[ExI] OpenAI Reaches A.I. Agreement With Defense Dept. After Anthropic Clash

[Keith Henson]

On Sun, Mar 1, 2026 at 4:39 AM <spike at rainier66.com> wrote:

> -----Original Message-----
> From: Keith Henson <hkeithhenson at gmail.com>
>
> > >>> Do you have a clearance?
> >
> > >> Nope. Do you?
> >
> > >No, and that’s the point.
> >
> >...Not arrogant, John is just right.  I am absolutely confident that the military does not have a massive AI data center...
>
> How massive do they not have?

none

 Do we know?

Yes, for many reasons.

>Can AI be trained in a slightly massive AI center?

yes, the Chinese did it and released it.  Caused a trillion dollar
drop in the stock market a few months back.

> How about if it is distributed into a number of micromassive facilities?

The speed of light delays cause such configurations to be intolerably
inefficient and slow. That is why AI data centers in space run into so
many problems.

>  Would we know what they are doing there?  How?

You can see what is there from orbit.  There are no GW scale power
plants out there or huge power lines running into that place.  Plus,
it is known where the high performance AI chips arg going.

> Consider that example of China Lake NWC.  Do locate that on a map pls.

Mostly a lot of nothing out there.  No rivers to cool chips, no huge
cooling towers.  Current AI training takes massive power and huge
cooling.  There may be ways around that eventually, but that is the
situation now.

> Out there they have controlled access, they have restricted airspace, natural gas in arbitrarily large supply, there are ways to vent the waste heat in way which would be difficult to detect by satellite.  We don't know that Musk is doing that, or if he is, he would talk about it.

Please go into details on waste heat.  It is an area I know, and your
statement mystifies me.
>
> >...In any case, "training at West Point" is not the question.  Training an AI on the material at West Point, along with all the text in the world, has already been done.  What the AI companies are trying to do is shape the AI after training to psychologically react to humans morally and ethically.  It is a hard task, one that I suspect will fail to humanity's detriment, perhaps to extinction... Keith
>
> Agreed that is an existential threat.  Considering that, is there sufficient justification for insisting the US government does not have the means to create weaponized AI now?  Can we confidently assume the military in the USA and elsewhere have not already been buying and otherwise collecting technology and expertise to create weaponized AI?  Can we assume that the world's most successful businessman will not also be able to develop it, along with several of his rivals?

That is why I am so pessimistic about our survival chances.

> We don't know what will happen with weaponized AI.

You have not been watching enough Ukraine war videos.  When jamming
cuts the link to a human, an AI takes over flying the drone into the
target and killing any people who happen to be there.

> We don't know what the US government already has, what weaponized AI the Chinese government and even the Russian government already have.  We are being presumptuous to claim to know, based on reason.

Assume the worst, what do you do?

Keith

> spike
>
>
>