or rather the lack of it."  Having a better AI model bypasses the need for more electricity

No it does not! Regardless of how efficient an AI model is, the more computational resources it has available the more intelligent it will be, and the more new computer models you will be able to train. The same basic phenomenon is responsible for the fact that improving the efficiency of cars did not reduce the nation's consumption of gasoline because the less gasoline a car needed the more people drove them. It's called "Jevons Paradox" but it's not really a paradox, it's just an odd situation that can crop up in economics when improvements in technology are involved. 

Jevons paradox

John K Clark    See what's on my new list at  Extropolis

px6

[Brent Meeker]

On 8/22/2025 1:36 PM, John Clark wrote:

On Fri, Aug 22, 2025 at 3:33 PM Brent Meeker <meeke...@gmail.com> wrote:

>>> what if just more training produces only marginal gains in intelligence.

>> Then you're obviously using the wrong AI model to educate, and maybe even the wrong basic AI architecture. 

> Exactly my point.  You originally wrote, "The problem is electricity, or rather the lack of it."  Having a better AI model bypasses the need for more electricity

No it does not! Regardless of how efficient an AI model is, 

Yes it does!  Regardless of how much more electricity you have a more efficient model may be superior.  Races are only one-dimensional.

Brent

[John Clark]

On Fri, Aug 22, 2025 at 9:43 PM Brent Meeker <meeke...@gmail.com> wrote:

>>Having a better AI model bypasses the need for more electricity

 >> No it does not! Regardless of how efficient an AI model is, the more computational resources it has available the more intelligent it will be, and the more new computer models you will be able to train.

> Yes it does!  Regardless of how much more electricity you have a more efficient model may be superior. 

Superior to what? If your AI is not lousy then more computing power will result in more intelligence. If your AI is lousy then more computing power will enable you to find a new AI model, or even a new AI architecture, that is less lousy. So your assertion that a more efficient AI will lead to less demand for computing power is simply not true. 

That's why on January 27 when Nvidia lost $600 billion in one day because the Chinese had made a more efficient AI, the largest one day loss in Wall Street history, I took that opportunity to buy more Nvidia stock. The market had forgotten about Jevons Paradox, but I hadn't. I don't know which AI will end up winning the race but I do know that whoever the winner is he, she or it will never feel that it has too much computing power. In a gold rush the people who make the most money are not those who dig holes in the ground but those who sell picks and shovels. 

 

> Races are only one-dimensional.

Not if there are a huge number of branching ways to reach the finish line and it's not obvious which path will get you there in the least amount of time. 

John K Clark    See what's on my new list at  Extropolis

e4j

[Brent Meeker]

On 8/23/2025 3:56 AM, John Clark wrote:

On Fri, Aug 22, 2025 at 9:43 PM Brent Meeker <meeke...@gmail.com> wrote:

>>Having a better AI model bypasses the need for more electricity

 >> No it does not! Regardless of how efficient an AI model is, the more computational resources it has available the more intelligent it will be, and the more new computer models you will be able to train.

> Yes it does!  Regardless of how much more electricity you have a more efficient model may be superior. 

Superior to what? 

Other AI; with a less efficient model.  The same thing that more training (and electricity) aims at.  You seem to notice this, e.g. HRM, when it suits your argument.  

Reading the HRM paper, it's not clear to me how the "R" part works.  Is it just a recurrent neural net that's smaller and runs faster, or does it actually have some logic built in?

If your AI is not lousy then more computing power will result in more intelligence. If your AI is lousy then more computing power will enable you to find a new AI model, or even a new AI architecture, that is less lousy. So your assertion that a more efficient AI will lead to less demand for computing power is simply not true. 

I wrote: "Regardless of how much more electricity you have a more efficient model may be superior. " I didn't say it would lead to less demand for computing power, but that it might divert the race for more electrical power.  The demand for electrical power is driven by training.  At some point LLM's will have absorbed all recorded human knowledge, but will only be smart the way Encyclopedia Britannica is smart. 

That's why on January 27 when Nvidia lost $600 billion in one day because the Chinese had made a more efficient AI, the largest one day loss in Wall Street history, I took that opportunity to buy more Nvidia stock. 

So did I.  And so far it's paying off.

The market had forgotten about Jevons Paradox, but I hadn't. I don't know which AI will end up winning the race but I do know that whoever the winner is he, she or it will never feel that it has too much computing power. In a gold rush the people who make the most money are not those who dig holes in the ground but those who sell picks and shovels. 

 

> Races are only one-dimensional.

Not if there are a huge number of branching ways to reach the finish line and it's not obvious which path will get you there in the least amount of time. 

Right.  That was a typo.  I meant to write "Races aren't only one-dimensional."  Noting that more training in bigger LLM's isn't the only path.

Brent

[smitra]

On 21-08-2025 16:35, John Clark wrote:
> The problem is electricity, or rather the lack of it.
>

> HAS CHINA WON THE AI RACE ALREADY? [1]
>
> JOHN K CLARK SEE WHAT&#39;S ON MY NEW LIST AT EXTROPOLIS [2]31=
>

I think everyone has lost or is soon going to lose by making massive
investments in the current extremely inefficient systems. It's the same
story with previous technological innovations all over again:


https://www.youtube.com/watch?v=RfN7eFo5cbk&t=736s


Current LLMs are a dead end because reducing errors by a factor of 10
requires a factor of 10^20 more computational power:

https://www.youtube.com/watch?v=mjB6HDot1Uk

and we're already starting out from extremely energy inefficient
systems. Note that the human brain runs on just 20 Watts. With current
technology getting to a robot that has similar capabilities as a human
would probably require a lot more than 20 MW. This does mean that a lot
more technological innovation is opposable, so it is actually good news.
But it is bad news for people who have bought into the current AI
systems or have invested in AI companies. That's why I am short Nvidia,
Palantir, Tesla etc. as these are massively overvalued companies.

Take e.g. Tesla's market cap of $1.1 T. This is for a large part just
phantom. No financial institution that can afford to buy all the shares
of Tesla and make it a private company would consider doing that. It's
all good and well to trade Tesla stocks and they do that with a lot of
leverage that is pumping up the price of the stocks and is causing tis
enormous overvaluation:

https://www.youtube.com/watch?v=cQFQg-9b9EQ&t=1911s

And then you can wait until that leverage falls partially away due to
liquidity drawn away from the stock market to do more important things
that playing games on the stock market, like e.g. having to refinance
debts:

https://www.youtube.com/watch?v=8G1i0vstvTo&t=730s


And then the stock price of Tesla, Nvidia, Palantir etc. will slide down
and it will become clear that the multi-trillion-dollar valuations where
all phantom valuations. These companies were never worth that much,
investors who bought into that were participating in a de-facto Ponzi.

Saibal

[Cosmin Visan]

Yes, they won it, and they had the greatest orgasm in their lives. AI! AI! AI!

[John Clark]

On Tue, Sep 2, 2025 at 6:51 AM 'Cosmin Visan' via Everything List <everyth...@googlegroups.com> wrote:

> Yes, they won it, and they had the greatest orgasm in their lives. AI! AI! AI!

I see that our resident village idiot is back. I don't know where he's been, perhaps to a village idiots convention, but if so he hasn't learned anything from his fellow idiots. A village idiot can be a useful part of a community by providing comic relief, but  even a group of village idiots has a IQ distribution curve, and some idiots, such as Cosmin Visan, are going to be at the extreme low end of that curve, and as a result are too stupid to realize that after the 19th or 20th repetition of the exact same post it starts to lose is comedic value; assuming it ever had any to begin with.  

John K Clark    See what's on my new list at  Extropolis

5vb