Latest / Great Houses / 27. Cursus Honorum of the AI Future Part 2

Transcript

27. Cursus Honorum of the AI Future Part 2 · Great Houses

0:01
All right.
0:01
Hello everybody, and welcome to episode 27 of The Great Houses Forum.
0:06
Today we're gonna be talking about, uh, continuing to talk about, I should say, the, uh, the cursus honorum of the AI future.
0:13
So, uh, here at The Great Houses call, we are thinking and talking and trying to learn about the, the, the different trends in our world that are leading to what I, I think of as the, the emergence, the reemergence of great houses.
0:26
Great houses are these pillar entities.
0:29
They're, they're … Sometimes I call them a firm that has insourced, uh, the, the production of the key pieces of human capital necessary for its survival.
0:36
And, uh, we have, we have obviously been in a, a rather long period of, of trying to outsource the production of human capital, and I think, uh, and, and, and many other people think that, that that's all kinda coming back in, if you will.
0:48
So we are, we are looking forward to that.
0:50
And, um, and also, you know, there, there's just a bunch of, a bunch of changes.
0:55
Our, our world is changing in a, in a bunch of ways.
0:57
So we're talking today about AI, um, as we, as we have often done and as we will often do as this, uh, this kind of moves forward.
1:05
So thanks everybody for joining me, and we'll get into it.
1:07
So last time we talked about, um, how AI is a, a bridge technology.
1:11
It's a, it's a technology that allows you to interact with multiple domains with one interface, one tool set, one, one metaphor, right?
1:19
An interface, like a computer interface, is a metaphor in a sense.
1:23
Um, and that means that, that a single core set of skills now applies to what was formerly a, a vast array of, of disparate domains, right?
1:32
And so the, uh, the example that I, that I used, uh, for that was CAD, the, uh, the, the CAD space, uh, which, you know, they, they had this kind of really interesting dynamic.
1:42
CAD was invented, it starts to spread across multiple industries, and, and the floor drops.
1:47
Like a whole bunch of new entrants can get into this field- Um, and, and then you have kind of these, these, these, these ceilings rise, the space gets bigger, right?
1:57
Then the initial phase is you have, uh, what-- the egalitarian clubhouse, right?
2:01
Where a bunch of people are, are betting, "Hey, uh, we're gonna, we're gonna license access to a bunch of people who are potentially not very good at actually using this underlying technology." But there's a bunch of interest, there's a bunch of people moving around.
2:13
Um, and that tends to be, and that, that is a, a short-term strategy, right?
2:17
Uh, those, those, those egalitarian clubhouses, they spawn and then they die.
2:21
And then what, what emerges is a ranked hierarchy, right?
2:24
Because this, this common space that everyone can operate in means you can observe skill levels between people that formerly were not competing with each other, right?
2:33
So across the… You have a ranked hierarchy that emerges across the whole of the new playing field.
2:37
And what that means, and well, this is kind of the theme of this episode, is that social complexity drops across the entirety of the new space, okay?
2:45
Now, that's important, um, because you need a certain number of specialties to, uh, to run a civilization, right?
2:54
So, you know, classically, you know, the, the, uh, the anthropologists tell us that a hunter-gatherer band had a few, you know, dozen distinct roles.
3:02
Uh, a guy named Joseph Tainter, uh, accounting for his study of, of collapse, found modern censuses recognizing 10,000 to 20,000 occupational roles.
3:11
You know, he's, he's thinking about, um, he's thinking about how complex can a society get before it just starts to implode.
3:18
Um, complexity is expensive, right?
3:20
Every one of those roles is a person who must be trained, paid, uh, paid and managed, right?
3:25
This is the complexity, the number of different roles of people that have to be… They have to think in a different way.
3:31
They have to have a different, you know, uh, vocabulary.
3:34
They have to have a specialized, uh, sense of, of what they're doing, specialized tools.
3:39
All of that, the number of those different boxes that we have in our world is, is exceptionally complex, right?
3:44
Um, another, another framework for this is that the federal catalog of fields of study names over 2,800, uh, distinct programs.
3:52
Medicine alone has 40 specialties and 90 subspecialties.
3:56
Um, and so, you know, let, let's say, you know, somewhere running our civilization takes between 2,500 and 10,000 PhD level specialties.
4:03
I, I, I, I am assuming, of course, when I say that, that like half of the jobs that, uh, that Mr. Tainter, uh, counted are completely useless and, and, and add nothing to our society.
4:13
I am assuming that about, about half of our, uh, PhD programs for, for… And, and I think I'm probably underestimating the amount there.
4:19
But the exact number matters less than the social requirements, right?
4:22
It, it, this is, this is something that is far beyond, uh, what any one human brain can possibly process, right?
4:30
Um, the, the, the precondition… This is the precondition that justifies our current managerial structure.
4:35
So when we say, "Why do we have managerialism?"
4:38
They talk about the complexity.
4:39
It's this thing.
4:40
We have so many different specialties, so many… It's, it's really hard in this environment to be a polymath, right?
4:47
Because every different, you know, piece of human knowledge becomes something that you can only study and interact with, with specialized tools, a specialized educational track, and they each kinda have their own little way of, of, of doing things and thinking about things, right?
5:00
I mean, this is why, you know, the, the woke stuff happened, right?
5:04
The woke stuff happened because each specialty lived within its own domain, and we were divided, and the communists kinda were able to do the long march through the institutions because you could go in and, and compromise or corrupt a, a discipline or a sub-discipline, and people outside of that discipline wouldn't really be able to know.
5:23
They wouldn't really be able to tell.
5:24
Like the disciplines, there's so many groups that had to police themselves.
5:28
It was very difficult, uh, you know, aside from kinda like broad civilizational scale outcomes to even tell, "Hey, you know, all these people are, have kind of stopped doing their jobs. They're not doing what they're supposed to be doing anymore."
5:40
Um- So contrast that with, with, with what I would call human scale organizations, the largest of which is, is Dunbar's number, and I've, I've talked about Dunbar's number before.
5:49
I, I think it's a really important concept for people to, to think about and know about.
5:54
Uh, Dunbar's number is, is roughly, you know, 200…
5:57
is roughly 150 people, at least on average, right?
5:59
There's people that, that think that Dunbar's number is, you know, larger for some people and smaller for other people, and some people that think it's, like, 150 for everyone, and I, I, I… it doesn't really matter.
6:09
Um, that's the, the, the maximum number of, uh, of stable relationships that, that a human brain can, can naturally maintain or, or can track, right?
6:17
This is the number of people… One of the, the kind of the fascinating shorthands for understanding Dunbar's number is this is the number of people for whom you would feel grief if they died, right?
6:26
And now, interestingly, humans, humans feel grief when a, a close friend, a loved one dies.
6:31
They also feel grief when somebody that they really hated or really feared dies because there's the, the, the…
6:37
it appears to be, like, the volume and intensity of emotion attached to this person that causes your brain to assign them a slot, and when that person passes away, you have to use… your, your brain has to use the, the g- grief chemical, the grief hormones to kind of open that slot up and allow you to, to, to assign it potentially to a new person, right?
6:55
So 150 is the number, and this is why, you know, villages, military companies, functioning church congregations, they keep landing at or near that number.
7:02
It is the largest group in which everyone can actually know everyone, right?
7:07
Um, and some people will say that, well, for lower IQ people, it's only about 75, and so 75 is, like, the core tribe, and then you can, you know, double that for, for…
7:15
if you have a decent amount of smart people.
7:17
Um, so And, and, you know, 150 is not the only hardwired number.
7:23
Uh, there's good evidence that the pair bond, the, the squad, which is that grouping of kinda three to eight, and then the, the platoon, let's call it the grouping of nine to 20, they each have actual brain architecture rather precisely calibrated to manage them.
7:36
Uh, one of the interesting things, the squad is the largest number of people.
7:39
Eight people is, is the largest number of people that you can interact with as individuals, right?
7:45
Beyond eight people, your brain, most people's brains anyway, can't really track the group as individuals, and so they start treating it as there's me and maybe m- a couple of people on my side, and then there's the group, which is this just psychologically undifferentiated mass.
7:59
It's a, it's a really kinda fascinating… Some libertarians have done some really good work on what happens when you, you push the group size above, you know, seven or eight people.
8:07
Um, so these nested groups are the natural building blocks a, a small AI era team can actually run on.
8:13
I think, I think, you know, the claim that I'm making basically is that AI is gonna push more and more of what we need to do to, to survive and run our society, which will grow in technical complexity, but a lot of systems are about to be absorbed.
8:29
A lot of social complexity is about to be eaten by AI.
8:32
So we're gonna get pushed down.
8:33
My hope is, my, my, my belief is that, that for good or for, or for ill, these, these, these, you know, socially complex managerial systems are gonna be pushed down back into these, these types of human scaled, um, human scaled things, right?
8:47
So, you know, right now, in our managerial system, the minimum viable social organization, you know, for, for just for producing the key personnel of our civilization.
8:56
So just like the university system exceeds Dunbar's number by two orders of magnitude, right?
9:01
Nobody can personally know what all of the specialties do, let alone understand the training groups and training processes for the specialties that their life depends on.
9:09
So we outsource trust to credentials, to bureaucracies, and modern financialized systems, and we live with what those systems do, right?
9:15
You know, and the, the famous line, you know, the purpose of a system is what it does, okay?
9:19
So again, I think AI is going to collapse that stack.
9:23
So, uh, these models absorb specialty after specialty and hand them to one or a much smaller number of operators.
9:31
So I, I believe that the distinct skills required to run the necessary machines to run our, our civilization are headed towards, you know, three to 10 skills.
9:39
I'm, I'm gonna pick three as I, as we talk about it today and in the future.
9:42
A- and, and the point is that's masterable by one person.
9:46
One person or maybe a small team can master all of the skills that are necessary to run all of the machines for our civilization to continue and replicate itself and move forward.
9:57
And that's the first time that that's been true- In at least 400 years, basically since the invention of, of modern chemistry and the ways that, that, that modern chemistry kind of, uh, changes, changes how, how human technology works, right?
10:10
So now I wanna big-- g- do a big caveat here, um, which is that, that I, uh, I may be wrong about which three skills endure.
10:19
I, I am almost certainly wrong on the technical details.
10:22
There's actually one, uh, one point.
10:23
Uh, so, you know, as I, as I was using Fable to, to help me research this presentation, there was a couple points at which it, it, it suggested that, uh, maybe I shouldn't publish this publicly 'cause it, uh, it was, you know, new or, or special, uh, you know, stuff.
10:38
May- maybe it would be bad for me to pub-- or maybe I was using my powers for evil.
10:41
Um, and if I continued, it would force me to use that, that horrible alternative, Opus.
10:46
Um, and, and I broadly overruled.
10:48
There was one there was one thing that it said, you know, "No one else has ever published any of the things that it can find."
10:54
It used up like 10% of my tokens searching for this online, which, you know, fabulous, I guess.
10:59
Um, that it, it just decided to do that on its own.
11:02
Um, but it, it went through and said, "No one else has published this analysis online that we can find. Do you wanna be the first guy to, to sort of put that out publicly?"
11:10
And I decided no, I did not.
11:12
Anyway, so, so there's, there's, there's many things about this presentation that, that are, are not technically precise.
11:18
Now, I have a wonderful audience.
11:20
I have some really, really bright people here.
11:22
And so anybody that listens to me, please feel free to reach out, help to, to make me technically better.
11:27
Um, but I, I, you know, I, I, I could be wrong about many, many things, and I don't think the things that I'm wrong about change the social implications, okay?
11:37
So I might be wrong about the count.
11:39
It might be 20 skills that we have.
11:41
It might be 50 skills that we have.
11:43
Even 50 skills, right?
11:45
If it, if it requires 50 different specialties, that can be a small team in one room with its own kind of A- AI architecture running locally, and they, they won't need the larger stuff, right?
11:55
They'll… I mean, obviously you'll still need, like, the commodities and the shipping, right, logistics to some extent.
12:00
But that's just a different kind of thing from 10,000 PhD specialties, you know, plus all the lesser jobs supporting them.
12:07
I mean, that's a continent of strangers.
12:08
Like, you need something the size of the United States of America or Europe, right, to, um, to, to compete in that, right?
12:16
I mean, that, that's, that was, uh, one of the things about, you know, NAFTA, right?
12:20
That, that, that Canada and Mexico- They're not small countries actually, and they didn't… They weren't able to compete in the global markets, and so they, they kind of came to us and said, "Well, we wanna participate in a, in a North American market," um, which was obviously implemented horrendously badly and, and, uh, and, and in ways that, uh, sort of betrayed our civilization.
12:40
But, you know, boomers, what can you expect?
12:42
Um, so anyway, the point is any number in that range, you know, if, if I'm, if I'm broadly right about, hey, the, the number of distinct specialties, right, is gonna dramatically reduce, um, it… That-that's gonna represent a massive change in how our society is organized, right?
12:59
E-even if we still have some kind of managerialism, AI-based managerialism is gonna be very different from non-A, from the, from the managerialism that we've had since the 1920s, okay?
13:13
Um, so as we move forward, last time we talked about… Oops.
13:21
A man in a tree line, right, in, in, in, in the Ukraine.
13:25
Um, today is…
13:26
We're gonna talk about h-how do you defend against disruptors like him?
13:29
The… This is, this is gonna be focused on the, the military and law enforcement piece, right?
13:34
Um, unfortunately, as I was going through this, I, I, I hit 60 slides, uh, and, uh, and realized, "Okay, we're, we're not gonna get to the civilian applications today."
13:43
Uh, by the way, if you, uh, if you join and get behind the paywall, then, then I, I publish, uh, my presentations in the, uh, in the show notes of the, the paid version.
13:51
So, um, and anyone who joins one of the, the sponsored communities gets access to the paid version.
13:55
So that's, uh, Radical Personal Finance, The Classical Society, and the Exit Group.
13:58
So, but we're not gonna get to these civilian applications today.
14:01
But the massive disruption in our underlying, you know, military tech, which is also effectively the law enforcement technology, is already happening, and the only question is whether we build the pipeline, the cursus honorum, to stay ahead of it.
14:13
So-
14:16
So last episode, I, I called the three skills logic, rhetoric, and command.
14:20
And so, you know, as, as I mentioned before, I, I, I was raised Baptist, and so I love for things to alliterate.
14:25
So the alliterative version of that is code, content, and command.
14:29
So code is the logic underneath, um, your software, right?
14:33
It's directing AI to write code that serves your goals, and basically what we're finding out is that in order to do that, you, you need to understand at a high level the architecture of the thing that you're doing in advance.
14:44
You can't… Like, AI does not help you if what you tend to do when you're coding is write a section, then kind of try and write the next section and see if that breaks the section you just wrote, and then kinda move sequentially through, right?
14:57
You're sort of feeling your way through the, the, the logical architecture.
15:00
That, I mean, you could, I suppose you could do it.
15:02
May- maybe you get some marginal improvements by doing that with AI, but it, you, you don't get the order of magnitude increases that, that, that some of, uh, the, the coders that are really getting into AI are, are, are enjoying, okay?
15:14
So then content is, you know, the rhetoric, right?
15:17
Uh, this is propaganda.
15:18
This is persuasion.
15:19
This is, you know, the Scott Adams and the Joshua Lysecks and all, all of that stuff.
15:23
You know, the, the TikTok videos and, and, and, and all the, the, the content creators and influencers on YouTube and the various, you know, other video, video platforms that, uh, that people are, are using today.
15:35
Um, content is a super important thing, right?
15:38
This is, this allows you, especially with AI tools, to kind of persuade people at scale.
15:44
Um, and, and, and I think a lot of people are focusing on the, um, on the ability of AI to, like, create new content, deep fakes, all of that.
15:52
I actually think that's less important than the, than the ability to, to…
15:56
'Cause I think we'll have ways of, like, digitally, cryptographically verifying whether some amount of video came, you know, from, from a, a particular object or a particular camera, et cetera.
16:06
Like, I think that'll be possible.
16:07
Um, but it's gonna be about cutting.
16:10
It's gonna be about understanding, okay, we've got- Effectively infinite amounts of footage, right?
16:16
We got so many cameras, so many, you know, drones, so many things that are, that are generating video.
16:22
What do we say, right?
16:23
What do we say with all that content?
16:25
What do we say with all that, that background?
16:27
And probably there will be a place for AI to kind of merge different, uh, different video streams and, and create new perspectives on, uh, on things.
16:34
But so content, very important, right?
16:36
And we'll get into… I, I actually don't think that-- I think that you are going to need the first two, two, uh, skills in order to effectively use the last skill, which is command.
16:45
And that's the decision-making, right?
16:46
That's directing the machines, making the decision about something is now going to happen because I pressed a button.
16:52
Potentially, you know, something economic, potentially something, uh, you know, even lethal in the real world.
16:58
This is gonna be the, uh, the, you know, at some level, right?
17:03
Whenever people talk about autonomous AI systems.
17:05
Well, I mean, yes, autonomous in the sense that they're implementing a, a command, but some person coded that command in.
17:13
Some person defined all of the things that, that, that, that, um, that AI is, is, is implementing.
17:19
And, and even if the specific code was written by another AI, it's like there, there is always a human originator at the back of all this stuff.
17:27
Um, and, and we are, I, I suspect, going to hold that person very, very, uh, legally responsible for, for the things that they, um, that they produce.
17:36
So just to, you know, since, since I am on the right wing and we'll, we'll have a little bit of fun with this.
17:41
Um, so you might think of when we think about these, these three, these three specialties, what does it look like for, for somebody to have all three of these, right?
17:49
The, the, the person that's a triple threat.
17:51
Well, that person, um, it would be-- we might call them a prime, right?
17:55
The triple threat.
17:56
Now, below the prime, we might have the single skill specialist.
17:59
Uh, the logic and code specialist is, uh, we might call him a mentat, right?
18:03
Love that word, mentat.
18:05
Um, the, the rhetoric and content specialist might be a priest or, or if you like, a, a, a psycher or a witch, right?
18:11
Um, and yes, I'm drawing from Dune and Warhammer 40K.
18:14
Sue me, I'm on the right wing.
18:15
And then, and then Paladin would be the, the person that we call-- some- that we call the person with the, the skill of command.
18:21
Um, but again, I, I, I, I sort of use Paladin and Prime somewhat interchangeably, um, 'cause, uh, as I'll get into, I, I don't, I don't think you can be an effective decision maker.
18:30
I don't think you can use command unless you have at least one and probably both of the other skills.
18:35
I mean, if not, if not mastered, then pretty well understood.
18:38
Um, and the basic claim that I, again, that I wanna make is, uh, I hope you'll see this as we go through, we are going to need a lot of primes.
18:45
We are going to need a lot of these people, and, and I don't think we're gonna have enough.
18:49
Okay?
18:50
So, so back to, you know, we, again, we ended last ep- uh, last episode talking about, uh, you know, somewhere along a tree line in eastern Ukraine, a man is crouching over a controller.
19:00
He's got a battery, he's got a laptop running a piloting model, uh, an LLM at the edge of the network, and an airframe that costs less than, than a used phone, like a, a drone, a drone, uh, airframe.
19:10
So two kilometers out there's a, there's a column of tanks which mo- is moving towards a military target, and each tank by itself is worth a thousand times this guy's entire kit.
19:19
The, the laptop, the drone, the, probably all of the, the, the, um, the, the training time, all the expenses that it took to, to train this guy.
19:28
Any one of these tanks is more expensive than, than, than the whole thing.
19:32
Um, and so, you know, he, he watches the lead vehicle through the drone's camera, and an AI brackets the scene where the hull plating thins and, you know, y- you can imagine this would be like a, like a carrying graphic from one of those old Star Fox games.
19:44
Um, he clicks an authorization button on the pre-planned kill run, and the LLM on his laptop, plus, you know, maybe some, some amount of inference on the drone executes it faster, uh, than, than human re- reflexes can track.
19:56
So w- a tank blows up, the column stops, it backs up, it looks for another road.
20:00
One person, uh, using a few thousand dollars of parts has just turned an entire tank group.
20:06
And, and, you know, that's, that's changing our world.
20:09
That's happening in Ukraine, right?
20:10
Uh, we'll get to, you know, we… Uh, Ukraine had some Abrams tanks.
20:14
You know, you didn't know that Ukraine could make Abrams tanks, but no, of course they didn't.
20:17
We gave to them.
20:18
Um, we pulled our, our Abrams tanks out.
20:21
Why did we do that?
20:22
Very interesting.
20:23
Um, the part of our, of this story that interests me is, is the part that will, you know, involve social reorganization, is how do you defend against that disruptor, right?
20:32
Um, a- and in particular, how do you defend against the fact that that disruptor does not need a nation state, you know, level group behind him?
20:41
Um, at least, at least it doesn't seem like that to me.
20:44
So, so what does it look like?
20:46
What… When we're defending this, what does that look like?
20:48
Well, you know, on, uh, September 14th, 2019, it was before, you know, all the, all the COVID stuff, 18 attack drones and seven cruise missiles struck the, uh, Abqaiq and Khurais facilities in Saudi Arabia.
21:00
They knocked out 5.7 million barrels, roughly 5% of the world's oil supply, which is the largest, uh, sudden outage in the history of oil, right?
21:08
Um- Then, uh, you know, more recently, uh, between, uh, July 14th and 16 of, of, of this, of last year, uh, um, drones hit five oil fields across Iraqi Kurdistan, including, uh, the DNO's Tawke and the Sarsang field.
21:25
Production fell by roughly 70%.
21:27
No one claimed the strikes, but the, the analysts pointed to pro-Iran militias.
21:30
And of course, you know, throughout the, the, the… this year as we've had the war in Iran, um, there has been just an incredible amount of, of drone attacks on various oil facilities.
21:40
Uh, the one that, that, that I, I recollect was there was a, a, uh, natural gas processing facility that, that processed natural gas, like out of the air, out of the exhaust from one of these oil fields, and it was like a billion-dollar, you know, facility.
21:56
And, and, and a drone, a very… a relatively cheap drone, I think it was one of, like, the $500 drones, went in and just took the whole facility out.
22:02
The- they're gonna have to rebuild all of the machines in factories in Germany, and they're actually not sure if they can rebuild the factories in Germany or th-…
22:09
because the factories in Germany are now surrounded by vast amounts of immigrants.
22:13
It's not clear that the, uh, technical skill is available to those guys anymore, okay?
22:18
Um, now, you cannot guard a, a facility with just sentries and a fence or, or even point-defense turrets managed by operators in a control room.
22:25
The- these threats are arriving faster than human reflexes can respond.
22:28
I mean, that's, that's one of the things, you know, that, that, that, that high-pitched buzz, zoom, right, that you… that we are starting to associate with drones.
22:35
Um, like, the final attack run takes, like, a third of a second for most… So, and it's crossing, like, 250 feet so fast.
22:44
You, you can't, you can't think that fast.
22:47
Uh, so what do you do?
22:48
Well, you put a swarm of drones hovering in a rough kinda half bubble around your facility and, and, and that swarm is gonna need an operator.
22:56
Now, when I say an operator, this guy is not taking over and piloting individual drones.
23:02
I mean, presumably he could.
23:03
Like, I assume we're gonna build that into the system.
23:05
But he is gonna be interacting with kinda like a three-dimensional game interface over his own facility, right?
23:11
He makes big picture calls, where to mass the swarm, which sector to thicken, when to, you know, kind of pivot around a flank.
23:17
And this is basically as, as, as, as far as the public is informed how Anduril's Lattice, uh, works, right?
23:23
So you got one operator, you got a live map, you got a bunch of autonomous machines.
23:27
And Anduril, of course, uh, the, the Lattice system was, uh, was Palmer Luckey's attempt to close the southern border, which we could do, right?
23:34
We, we know how to close the southern border.
23:36
You put the, the Lattice system in place, and relatively few people and, and amazingly cheap drones can patrol and, and effectively shut off just incredible amounts of territory, okay?
23:51
So, and again, you know, I, I mentioned that this is kind of similar to a, a gaming interface.
23:54
So you can imagine this is similar to like a real time strategy game such as StarCraft.
23:59
So when an object in a drone's field of view does something that the pre-coded rules define as indicating that it is an enemy, then the drone engages on its own, right?
24:06
In, in, in StarCraft or in these, you know, real time strategy games, you have your units.
24:10
They kind of default to being in attack mode, and that means if an enemy unit, you know, shows up in their, in their range, they just shoot at them, right?
24:18
Um, and, and they, they, they do that autonomously, right?
24:21
So the operator is still involved.
24:23
He's still kind of moving them around.
24:25
He's still… I mean, he could, he could turn that off.
24:26
He could say, "Okay, I don't want you to shoot at anything that, that moves or even things that are being targeted, that are being targeted by the, the watcher AI, as we think this is an enemy," right?
24:35
Because it's always one of the, one of the differences, right, between the game and the reality is, of course, someone has to code the, the, the parameters that, that make that, that unit, that moving object defined as an enemy, right?
24:49
Like, someone has to have sat down and looked at inputs and, and trained an AI that is distinguishing, that is making the determination, well, it's, you know, 80% likelihood that this is an enemy, right?
25:00
It's 90%.
25:01
It's 99% likelihood.
25:02
It's 99.999999, right?
25:05
Um, and so you've gotta, you've gotta think through that.
25:09
Like, how do you, how do you… Uh, who, who makes that decision?
25:13
And I, I would expect that it'd probably be basically the same guy who is, um, who is making the call at that, at that facility.
25:20
He's gonna need to, at least at a minimum, he's gonna need to be able to interact with those, um, those deep, those deep structures, okay?
25:28
And, you know, I made this point to someone.
25:30
They, they, they, they asked about point defense.
25:32
The-- I'm sure, you know, these facilities can have point defense, right?
25:36
Lasers, rail guns.
25:37
They have these new kind of like shotgun s- type cannon things that just send out a vast amount of little, uh, little, little balls and, and, and pieces of shrapnel that, that can kind of clear a whole chunk of the sky.
25:48
More traditional machine gun turrets.
25:50
The point is, even these will be incorporated into an AI-run defense network that is likely guided by a singular human or a very small team.
25:58
Th- It's not gonna be like a traditional, like you see in, in, in the old movies with the, the battleships, um, and you've got different crews running a gun and the, the, the commander is telling people what to do, but the guns, you know, you have to have somebody there to load it or to fix it or whatever.
26:12
Um, you're not gonna have that.
26:13
This is all gonna be automated, and it's gonna be c- controlled, uh, at a very high level by somebody that's sort of pointing and clicking on a computer interface, not somebody that is there steering a mechanical, uh, interface and, and a- and aiming the gun, right?
26:29
The other thing is that, um, you know, point defenses have a seam, right?
26:34
So you, they, they ha- they have weaknesses.
26:37
Um, one of the things that, that seems likely, like everyone has seen the, the, the drone shapes, right?
26:42
You've got like a dragon, right, flowing through, and it's, it's a bunch of different points, uh, that are, that are cooperating modularly to, to create that, that vector.
26:50
You're gonna have some kind of, you know, three-dimensional phalanx.
26:53
You're gonna have, uh, you know, drones out front.
26:56
You fly some cheap drones out front.
26:58
They, they can have chaff, which is basically just, you know, strips of aluminum foil.
27:02
Um, that means it's really hard to see them.
27:04
It's really hard to aim with radar.
27:06
I mean, it might even work on lasers, right?
27:08
Uh, so it's, it's, it's unclear how much we could degrade lasers by just having like two or three layers of, of drones with a bunch of aluminum foil around them get in the way of the laser.
27:19
Okay?
27:19
So now the escorts are probably gonna die doing that.
27:24
But again, these are cheap drones.
27:26
Who cares?
27:27
Um, if they, if, if, if five or 10 or 50, you know, $50 drones get killed in order to get the, the rest of the drones or to get a weapon system within striking range of a facility, I mean, easy.
27:41
That's, that's easy to do.
27:41
These, these, these facilities are worth millions or billions of dollars, right?
27:45
Um, and, and I'm…
27:46
This is not just me doing random research on the internet.
27:48
There's a, the Strategic Studies Institute basically put something like that out in, um, in January of twenty twenty-five, uh, so this is just kind of an unsolved problem.
27:57
So our, our… I, I mean, it's not an unsolved problem.
27:59
We know, we know more or less what the solution is, it's just that our…
28:01
it's not gonna be point defense.
28:02
Our point defense systems are designed to stop missiles.
28:05
They are not designed to stop a modular swarm of cheap drones arriving together when many of them are just there to be shot, basically.
28:12
Um, and so that means that, that the defensive swarm is going to be the core defense for every piece of critical infrastructure in a threat zone, like, I mean, until somebody invents an actual force field, right?
28:22
Which, you know, uh, they, is like 50 years away, right?
28:25
They- it's been 50 years away for a long time now.
28:28
So then the question is, well, uh, how big is the threat zone, right?
28:32
How big is the threat zone?
28:33
Uh, well, the Iranian Shahed one thirty-seven, or excuse me, one thirty-six, flies about, uh, two thousand kilometers.
28:40
Uh, the Shahed one thirty-six B, unveiled in September twenty twenty-four, claims four thousand kilometers, which would be enough to put a bunch of European bases and cities in range.
28:50
Um, you know, Ukraine now strikes Moscow and beyond hundreds to over a thousand miles deep, uh, with a June twenty twenty-six raid sending nearly four hundred and fifty drones, um, at, at the capital of Russia, right?
29:01
According to NPR.
29:03
That 2,000 kilometers is a long way, right?
29:07
Um, so at 2,000 kilometers, and, and these, you know, defense analysts, European defense analysts have, have looked at this.
29:12
That, that means that basically every European capital is ex- inside the threat envelope, uh, with, I think with the exception of Lisbon, right?
29:21
Um, and that means that, that, you know, these… We think of drones as kind of like a short range weapon, but you get one of these up, you get it kind of going at the right speed, especially if they can find wind that's going in the right direction, they can go a very, very long way.
29:34
The, the, the, the sort of the power to weight ratio of these units is, is, is really impressive, right?
29:40
And I mean, that, y- you also gotta think about, you know, delivery systems, right?
29:43
So, uh, March of this year, uh, Chinese state television aired a, a, a demonstration of they call it the Atlas Swarm system.
29:50
The Atlas Swarm system.
29:51
They have this truck that, uh, that drives up, and then 48 drones get launched from the truck.
29:56
Um, it takes about three seconds per drone, and then they, they go out searching the, this field full of decoys, and they find a command vehicle.
30:03
Um, and, and an AI, uh, running, presumably running in the truck itself.
30:07
It's not clear.
30:07
You know, this, this is all Chinese propaganda anyway.
30:10
Um, pick the target.
30:11
Uh, and, and, and, and what that means is, again, a human decision-maker gave a broad target, and then all of the detailed instructions were carried out by AI, okay?
30:20
Um, China has also shown a, um, a 200-drone swarm commanded by a single soldier, right?
30:27
Um, the, the algorithm and, and that, that, that swarm actually was set up so that even if they lost, uh, contact with the operator, the, the AI on the units and, and cross-link between them could keep them going for, for a while, uh, even if they didn't have direct long-term contact with the operator.
30:43
I mean, again, the idea would be maybe you're experiencing jamming, and you can briefly punch through or have some kind of like laser-based, uh, communication system where you, you paint one of the, the, the drones and communicate something to it, and then that, that, that connection is gonna break.
30:58
But that's okay, 'cause it's just basically giving a packet of predetermined instructions that, that the, that the AI of the swarm already knows what to do then, okay?
31:09
So, you know, let's, uh, let's, let's sort of split the difference.
31:11
Let's take a conservative, uh, view.
31:13
Let's say, let's just say a drone truck with, with a practical range of 500 miles, right?
31:18
Um, let's say we take one of those trucks and it, it gets landed through lax Mexican customs or a corrupt Canadian port, of which there is a lot, and it drives to their US border.
31:27
From, from those borders, what can it reach?
31:29
Well, within 500 miles of Canada, just to kind of give you an idea, is Seattle, Portland, Minneapolis, Chicago, Detroit, Cleveland, Pittsburgh, Buffalo, Boston, New York, Philadelphia, and Washington, DC.
31:42
Within 500 miles of Mexico is San Diego, Los Angeles, Phoenix, Las Vegas, El Paso, San Antonio, Austin, Houston, Dallas, New Orleans.
31:50
Um, the only major… And then obviously everything in Florida.
31:53
Um, the only major metros outside that band sit around Denver and Kansas City, and a, and a relatively, uh, small improvement in range fixes even that, right?
32:08
So, um, if almost every American city is inside that range, um, then the vast amount of, of our critical infrastructure, substations, refineries, ports, and stadiums are, are inside the threat zone.
32:20
Um, potentially all of them will need local inference and a, and a local operator holding a defensive swarm overhead.
32:27
And, you know, every critical node is gonna need an AI paladin or, or, or some small team that can make this work on over- on Overwatch, right?
32:37
Um, now one of the things that I wanna, I wanna think about with you is, uh, w- This technology tends to be framed as, well, it's making things obsolete, right?
32:48
Now, so we ask the question, is the tank obsolete, right?
32:51
Uh, Abrams, as I said, deployed… were deployed in Ukraine.
32:54
I think there was, like, 31 of them.
32:56
Um, and they were deployed in, in Ukraine.
32:59
They were pulled off the front in April 2024 because drones had… were hunting them very successfully.
33:03
Five of the 31 had already been lost, and, uh, and, and so people, you know, had headlines asking, "Is the tank obsolete?"
33:10
A- a- and, and, you know, it… for those that aren't aware, the tank has been declared obsolete many times.
33:15
Many, many times.
33:17
Uh, so a, uh, a, a person with a, a, uh, a X handle that I won't repeat on, on a recording said, "It is 1919.
33:26
Tanks are declared obsolete due to anti-tank rifles.
33:28
It is 1936.
33:30
Tanks are declared obsolete due to rapid improvements in anti-tank gun technology.
33:33
It is 1949.
33:34
Tanks are declared obsolete due to shaped charge anti-tank weapons.
33:38
It is 1953.
33:39
Tanks are declared obsolete due to tactical nuclear weapons.
33:41
It is 1973.
33:43
Tanks are declared obsolete due to anti-tank guided missiles.
33:46
It is 2006.
33:47
Tanks are declared obsolete due to IEDs and fifth generation warfare.
33:50
It is 2024.
33:51
Tanks are declared obsolete due to F, uh, FPV drones." And, you know, so and he just kind of says, "To be continued" at the end of it.
33:58
Um, and by the way, he was doing this… this, this was in response to somebody that had basically said, "Is the tank obsolete?" And that, that person deleted their account because this tweet went viral.
34:09
Good times.
34:10
So anyway, the lesson is not that tanks are obsolete any more than humans are obsolete, right?
34:14
What… but the, the tank does need an AI drone swarm as, as flankers and as force multipliers.
34:20
I mean, this is basically the way heavy cavalry has always needed flankers.
34:23
Um, and now it makes sense, m- it makes much more sense, I think, for a single human to command both the tank and its flankers, especially because this AI stuff means no one has to drive the tank anymore.
34:36
All right?
34:36
No one has to be there pulling the levers, okay?
34:39
Like, we can, we can have a box, the same box that can run, run the drones can, can tell the tank, "Drive here, drive there." Okay?
34:47
Now, um, as we're thinking about stuff like this, who, who are we worried are?
34:52
We, we worried about China?
34:53
I mean, yes, to some extent we're worried about China, but we're, we're also worried about the cartels, right?
34:58
Uh, weaponized drone attacks by Mexican cartels rose from about five in, in 2020 to 260 by mid-2023, and that was, uh, from a Book- Brookings report called Small Wars Journal in, uh, uh, in February of this year.
35:13
Then in, in March of 2024, one group ran a 72-hour campaign, dropping 40-plus explosives to terrorize and displace civilians.
35:21
It was called, uh, the, the, uh, the, uh, CGNG- CJNG, which is one of the cartels, flew its first FPV attack in April of 2025.
35:29
And, and basically what they did is they, they filmed it, they filmed them releasing, you know, these attacks, and then they released it to the public, I think it was on YouTube, right?
35:39
Basically to intimidate and, and, and recruit and all of this stuff, right?
35:43
So they're already doing these three things.
35:46
They're coding, right?
35:47
Somebody is, is, is getting these, these drones.
35:49
They're, they're, they're setting them up.
35:51
They're, they're actually doing it.
35:53
They're doing the command piece, and then they're doing the content piece, right?
35:55
This is not even a, a nation state.
35:58
This is a criminal organization.
35:59
And that means that we will need… Like, it's not just, we're not just worried about, uh, groups that, that, you know, we're not just worried about China and Russia, kinda like strategic, uh, rivals.
36:10
We're worried about a whole bunch of sub-nation state actors that are gonna be able to get access to this technology and gonna be able to do incredible things with it, right?
36:19
Um, another, another feature of this is, is not just, you know, kind of flying drones, but, but, you know, drones that are walking around.
36:26
So they had a rifle-armed robot dog, and, and I'm, and I'm using a lot of sources from China.
36:30
You know, one of the things I'll say about Chinese sources, um, you know, you never know whether what you're seeing is, is actually real or it's just, you know, some kind of, um, some kind of bluff that they're put- they're putting out.
36:43
Um, but the fact that they felt like they needed to say that is, is kind of interesting and important.
36:50
Um, so, so, you know, by early 2026, reporting had them carrying missiles and recoilless weapons, not just, uh, rifles.
36:58
Um, and, and, and, you know, one of the clips, you know, went, went viral on, uh, on, on, um, X.
37:03
Basically, it was, it was the, the, the robot dog, right?
37:06
It looks like the, the Boston Dynamics robot dog, but they had put a rifle into it, and it was shooting, and it was holding it really, really steady.
37:12
Um, so the, the armed ground robot is a solved engineering problem.
37:16
And this is, you know, this is the dog, so it's not a tracked, uh, vehicle.
37:19
Like, it doesn't… It, it, this thing can jump, it can run, it can go over rugged terrain.
37:23
Uh, it, it can climb pretty effectively.
37:26
Um, so the, uh, so the question would be, can the cartels get those?
37:32
You know, are, are the cartels already fly Chinese drones?
37:35
Are, are we sure they cannot buy or be handed the armed versions?
37:38
Um, and, and, you know, even if a, uh, even if a, um-
37:44
Even if our, our strategic rivals, you know, don't wanna use, they don't wanna, they don't wanna be responsible for using it on us, they have every incentive to arm the groups that are already destabilizing our southern border.
37:55
And as drone parts become more and more ubiquitous, it will become even harder to ferret out people buying drones for malicious purposes, right?
38:02
So 'cause, 'cause basically what we're doing right now as a, as a culture, you know, uh, is, is we're relying on, on mass surveillance, right?
38:09
So in June last month, uh, the FBI and Secret Service announced that they had foiled a planned drone and sniper attack on the, uh, the, the UFC fight that happened at the, at the White House.
38:18
Their plan was to kind of send in attack drones and kind of drive people into these, um, into these kill boxes, and they were gonna shoot them.
38:24
Uh, that was, that was their plan.
38:27
Um, but, uh, it was stopped by, uh, investigation and surveillance 10 days out before a single drone left, left the ground, right?
38:36
And, and the question is how many, how many times can we do that, right?
38:39
This is a classic asymmetric warfare type of a thing, right, where we have to win every single time, and they only have to win once.
38:49
Um, if they win even one time, if they get drones launched even one time, then man, that, that, that is a crazy force.
38:56
Or, or just think about, you know, like the criminal element.
38:58
If you've got a, a drone and you've got somebody that, that's good with their hands, a kid that, that, that's, that's a good technical person, right?
39:05
How confident are you that he can't take a 3D printer and some drone parts and, and the pieces of a gun, right?
39:11
You know, the, the, the, the lower receiver, right?
39:14
Which is, which is technically the gun according to the, the ATF, right?
39:17
Uh, as all the people that have followed, like the Ghost Gunner, uh, guys have, have, have found.
39:22
How confident are you that, that, that that's not going to result in he… that he's not gonna be able to put something together that can point and aim and fire a gun fairly accurately?
39:32
And if, if it can, then that can happen so fast.
39:36
It can happen so fast.
39:37
Yeah?
39:39
Um, now one of the things I wanna, I wanna mention is that, that these
39:47
These interactions, they, they are going to be in the theater, right?
39:52
They're going to be in the specific location where you're launching the drones, um, at least for on defense side, right?
40:00
Um, be- and, and, and that there is a, there is a, there's a, a fair bit of risk here, right?
40:05
So if, if you're running local inference, that which is the cheapest, the fastest, and, and the hardest to jam, right?
40:12
That means that you have a price, right?
40:14
A, a man who is signaling to his swarm is, is probably gonna light up the RF frequency, right?
40:18
If you're sending a laser-pointed, uh, you know, communications system, you're, you're communicating instructions via, via lasers, probably you're gonna have encryption, so they can't break that.
40:28
They may not be able to break it, at least not quick enough to matter, but they're going to know where you are, okay?
40:33
To the enemy sensors, likely this guy is the brightest thing on, on the field.
40:36
He is not, you know, he, this is, this is not traditional drone, um, like what we think of as like the, the, kind of the US drone programs where you have guys that, that are, that are in the United States frequently that are piloting stuff around the world.
40:50
Um, but he, you know, this guy is right there.
40:54
Um, and, and I think this is gonna be the case.
40:56
You know, it's entirely possible that, that the US has some super special communications technology that, uh, that allows you to, um That allows you to punch through jamming or, or have, you know, something that can't be blocked.
41:10
Um, but I think, y- you know, again, the, the thing that, the thing that's gonna happen is not all of these players are going to be run by the US military or the, the, the, the intelligence services.
41:21
A lot of this is gonna be local.
41:23
A lot of this is gonna be local police departments, local, uh, you know, or state level, uh, players that are, that are trying to protect their, their infrastructure, that are trying to deal with criminal e- elements that are going to be wildly empowered by this technology, and they're going to be on the board, right?
41:40
Um, and again, just, just a note on the jamming, the jamming stuff.
41:44
You know, Ukraine has kind of settled the question.
41:47
Jamming works really well, right?
41:49
If the drone is being piloted by a model on a distant server or being piloted by, you know, somebody using controller, a- as soon a- it's, it's very easy to kind of just light up the, uh, the electromagnetic spectrum and the radio spectrum, and then the drone just falls out of the sky.
42:04
Uh, so the defender can't… you can't command from a bunker three states away.
42:07
The- you have to be right there.
42:09
Um, you have to be, you have to be local, on site, you know, close enough to feel the hardware hum, as they say.
42:15
And what this means is, is that local command means this is a, a, a game of StarCraft where the loser does not get a rematch, right?
42:21
If your, if your, your swarm is fighting another swarm, when the last of your drones falls, then nothing stands between the other machines and, and where you are, you know?
42:30
And so you potentially die, like, half a second later.
42:32
And, and I'm sure we'll come up with ways of, of, you know, defending and, and, and escaping and all of those things.
42:38
Um, but this is a different, this is, this is a different kind of thing, um, than, than what we have thought about from, from drones in the past, okay?
42:46
So this is the AI of Halibut.
42:48
Like, I'm, I'm giving you all these things to kind of describe the way that I think this will work, right?
42:52
The guy who can build the swarm, command it, and survive being seen is, is the new apex human on defense and offense.
42:59
And, and, and again, this is code, content, and command, and maybe some other skills that I'm not aware of and not, not thinking about right now, either in one individual or in a small team.
43:09
This is the most valuable, most dangerous person that, that, that our world is gonna produce in, in this new era of AI and drones and robotics and all this stuff, okay?
43:19
Um, so and, and what's interesting is this, this is, this is not just usable against other drones, right?
43:26
So Ukraine's Sting interceptor, uh, went viral in April of twenty twenty-five, killing Shaheds midair.
43:32
Uh, the US is, you know, racing to reproduce it.
43:34
Y- you picture a small drone, uh, with two long metal tubes jutting forward, right?
43:39
You, you can see videos of it.
43:40
It's, I mean, people have kind of talked about this is like drone jousting, right?
43:43
Um, and, and I think their, their long-term idea would be how do we get a drone with, with, with a, a lance that we can, you know, use that to, to destroy an enemy and not kill the drone in the process?
43:54
Then they can make, you know, much, much more, uh, expensive drones and much more capable drones.
43:58
Um, that, that same interceptor threatens, you know, more than drones.
44:01
Uh, short-range missiles worth, you know, thousands, um, and, and you know, e- potentially even fighter jets, right?
44:08
So the… There was a F-15, a US F-15, uh, that, that, uh, was just, was knocked out of the sky in Iran, uh, this year, and one of the reports described it.
44:17
There was, there was a jellyfish swarm of drones in the sky before the jet went down.
44:22
It's not clear if, like, a drone flew into his engine or if the drones were somehow guiding, uh, uh, shoulder-mounted missiles from the, from the surface.
44:31
Uh, but the point is, like a, a five thousand to, you know, five hundred to five thousand dollar drone, if you can exchange, you can exchange a lot of those against a fifteen million dollar fighter jet, right?
44:43
Like that is a, that is a deal that you take every time, okay?
44:48
Now, again, the, the, the point that I'm making here is not that the fighter jets are obsolete, okay?
44:53
That's not what I'm saying.
44:53
That's not… So it's, it's that the fighter now needs probably a local swarm flying cover ready to kill the enemy swarm or interfere with it before it reaches the main delivery system.
45:03
And, and if you have a drone, you know, system, then you're probably less and less piloting the, the, the, the, the main fighter jet.
45:15
You're piloting the swarm.
45:16
You're piloting using AI to control a group of machines, one of which you happen to be in, right?
45:24
And this is all possible because, again, one, one of the things I wanna, I wanna bring us back to is the interface, right?
45:28
So th- there, there's an intuitive interface going on here.
45:31
Because people say, "Well, you know, it's not efficient. You couldn't, you couldn't control things that quickly, that fast." You know, i- in December of 2025, a Chinese creator, Fan Xisan, flew 50 sho- sword-shaped drones with hand gestures, wuxia style, right?
45:45
Like, he, he points, right?
45:46
He has kind of hand gestures and movements, and the AI is watching him.
45:50
The AI, uh, I think, I think it was actually, uh, on some of the drones.
45:54
And it has this group of drones that they, they fly, they assume different shapes, they go different places based on what he is doing.
46:01
He pushes a palm forward and the blade advance.
46:03
He pulls it back and they return in formation.
46:05
No joystick, no code.
46:07
The machine is interpreting his movements, right?
46:11
And turning that into machine-readable instructions, okay?
46:14
The same thing is what is driving, you know, the, the new mechs, right?
46:17
So in May, a couple months ago, Unitree unveiled the GD01, a 2.8-meter manned mecha piloted from an open cockpit, walking on two or four legs.
46:29
Um, it was reported in Euro News.
46:30
So its CEO climbed in and smashed through a wall on camera.
46:33
Now, now that, that mech is like, you know, almost three-quarters of a million dollars.
46:38
Uh, but it is in production.
46:40
Um, and, and without … The point here is without AI, piloting this mech would take a full team of operators in a van.
46:46
It's similar to how, like, movie special effects teams works, right?
46:49
Right, so you got, you've got robots that do really pretty cool things, but there's so many, you know, knobs and dials and, and things that you can move.
46:57
Y- you have to have a whole team of people in a van to pilot one robot one time.
47:02
Now the onboard AI can translate subtle human action into precise machine-readable instructions, okay?
47:08
And so the interface is the point.
47:10
So AI translates human intention into machine action.
47:13
So you no longer need an individual pilot for any of these machines.
47:17
You don't need, uh, each tank to have a crew.
47:19
You don't need a, a- an individual fighter pilot.
47:21
You don't have to have one operator per drone.
47:24
Now, I'm not saying that, like, we're gonna see
47:26
Like, m- there will probably be manual controls on, on the command vehicles, on any c- c- vehicle that has a human in it, right?
47:33
Um, but I would assume that the tactical doctrine will default to flying the command vehicle as just one unit in the swarm, right?
47:42
Under normal conditions.
47:43
Um, so anyone who exer- wants to exercise, um
47:46
Now, so, so, yes.
47:48
That … I think we, we … I wanna, I wanna rest on that point for a second about- You're going to interact with it like a real-time strategy, right?
47:57
Like a real-time strategy game.
47:58
It's, it… You're gonna have units that can't be killed, right?
48:00
Because one of them is, is the one you… One of these units is what you're in.
48:03
But you're not really interacting from your singular viewpoint.
48:06
You, you have a whole bunch of cameras that are feeding you information, and you are piloting the group, right?
48:12
And you're, you're sort of switching between modes.
48:15
That's what you're doing.
48:16
Okay?
48:17
Now, when we think about this, I, I, I wanna emphasize that I think these skills bundle, okay?
48:23
And that's because anyone who wants to exercise command had better have mastered rhetoric.
48:26
So every machine that the commander directs carries at least one camera, I would assume, and, and, and let's assume that every frame is saved somewhere.
48:33
So any kind of lethal command decision or, or, or non-decision, especially in like law enforcement context, is going to face a political inquest.
48:42
Uh, uh, we talked about this in the Q&A last time.
48:44
Um, and, and the inquest will likely get access to, to the paladin's own footage.
48:48
So every life-altering decision needs a defense in the form of presumably a video essay.
48:53
You need camera angles, narrative, persuasion, likely released in, in like a YouTube or TikTok format.
48:58
So rhetoric in the AI world is knowing how the footage will read, like, as you're shooting it, okay?
49:04
So convincing people that you are the good guy is, is a trained skill, and it's part of the, the, the what you need to do to be successful here, right?
49:12
It's, it's not an option.
49:14
You're going to… You're doing things on camera.
49:16
People are gonna see it.
49:17
People are gonna get access to the footage.
49:19
You are going to need to convince them, um, that, that, that you're doing the right thing.
49:24
I mean, uh, potentially we, we don't have anything resembling a democracy, um, in, in the years, in the coming years, and maybe, maybe that would make this less important.
49:32
But even then, there's gonna be decision-makers, and you're probably gonna want to present your side of the story in a, in a compelling and persuasive way, right?
49:41
So this means that the successful commander will position his machines not just to accomplish his mission, but to generate the camera angles to support his defense.
49:49
The direct action piece and the propaganda piece therefore are different types of warfare, different, different, different things that are necessary for the same operation.
49:57
And hopefully you see th- this type of multivariate a- analysis runs most efficiently inside one brain.
50:04
I mean, even the logic and moral judgment embodied in the, the predetermined code, uh, will likely be analyzed and need to be coherent, kinda line up with, with what you're doing, right?
50:14
And again, I'm not saying this has to be one person, but think about the difficulty there, right?
50:18
If the, if the person that's running your PR defense is not you, then you're gonna need someone that is basically irrationally on your side, right?
50:27
And, and, and so the skills bundle.
50:29
So if you cannot get all the necessary skills into one brain, you want the smallest viable team you can.
50:36
You know, a, a squad, ideally a marriage, right?
50:38
You, you want somebody that, that, that, that is deeply, deeply incentivized for you to win no matter what, right?
50:45
Um- You know, they're, they're as, as one of the people I was talking to, they said, "You know, uh, we, we … People make a lot of, of, uh, noises about how, how bad it is that, that women are irrational in some senses."
50:57
Um, there are advantages to, to women's irrationality in the right context.
51:02
Okay?
51:02
Uh, if you've got … You know, if you need someone to say, "Hey, I believe that this person is gonna be on my side," well, um, you know, a wife is i- it ca- at least can be that person, uh, for, for … in the right marriage.
51:15
And, and certainly that's what you would, that's what you
51:17
If you can't get them all in, in one brain.
51:20
Um, so So again, back to the social implications.
51:24
The point of this episode is to think about the implications of the leveling of the social graph, right?
51:28
The, the, the shrinking of the social graph.
51:30
When 10,000 specialties become a human-scaled, uh, number of skills, then the necessary crew fits in one room, right?
51:37
The question is what becomes… what organization does that crew need?
51:41
And the answer is, is pretty small, I think, and, and getting smaller.
51:45
So the other point that I wanna make is, you know, scale starts to lose its grip, right?
51:49
A $400 operator, when a $400 operator can routinely defeat the multimillion-dollar platform, um, then we, we have, we have a change in the game, right?
51:58
Uh, when, when the metric is, is the quality of the prompt and, and your ability to remember the internal logic of the pre-coded instructions that you coded, you know, a lot, a lot further back than last week, right?
52:11
Uh, then, then what that means is no amount of midwits can compete with one genius with a true grasp of the system.
52:17
It, it means that, that these peak users, these, these, these paladins, these prime users, those people are going to be, uh, incredibly, incredibly valuable, right?
52:27
The decisive unit is the man, not the budget.
52:30
Cap-capital is going to lose to tech in this instance, 'cause we're gonna need this man for so many things.
52:35
Every plant, every grid note, every hospital, every protected person is, is a post.
52:40
And, and behind, you know, each post is gonna need to be a, a team of people that, that make that guy work.
52:46
Those posts will multiply faster than, than sort of natural prodity- prodigies.
52:50
I think there will be many places that are backwaters, right?
52:52
That, that are unable to get their own active defense.
52:55
The point is that in every scenario that I can think of at this point, AI is a force multiplier and a social complexity reducer.
53:02
So the remaining roles, whatever they are… And, and again, I, I could be wrong about how many or which ones, um, but it's gonna be much, much smaller.
53:09
There's gonna be much, much fewer specialties, and the specialties that remain are going to be, you know, linked by AI into small teams that will need to, and therefore be able to understand the other skill sets intuitively, which is all that you need for cross-ranking, right?
53:24
There, there won't be all of these siloed little knowledge domains.
53:28
AI is going to eat the difference between those knowledge domains, and it's gonna be skill in using AI, skill in, in, in kind of operating in the world.
53:35
And, and again, I think it could be as simple as code, content, command.
53:39
And tho- And, and ideally, you know, wh- when we talk about rhetoric being useful for command, rhetoric being needed for command, what we're really saying there is these are just two aspects of one skill.
53:52
What we're saying, this really does kind of reduce to one ability to use AI well and to use the machines that the AI will allow you to control well.
54:00
Um, so we have a need.
54:03
We, we need to produce a bunch of people with a, with a unified skill set, a skill set that can be ranked basically on one, one leaderboard, right?
54:12
And we know how to do that.
54:13
That's called a cursus honorum.
54:14
The purpose of a cursus honorum is to produce as many men as possible who hold a distinct, hyper-valuable skill set.
54:20
It is, it is insourcing at civilizational scale, right?
54:23
Instead of importing or renting or just hoping that the rare capability appears magically and sorting for it, the polity says we're going to manufacture our own supply.
54:32
So when we, when you see these ladders, um, the point of them is this is not about gatekeeping people out.
54:39
It is to maximizing the number of trustworthy apex operators, okay?
54:44
We don't… Like, the reason why we have a cursus honorum is because we don't have enough of something, right?
54:50
And again, this is gonna, uh, as these tools converge, the number of distinct hierarchies collapses.
54:55
So thousands of separate ladders, thousands, uh, of separate siloed groups kind of fold into a single leaderboard.
55:02
A leaderboard is, is the, the, the, the term that comes from, like, the gaming world, right?
55:06
So, um, everyone is scored on the same board, similar to how video games in the same genre allow easy cross-ranking, right?
55:12
Um, and, and certainly within one game you can, you can have a leaderboard, but then people will start to understand, hey, skills translate between different games of the same genre.
55:21
So somebody that's, that's in a, uh, like a first-person shooter can move into a new first-person shooter, and maybe there'll be some slight variation.
55:28
But the broad tiers of skill and capacity are really durable, right?
55:33
So social complexity drops.
55:35
Rank becomes legible across every domain that the tool touches, okay?
55:39
And again, we're gonna dive deeper into the civilian applications, uh, next week, but hopefully you see that a single person who reaches the Prime level, who has all three skills, is incredibly economically valuable.
55:49
Uh, I mean, uh, hopefully as, as we've talked about this, like, yes, we were talking about kind of the, the, the military applications.
55:55
All of these things and a lot more have, have massive economic civilian applications, to the point where one successful trainee developing into a Prime m- will, will be, will create more than enough economic value, not just to pay for the training of the whole class, but potentially a lifetime of support for the rest of the cohort.
56:15
Now, that doesn't mean that we're gonna, like, that we're gonna uncritically support washouts, right?
56:19
That you're just guaranteed a salary for the attempt at becoming a, a Prime.
56:23
But the fact that the economics mean we could is really interesting, all right?
56:28
And again, uh, uh, one of the reasons why we've spent the last couple weeks, um, and we'll spend at least one more, uh, talking about this, uh, before getting into what I think the, the cursus honorum will look like is you can't build a ladder until you know what stands at the top of it, right?
56:41
How its value has, has, has captured.
56:42
So we, we… I want you to get a very clear idea of the, the person that's gonna be at the very top- of this, of this process, the, the true peak user, and then we, we can think together about, well, what would it look like if you have large amounts of capital, right?
56:57
'Cause, you know, I, I was talking with, with, with, uh, someone a- a- about this idea, and he's like, "Well, I mean, how many people have the, ha- have access to, to the capital even if they've got the native skill?"
57:07
And, and again, I think the gatekeeping concern is misplaced.
57:12
Like, if, if this is how this shakes out, if there is one, you know, basic group, one basic way of ranking people, and we need as many as we possibly can of the peak users, th- there isn't going to be enough room.
57:27
There isn't going … You know, sort of, sort of like, uh, um, it, it, it's unbelievably stupid to have, um, somebody that is capable of, of, of wielding the technology at, at the level that they're capable of, um, when, when, when people like that are scarce, and then to just keep them down.
57:44
A- a- I mean, this is true.
57:45
You know, people even, even during, you know, the, the, the much maligned aristocratic, you know, world.
57:50
Um, so for most of medieval Europe, if you could … You, you could walk into a, a bunch of different, uh, places, guilds, training facilities.
57:57
If you could hold a sword out straight in your hand and do this kinda like isometric exercise, you could hold a full broadsword for, I think, 30 minutes, they would immediately take you as a knight, no
58:08
uh, as a, as a squire and train you to be a knight, no questions asked.
58:12
Because that ability to wield a sword was so valuable, right?
58:17
That … Just the physical toughness to be able to do that was so valuable, we wanted … Like, we don't care where you come from, right?
58:24
Um, now, when it comes to, like, bloodlines, we're gonna invest in certain bloodlines 'cause they have the highest rate of producing people like this, but if somebody shows up that's got the goods, well, man, we, we want those people.
58:37
That's what a healthy system looks like.
58:38
That's what a system that is, that is based in objective reality looks like.
58:41
It's not how managerial systems operate.
58:44
Um, but I mean, and even, even that, like, it's not how our managerial system operates in its last legs a- at, at the end.
58:50
Uh, but so that's why we're talking about this.
58:52
And so be thinking, okay, as we're thinking about this, if this is what the peak user looks like, what will training systems look like?
58:59
What will, what will it … What will the, the, the way when we reorganize our society to facilitate the creation of more of these people, when we start searching for them, how do I get my kids To be able to, to, to do well in that structure.
59:14
How do I train them?
59:15
How do I give them the best possible chance to either be one of these people or effectively serve one of these people, right?
59:22
So again, Rome's name for the institution that made capable, trustworthy men was the cursus honorum.
59:26
It was a ladder of offices that tested a man at every rung before he climbed.
59:30
Now, uh, next week, part three will turn to the civilian economy, right?
59:33
Which again, I see… think that the, the Prime reshapes, and I think that'll be heavily influenced and maybe even directly subsidized by how you train a Prime to be a child.
59:41
And, and this is the logic that will demand a cursus honorum as we move into the AI future.
59:45
So thanks, everybody.
59:47
Uh, this concludes the, uh, the unpaid portion of, uh, of this discussion.