Latest / Great Houses / 30. Cursus Honorum of the AI Future Part 3

Transcript

30. Cursus Honorum of the AI Future Part 3 · Great Houses

0:00
Gregory Treat: All right, everybody, thank you so much for joining me, uh, for
0:03
episode thirty of The Great Houses Forum.
0:05
So we're continuing to discuss, uh, the, the cursus honorum of the AI future
0:12
is what I've, what I've titled these.
0:13
And, and, you know, again, basically what I'm, what I'm trying to answer
0:17
is the question of what does it look like for, um, there to be…
0:23
You know, we're gonna, we're gonna have some profound changes
0:25
in our world that come from AI.
0:28
What, what do those look like?
0:29
And especially, you know, the, the ultimate thing that we're gonna try to
0:32
get to here is what does that change in terms of how do we raise our children?
0:37
How should we think about that?
0:38
Are there things we can do to pos-- position ourselves?
0:40
And I, I think they-- that we can, and so that's, uh, that's what
0:45
I'm hoping to bring to you guys.
0:46
So last time we talked about how I think one of the key things we're gonna
0:50
see here is a collapse in specialties.
0:52
And right now we have this hyper socially complex, uh, world.
0:57
We have thousands of occupational roles, uh, maybe tens of thousands.
1:01
I mean, again, it depends on kind of your, your opinion of many of the,
1:05
uh, the PhD programs that our fair society produces at the present time.
1:09
But, but certainly in the thousands and maybe in the tens of thousands,
1:12
um, of roles that are just necessary to keep our machine, machine going.
1:17
Um, and that's, you know, very different from earlier societies.
1:20
You know, hunter-gatherers had, you know, a few different distinct roles.
1:24
And then, you know, as we've kind of, we've kind of grown in complexity, uh,
1:28
we ha-- we have more differentiation, more specialization in our social
1:32
arrangement than, uh, I think anything since the Roman Empire, basically.
1:37
Uh, which, which is, you know, one of the reasons why people continually
1:40
make comparisons to the Roman Empire.
1:42
And I, I think that we're gonna go down to-- back to a human
1:47
scale number of specialties.
1:49
And, and I think that there's going to be certain key specialties with AI, certain
1:54
master-- certain skills that if you master them, you will be hyper dominant in a
1:58
whole bunch of different fields, right?
2:00
And so I think the, the main skill sets that we need to think about are code,
2:04
content, and, and what I'm calling command, mostly so that it alliterates.
2:08
Um, but, but this idea of can you organize your thoughts logically?
2:12
Can you give a, a description and an architectural flow that a, a machine,
2:17
an AI can take and turn into, you know, machine-perfect language?
2:21
Then content, can you understand rhetoric?
2:23
Can you understand how to interact with people?
2:25
And then command, can you make a decision and, and own the results
2:28
and, uh, and be able to, to kind of absorb the, the socially necessary
2:33
function of responsibility?
2:34
So that's what we're gonna be talking about today.
2:35
And, and one of the words that I have, uh, for that is, is the AI paladin or,
2:39
or somebody who's, you know, somebody who's got all three of those skill
2:42
sets, we might call them prime, right?
2:43
They're a triple threat.
2:45
Um, code, content, and command.
2:48
Oh, goodness.
2:49
Um, it, it might take some time to recover from the compliment
2:53
that Joshua just paid me.
2:55
So, uh, one of the other things we talked about last time was, you know, we talked
2:59
about tanks, we talked about fighter jets, we talked about what those things
3:03
look like in, um, in the AI context.
3:08
And, and again, and I'm gonna say this probably ten times today,
3:11
it's not that X is obsolete.
3:13
It's not that tanks are obsolete.
3:14
It's not that fighter jets are obsolete.
3:16
It's that the cheap integrated unmanned unit is as good or better, um, especially
3:22
if, if there's kind of a, the, the willingness to, to be suicidal, right?
3:27
If a, a drone that is willing to, to kill itself flying into a missile or into the,
3:32
uh, the jet engine of a fighter jet can really, uh, compete, can, can make it very
3:37
difficult for, um, for traditional or the existing kind of military setup, unless
3:43
you have flanking units of your own, which are gonna be equally cheap, right?
3:47
But once you've built those things, once you have… Well, in order to have a jet,
3:51
we have to kind of have an, a cloud of AI defenders, uh, AI drone defenders,
3:55
anytime we enter an actual theater of war.
3:57
Well, then you're gonna ha- start having discussions about, well,
4:00
are you piloting the jet yourself or is an AI piloting the jet?
4:05
Are you manning the tank, you know, controlling the tank yourself, or is
4:08
an AI controlling the tank the same way that it's controlling all of the
4:12
flanking units that, that are, you know, that don't actually have space
4:14
inside of them for humans, right?
4:16
Um, and so this is gonna change things.
4:19
Again, it's not that tanks are obsolete.
4:20
It's not that fighter jets are obsolete.
4:22
It-- Almost none of the things that we understand as being, you
4:26
know, dominant weapons platforms or dominant useful things.
4:30
If they're useful, they're gonna continue to be useful.
4:32
It's just that they're, how they are interacting with humans is gonna change.
4:36
Okay?
4:38
So the scarce thing is about to become people who can coordinate the AI that
4:42
controls the robots, which will shortly include, you know, like I said, uh, light
4:45
screening robots and things like the tanks and the fighter jets themselves.
4:48
Okay?
4:49
You know what, what I said about tanks also applies to people, right?
4:52
Humans are not obsolete.
4:54
I, I, I am, I am coming to the point where I'm, I'm beginning to become
4:58
increasingly frustrated with people who, who are saying, well, the-
5:02
there's, there's not… You know, are we gonna have humans in the loop?
5:05
I assure you, there will always be humans in the loop, okay?
5:09
There will always be someone who initiated the thing.
5:13
I-- You know, we, we-- There was this recent thing where allegedly, um, a,
5:18
a, an OpenAI model, uh, was trying to pass a test, and it couldn't pass
5:22
the test to its satisfaction, uh, in, in this sandbox environment.
5:26
So what it did was it broke out of the sandbox environment, got access to the
5:30
internet, went to Hugging Face, which is a, a, a, a website from which you
5:35
can download many different, um, AI LLMs, and, and attempted to download
5:40
other LLMs and run it on Hugging Face's software, uh, is my understanding, or
5:45
their servers, in order to get another AI to help it answer the question, right?
5:48
So it, it did all of this, this incredible, you know, software
5:53
exploits in order to cheat on a test, is the, the, the way that,
5:56
uh, the people are framing it, okay?
5:58
What I want you to understand is there was still a human in the loop.
6:00
There was still a human that was making the decisions to ask some computer
6:04
to do something, to do something, to do something that resulted in that.
6:07
That it wasn't-- These things don't do anything on their own.
6:11
Now, the, the, the methods by which they may attempt to fulfill the directives
6:16
they have been given might surprise us.
6:18
They might not be, you know, obvious to us why it's doing that.
6:21
But, you know, it's a, it-it's an incredible thing that,
6:24
that's happening here, okay?
6:25
So, so but there is always a human in the loop.
6:28
Like, that, that, that's the point that I wanna make.
6:30
There is always, at some level, um, a, a human being that is, is going to have
6:35
been the originator that, that start, that starts this, this process, okay?
6:41
so I want us all to ask the question, so where is the human, right?
6:44
When, when you say there's no human in the loop, when you, when you talk about
6:47
automaticity or you're saying, "Well, this is fully automated," what you're saying
6:50
is the humans that initiated that process are in some way, you know, insulated
6:55
from responsibility or liability, okay?
6:57
That's really what you're saying there.
6:59
That, that, that… A-and probably you're, you're, you're following some
7:03
kind of social rules set by the corporate environment that, that we live in, by
7:08
the social environment that we live in, and you're saying, well, these
7:10
humans are sufficiently separated from the work that they're not responsible
7:14
for what their AI agents are doing.
7:17
And, and, and I basically think that is, that is just gonna get massively,
7:21
massively tested, and people are gonna-- When, when, when that attitude results in
7:26
claims of, well, we're not responsible for these massive harms that have been done
7:30
by AI, I, I think that that legal doctrine is just gonna, just gonna go away.
7:35
I don't think it's gonna be able to, to stand the test of time.
7:38
So we can, you know, when we're talking, thinking about AI or thinking
7:41
about knowledge work, we can make an analogy to hand tools, right?
7:44
Uh, physical work began with hand tools.
7:46
You had, you know, you're carving wood and stone, you're shaping metal with hammers.
7:51
Um, the craftsman hel-holds the entire act.
7:53
Like, he, he is the person doing this.
7:56
It's his hand.
7:56
He, he, the, he has incredible fine detail control.
8:00
And you look at these old sculptor, uh, tools, they had, like, a dozen
8:05
different tools for, for a, a huge amount of, of different ways of
8:10
making cuts on a piece of stone, okay?
8:13
And it was, you know, slow, it was intimate, it was
8:15
limited by muscle and time.
8:16
By its very nature, you know, every piece was custom.
8:19
It was highly specialized.
8:20
You know, you couldn't, you couldn't make a mass-produced piece, right?
8:24
So then we got to power tools, right?
8:26
Lathes, power hammers, presses.
8:28
Um, so at that point, you are-- you still have a human at every workstation,
8:34
but it's, it's not their strength.
8:36
It's not their muscles.
8:37
I mean, their muscles are still involved.
8:38
They're still guiding the machine, right?
8:40
But y- the, the main skill was understanding the strengths and
8:43
limits of the machine, right?
8:44
And probably you're gonna, you're gonna place a high value on keeping
8:49
the tooling in place, right?
8:50
So we, you know, in this, the, the, the early powered lathes, people
8:54
figured out incredible ways to cut things without having to stop
8:59
and change the head on the lathe.
9:01
Like, they… And, and that was a r- there was, there was a,
9:03
a incredible premium on that.
9:05
Um, 'cause changing the head of the lathe obviously was a, was
9:08
a massive human-driven process that, that, that way slowed down
9:12
whatever you were doing, right?
9:13
And then ultimately, this is what led to the assembly line.
9:15
So if you wanted to do something with one head, you had it at one
9:18
station on the assembly line.
9:20
And if you needed a different head, if a different, different w- tooling
9:22
on the lathe would be better, you just shift it over, you know, to the
9:26
next, uh, station on the, on the line.
9:28
And then, you know, that's how we get the, the assembly line.
9:30
You get, you get jigs, right?
9:32
So in, in initially, you had kind of mechanics that are, that are
9:35
shaping metal on, on kind of free lathes, and then you have jigs,
9:39
and then you have automated jigs.
9:40
So now the job, instead of actually guiding the machine, is just, well,
9:45
we're-- you gotta watch the machine, and if it starts to break or if something goes
9:49
wrong, you've got to jump in and fix it.
9:51
Um, but you're doing less and less, and then ultimately, you know, as, as these
9:54
machines became more reliable, we were able to step away more and more, right?
9:58
Um, and then you get to, you know, the modern world, CNC,
10:01
industrial metal shaping.
10:02
You have program paths.
10:03
You have a computer controlling all of it.
10:05
So the comp-- the human is writing the program, sets the job, and then, you know,
10:08
he's, he's responsible for any screw-ups.
10:10
Um, but, you know, you still have guys on the shop floor, okay?
10:14
And, and that, and that's again, one of the points that I wanted to make, right?
10:17
Like, the fact that no one is-- no, no human is making the cuts, right?
10:22
There's no knives involved, does not mean that humans are not in the loop, right?
10:27
No CNC machine has ever just decided to cut something, right?
10:32
It's-- it, it, it follows the pre-programmed paths.
10:35
There is still a human in the loop.
10:36
And so I think knowledge work is going to, uh, go through the same trajectory.
10:40
We, we might think of unaided drafting, research, judgment.
10:44
These are the, the hand tools of the mind, right?
10:46
And then we get, you know, spreadsheets, search, early automation templates.
10:50
These are, you know, roughly analogous to power tools.
10:53
And AI might be the CNC of knowledge work, right?
10:56
That, that, that might be what we're about to all go through, um, which is
11:00
obviously gonna, gonna result in an enormous change in what we're doing.
11:04
But again, humans are not obsolete.
11:06
It's just which humans and where are they sitting in the stack, And
11:09
there's that old, old IBM line, "A, a computer can never be held
11:12
accountable, therefore, a computer must never make a management decision."
11:16
Right?
11:17
And that is where we find ourselves, right?
11:19
There, there… And, and, and I think that there's a lot of, there's a lot
11:22
of content that goes by the name of management that is not really management.
11:26
It's just sort of shoving risk around.
11:28
Um, and that is, you know, shoving risk around if your job is to
11:31
push paper, um, then, then m-much of that the AI can do, right?
11:35
It can generate reports.
11:36
It can say, "Well, if this then that." It can say, "Well, if we ch-choose A, B,
11:40
and C, here's the different risks that are associated with that." It can do
11:42
that, and it can do that really, really fast, especially if it's given access
11:46
to, you know, research s- um, research systems and your business systems.
11:50
But it can't actually make a decision , right?
11:53
AI can never say, "Okay, this is, this i-- Or, I mean, and when it can, um, I
11:59
think we have to recognize it's, it's wildly, wildly more expensive from
12:03
like a token perspective. Um, so if you, if you say, "Hey, AI, I want you
12:07
to give me the top three or the top five or the, or the top ten options
12:11
and build me out a comprehensive plan.
12:14
Tell me what all the risks are.
12:15
Lay it out for me," right?
12:17
That's a certain amount of tokens.
12:20
And to get it to tell you, "No, this is the only thing you should do,"
12:24
is incredibly more expensive, you know, depending on who you talk to.
12:27
Some-somewhere between a thousand times more expensive and like a million times
12:32
more expensive in token usage, right?
12:34
It is very, very, very complicated and very hard for this machine
12:39
to figure out like how… W-what, what is the final decision?
12:44
Yeah.
12:44
And even then, it, it, um, you know, if you, if you have any kind of safety
12:49
guidelines at all, it's much easier.
12:50
Even if you say, "Well, narrow it down to one out of the t-top ten," um- It's
12:56
much easier for it if you say, "Okay, now, now ask the human to approve."
12:59
That wildly reduces the token cost.
13:01
Oth-oth-otherwise, you know, you get, you know, bad outcomes.
13:04
That's just how it's programmed.
13:05
So, um, so the things I'm gonna try to-- I'm gonna attempt to convince you
13:10
of today is, is that humans, number one, are not obsolete in, in either
13:13
civilian or military applications.
13:15
There will still be a human in the loop.
13:16
Um, number two, AI is going to reshape every industry.
13:20
I'm gonna kinda set out how, how I think that's gonna, gonna go.
13:23
And, and, and, and basically the, the point there is, is that a, a, a
13:27
human plus an AI is going to be better than a very large number of humans,
13:34
um, in almost every industry, okay?
13:36
In e-- in every industry that I, I can think of at this point, right?
13:39
Um, and then the third point that, that I'll, I'll, you know, just kinda
13:42
note at, at different, different places is that the-- these core
13:45
remaining skill sets, when you shrink down the, the, the complexity,
13:49
all of these skill sets basically have military applications, right?
13:52
All of these skill sets are things that are also useful and valuable
13:57
in the, in the military context.
14:01
Um, and so you, you, you have less and less separation.
14:04
You have less and less of a civilian track and military track, and you
14:08
start having more competence tracks which have a, which have a, a variety
14:12
of different, uh, implications.
14:14
And again, I think, there's a lot of stuff, and maybe I'll
14:16
take an episode and go into it.
14:17
There, there is actually a great deal of social problems that we suffer
14:22
through right now because there is so much of our society that is completely
14:28
insulated from military applications.
14:29
There, there is no overlap in the social or the economic reality between
14:35
the people who, um, do many things in our, in our, you know, academic sector,
14:41
in our finance sector and, and the, the guys that are at the sharp end.
14:45
That's a bad thing for our society.
14:48
That's a… You know, there was this famous, uh, quote from a, from a poem,
14:52
um- His name escapes me right now, but basically there's a, there's a, there's
14:56
a guy who's talking to, to a, a Muslim ruler, a caliph, and he says-- and the,
15:01
and the caliph says, um, you know, "What will happen if someday a people arise who
15:07
their, uh, their scholars and their, their warriors, you know, don't, don't talk to
15:11
each other, don't, don't know each other?"
15:13
and the, the guy responds, " Even if they go to the moon, those people will
15:18
be a blight on the world." Which is a fascinating, a fascinating thing for a
15:22
guy in the, in the medieval period to say.
15:24
I, I have questions.
15:26
All right.
15:26
So moving on, um, I think the, um, the first thing that I want to, um,
15:32
the first vignette, I'm gonna share with you several vignettes, uh, that,
15:35
that I, I believe are showing this, this tendency, this, this way forward.
15:39
So on May 11th, twenty twenty-six, just, you know, a month, month or
15:42
two ago, uh, the a16z show aired an episode titled Marc Andreessen on
15:47
Builder Culture in the Age of AI.
15:49
So about twenty-seven minutes in, they're talking through AI's effect on
15:52
programmer productivity, AI vampires, which is a hilarious concept, and
15:56
you should-- you'll go listen to it.
15:57
And the-- you know, then they're talking through, you know, a sketch
15:59
of kind of future tech company jobs.
16:01
Andreessen, you know, starts talking about, okay, the jobs
16:03
themselves are changing, right?
16:05
And he describes how programmers, product managers, and designers are
16:09
collapsing into a single specialty.
16:11
Now, Marc Andreessen, of course, co-founded Netscape and then co-founded
16:14
Andreessen Horowitz, which is, you know, there's sixteen letters between the A
16:18
and the Z, and that's where they get the, the acronym that they use, with Ben
16:21
Horowitz in, in two thousand and nine.
16:23
Uh, the a16z portfolio includes Airbnb, Coinbase, Slack,
16:27
GitHub, a long list of unicorns.
16:29
Um, it, it's one of the top venture funds of our time, managing on the
16:33
order of ninety billion in assets, um, including its most recent round,
16:36
raising fifteen billion dollars in new funds this last January.
16:40
So this is a very successful investor.
16:43
This is a guy who is deeply tied into the tech world.
16:46
You know, this is a guy who is seeing this stuff happen in real time, and he, he has
16:50
both the, the legal, uh, right to know and the-- and clearly the, the intellect
16:54
to understand what he's being told.
16:55
So quote, "I'm seeing it in a bunch of the early, uh, leading-edge
16:59
companies in the Valley.
17:00
They're circling around a job title loosely called builder or something like
17:04
it." And basically, the, the idea is that you had these separate jobs in the
17:07
past, the programmer, product manager, and designer, and then him, him and
17:11
the, uh, the, the other host kind of go on a side quest for a little bit.
17:14
But later on, he comes back to the topic and he says, quote, "Sort of this
17:17
three-way Mexican standoff where the programmers think that they don't need
17:20
the product managers and the designers anymore 'cause they can have AI do that,
17:23
and then each of the other two doesn't think they need the other two either."
17:27
Uh, and he concludes, "I've been predicting that they're all correct.
17:29
The product manager can generate code and design now, and each of them can do
17:33
the job of all three. The idea is the job cha-- jobs change, so now the job is
17:37
builder." Builder, which is a great word.
17:40
Um, I, I still think it's, it's, it's too small.
17:44
Um, you know, I still think there's too much identity, there's too
17:47
much limitation in, in saying these people are just builders.
17:51
You know, my word for this is, is the code skill, right?
17:53
And I think that this collapsing of silos is not gonna stop, right?
17:56
There's gonna be more and more and more things that are distinct specialties
18:01
right now, and they are gonna all get swallowed by this mega skill, uh, which,
18:06
which is basically reducing to people who can architect logical flows and,
18:11
and communi-- make the big picture architectural decisions that AI can
18:14
then turn into a bunch of things, right?
18:16
The scarce unit here is the human who can define intent, right, define that
18:20
logical flow, make the human-to-machine interface effective and pleasant.
18:23
That's the UI, UX, the designer component, and then allowing
18:26
high-quality human decision-making at the necessary scale, uh, and speeds.
18:30
You know, that's the, um, that's the, the product manager component, and
18:35
then carry that loop into, you know, various industries: marketing, medicine,
18:38
law, manufacturing, and agriculture.
18:39
So, um, I'll g-give you a marketing vignette, right?
18:42
On July third, twenty twenty-six, the beginning of this month, Alina
18:46
Uverova, a CMO and co-founder based in Portugal, published a
18:49
first-person account on LinkedIn.
18:51
For three months, she ran marketing with AI agents, and two of those
18:54
months ran near full autonomy, right?
18:56
The team, in her words, was one marketer, herself, and the agents.
19:00
The quote: "For three months, I ran my marketing with, with AI agents.
19:03
Two of those months ran on near full autonomy.
19:05
The team was one marketer, me and the agents." her monthly output was
19:08
twenty blog posts, a hundred and ninety-five social posts across seven
19:12
platforms, four newsletters, forty-three influencers in the pipeline, four
19:15
hundred and thirty Reddit comments, and then two always-on ad accounts.
19:18
And her results were she got seven times organic referral, uh, organic
19:23
results, ten times referrals, thirty percent lower cost per lead at the same
19:27
spend, and a hundred and thirty-five thousand views on Reddit, which I would
19:30
view as a, you know, personal failure.
19:31
But, you know, you're-- so no accounting for taste.
19:34
While product changes may have mixed in the, into the metrics, this is,
19:37
this is an incredible accomplishment.
19:39
though I will disagree with her in one sense, right?
19:42
This is not autonomous AI marketing.
19:44
It is a system that multiplies the output of a high-quality
19:47
human decision-maker, right?
19:49
The staff department, like if you had a marketing department, they are
19:52
going to lose on volume and iteration speed before they lose on taste.
19:55
A-and they are gonna lose on taste as well.
19:57
but y-you're seeing-- hopefully, you're seeing this same three-skill thing.
20:02
You have a code that builds the harness, you have content, that's the payload,
20:05
and then there is a human intimately involved in the design choices.
20:09
So there's, uh, another marketing, uh, vignette is Timofey, uh, I don't know
20:13
how to pronounce that last name, but, uh, Aiden's lab, which had tens of
20:16
thousands of, of AI ad and Mr. Powerz with a Z, their, their case study
20:23
crossed a hundred and twenty thousand dollars per month within six months.
20:26
And then this is the same pattern as Uvarova.
20:28
They had one human governing a content factory.
20:31
I use, uh, AI to research and do drafts.
20:33
Usually, it starts with research that I pulled from books or papers that I've
20:36
read, or depending on the subject, tweets.
20:38
The LLM takes my outline, goes and does further research on the topics that
20:42
I've given it, and generates initial drafts, which I edit intensively,
20:44
both at a line-by-line level and increasingly at a structural level.
20:48
So one of the cool things that's been happening recently is I'll, I'll go
20:51
through and I'll find these AI-isms, and I'll tell it, "This is wrong.
20:56
Fix it throughout the whole rest of the piece." And
20:58
increasingly, it actually g-- does.
20:59
It goes through the rest of the piece as I'm kind of going through line
21:02
by line and, and, and it fixes it.
21:04
Does a good job.
21:05
So, and, and I think part of this is, you know, the way I frame it, the mental
21:09
framework that I have for this is, uh, this is essentially the same workflow
21:12
as the law firms I was trained for.
21:14
You know, you have a paralegal or a junior associate.
21:16
And by the way, the, the law firm as I was trained for it in law school doesn't
21:19
really exist anymore, which is hilarious.
21:21
but you have the paralegal or junior associate which finds the
21:23
case law and drafts the brief.
21:25
It kind of pulls together, you know, the, the, the different examples, and then
21:28
the senior lawyer revises, reframes, and puts a final spin on it and then signs.
21:32
So, uh, you know, my writing stack is that pattern.
21:35
AI does the grunt work of putting connective tissue between my big
21:38
ideas, but it is my content and, and I think fairly noticeably
21:41
different from anybody else.
21:43
Um, now I don't use AI very much in my actual legal practice.
21:46
A part of that is getting models and hardware where I can prove that
21:49
the data isn't processed in a way that violates the ethical rules,
21:52
which is possible but cumbersome.
21:54
Um, and, and it's irritating to me 'cause I use, you know, frontier
21:58
models on, on the content stuff.
22:01
It's irritating to me to use the, the, the legal models, um, the, the ones that
22:05
are kind of specifically designed for legal use 'cause they're like, you know,
22:07
two, two or three iterations behind, uh, where the frontier models are.
22:12
So a-another part of it is the specific types of law that I practice.
22:15
So though many, like when in estate planning or transaction work, these
22:19
are already highly templatized fields.
22:22
Um, so you know, a template is basically we've, we've reduced this, this workflow
22:26
to a series of questions that you answer.
22:28
Um, and it's, it's pretty efficient, right?
22:31
So the speed gains are not that great, uh, at least not yet.
22:34
Um, and you know, much of document produ-production, you know, speaking
22:38
of templates, it, it, it behaves like an AI, AI-- like a template
22:41
filled from a fact pattern.
22:42
So, uh, the, the AI can make that explicit.
22:45
It can reduce the deal, the pleading of the form to a set of clean specs,
22:48
and then you have it, the separate generation pass, which then writes
22:51
the document from those specs.
22:53
Um, and that's much faster, uh, than retyping the prior file.
22:57
So you have very interesting, uh, very interesting claims being made on Twitter.
23:01
So on April ninth of this year, a guy named Hello-- @helloparalegal on
23:05
Twitter posted a Harvard Law solo using Claude Code to ship a thirteen-page
23:09
intelligence report that beat pitches from fifteen attorney firms.
23:14
Um, and, and, you know, there's number one, this is sales and
23:17
business development content.
23:18
Uh, so, you know, you always take that stuff with a grain of salt.
23:21
But I, I mean, I, I think it's really interesting that he's able to put
23:25
something together that's competitive that, that clients are liking.
23:28
Um, and, and I would also note, you know, parenthetically, that there are
23:32
obvious implications to being able to pull together an intelligence report.
23:35
Okay?
23:36
Another example is, uh, GetDynasty, which is a qualified small business, uh,
23:40
stock, uh, which is a way of basically not if you, if you start your business
23:45
as a C corp instead of as an S corp or if you convert, um, uh, appro-- at an
23:50
appropriate time, uh, there, then you are able to not pay, uh, capital gains on a
23:55
certain amount of your, your net value if you then later sell the business.
23:59
So QSBS is a, is a big area.
24:01
Um, now I, I, uh, wanna explicitly say that, uh, q-- the GetDynasty.com,
24:06
the framing that they're taking is, is more aggressive than I
24:10
would recommend for my clients.
24:11
And so, you know, they're-- I think, uh, we will all find out in the next couple of
24:16
years if the IRS disagrees with the, the, the, the, the stance that they have taken.
24:21
And if the IRS successfully disagrees with them, well, then the-- it'll be
24:24
really interesting to find out just how similar the AI, um, generated
24:29
stuff for all their clients was.
24:32
but what I want, what I wanna think of it for you is, the, the way I want, I
24:35
want you to think of this example is there is still a lawyer involved in this.
24:39
This is essentially a lawyer who figured out, a lawyer pr-probably plus some,
24:43
some software guys that figured out, hey, this is turning into, this QSBS
24:49
space is turning into a volume business.
24:51
It's turning into a business where I can't really make money doing
24:57
high-level expensive bespoke work.
24:59
you know, which means I can't make money structured as a law firm, right?
25:03
So what I wanna do is make an AI product that can do the volume piece, right,
25:08
without having to s-- get a bunch of funding and spin up, you know, a, a big
25:11
law firm with, with bunches of attorneys kind of churning, which leads to its
25:15
own, you know, quality control problems.
25:16
Pro-- frankly, like, if you were to just go to, you know, Robert Half or one of
25:20
these, you know, attorney, uh, document review kind of aggregator firms where
25:24
you can go and just, you know, if you have enough budget, you can go and say,
25:26
"Hey, I need five hundred attorneys to be able to log into this portal and,
25:30
and, and review documents," right?
25:32
And they'll, they'll give you a number, and they've got those attorneys, right?
25:36
But the, the quality of those attorneys, you know, with respect to them, um, and
25:40
I've worked in some of these subpoena mills in the past I, I, I would question
25:44
if it would be, if it would be worse than what the AI is gonna give them.
25:48
Okay?
25:49
So there, there, there is that.
25:50
But I also think that, like I said, there, there's, there's some interesting
25:53
things with, uh, with Get Dynasty and it-- and, and, you know, we will have to
25:58
see big picture if the bet that they have made on the, the, uh, on the QSBS, uh,
26:03
s-trust stacking, it works out for them.
26:06
another example and one that I, I think is probably a little bit more, I, I,
26:10
I don't have as much concerns about just the fundamental risks that they're
26:14
taking is FundLaunch, fundlaunch.com, uh, which opens, like you go to their
26:18
website, it opens as a chat skin.
26:20
You can pick among six to eight fund types: real estate, buyout,
26:23
long/short, private credit, search fund, et cetera, crypto funds.
26:27
And, you know, from-- you, you, you're interacting with this chatbot, so you
26:30
kind of s-select what type of fund it is, you select how big it is, and then
26:34
it asks you questions kind of in like a, like the, uh, like ChatGPT, right?
26:38
Where you go and you just sort of do a, uh, a brain dump on ChatGPT.
26:42
You just give the brain dump to this, this FundLaunch AI, which I, I, I, it,
26:46
I think is a wrapper of one of the GPTs.
26:49
and it figures out what you want, right?
26:51
And it, and it generates documents, right?
26:53
And so they have somewhere, uh, you know, s- I saw some, something to
26:56
the effect of like, we, we-- "This is reviewed by actual attorneys,"
26:59
which reviewed is a fascinating term.
27:02
I'm not sure that it has a hard legal definition right now.
27:05
the, the-- I, I would assume that there is a, a lawyer certainly
27:08
before they file it, right?
27:10
So there's, there's an add-on price.
27:11
They'll give you the, they'll give you the drafted docs, and then if you want them to
27:14
file it, then they're gonna have an actual attorney come in and kind of review it.
27:17
I think the most interesting thing to me about FundLaunch is You know, most
27:21
of the way that people frame the AI discussion, the way I often frame the
27:25
AI discussion of what it's going to take is, well, the human interaction
27:29
pieces are going to stay, right?
27:31
But FundLaunch is taking a lot of the human interaction
27:34
out, and people like that.
27:36
Like, that's the thing that I find.
27:38
The people that really like FundLaunch, many of them, many of them are people
27:42
that find interacting with, uh, with the types of high-powered attorneys that
27:46
do fund structuring deeply unpleasant.
27:49
And so they really enjoy just talking to this, this chat that is very nice
27:55
and unfailingly polite and always encouraging and tells you that you're,
27:59
you're, you're brilliant and, and are going to be successful and gets
28:02
all the information out of you over a timescale that, that, that a lawyer just
28:07
probably couldn't make money if they spent that long with a client, right?
28:10
In ma- in many ways, these kind of high-powered attorneys are finding
28:13
themselves in the, the same type of situation that doctors were in, you know,
28:17
twenty, thirty years ago as they were, you know, starting to really push the, you got
28:21
to get the interview down, you know, the, the time with the doctor down to fifteen
28:25
minutes and then five minutes, right?
28:26
That's, that's-- They're not doing that in many ways, or the, the firms
28:30
are resisting that, and they're, uh, they're better financed than, uh, than
28:33
the, than the doctoral group, the, the medical groups were at the time.
28:36
But they're, they're under that same kind of economic pressure, You
28:40
know, and another way to think about this is, you know, every business is
28:42
becoming a content business twice over.
28:44
Once in marketing, right?
28:45
You gotta, you gotta generate demand, you gotta solve the distribution problem.
28:49
And then a second time in compliance, right?
28:52
Footage or any other computerized artifact is becoming the permanent
28:55
record of what was sold, what was done, and how it was done, right?
28:59
And, and, you know, again, as I've noted in the past, the software
29:02
that finds a camera angle for a narrative mirrors the software that
29:05
finds a firing angle to a target.
29:07
This is the same geometry.
29:08
And, and a-another, another thing to, to think about is that the same loop applies
29:13
to co- applies to copyrighted works.
29:14
So, um, you, you extract the specs, and then you regenerate from the
29:19
specs alone in a separate AI process.
29:22
And, uh, for copyright purposes, that regenerated artifact is, is,
29:25
at least under current law, legally independent of the prior file, even
29:29
when it does all the same things, right?
29:31
Um, so templates were already doing this in the, in the areas
29:34
of, of law that could be templated.
29:36
Um, but AI makes the, the kind of the spec-to-work loop cheap enough
29:39
to change the business model in, in a whole host of, a whole host of ways.
29:43
So let's talk about medicine.
29:45
Um, in, in, also in April of this year, Doximity OpMed, uh, run by Oren Foss, an
29:51
MD who runs a solo cataract micro practice in Denton, Texas, which is, you know, very
29:55
close to me, has zero full-time employees.
29:58
AI routes incoming faxes into his EMR.
30:01
Uh, he self-schedules patients, and he automates the insurance verification.
30:05
He aims at about five new patients and ten cataracts a week and
30:09
gives each patient about an hour.
30:10
So that-- this actually, this model has, has allowed him to spend more time with
30:16
the clients because he's spending almost no time doing the paperwork, right?
30:20
Uh, it's one guy plus this AI harness, and that's, that's taking all of the, um, all
30:26
of the other people out, which means he doesn't have to do the volume that he did.
30:30
So again, in, in law, the, there, there's this, this pressure to
30:34
spend less time with the client.
30:36
In medicine, they've kind of hit their, hit their rock bottom, and
30:38
AI is kind of helping them come back to spend more time with the client.
30:42
Uh, similar things are happening in, in manufacturing, right?
30:46
So there's a company called Figure AI.
30:49
They have a, um, they have a factory called BotQ, and they,
30:54
they, they, they have a robot they, they're calling the Figure 03.
30:58
And they were able to ramp up the production of their robots from one
31:01
robot today-- per day to one robot per hour, about twenty-four times in under
31:06
a hundred and twenty days with more than three hundred, uh, fifty robots delivered.
31:09
And the way they did that was they took their humanoid robots
31:12
that they're building, and they put them into the workflow.
31:15
So, um, so their, their Figure robots, the same robots that they're selling, are in
31:20
their factory assembling key components and moving material between stations.
31:23
So you have robots building robots.
31:25
And, and, um, you know, it's a really interesting setup.
31:27
They have a custom MES across a hundred fifty plus workstations
31:31
plus that hybrid line.
31:33
And, and, you know, one of the things that we're learning is that, that
31:37
the AI robot stuff is gonna change how we organize our factories.
31:44
And I, I would argue that it's gonna have a similar impact on, on our corporations.
31:48
And this has happened before, right?
31:49
A century ago when, uh, when we moved from, from steam, um, steam engines
31:55
or sometimes water wheel engines.
31:56
The way factories were set up, you had one kind of source of mechanical power
32:01
on the wall kinda coming in from either this-- the boiler room or again, from like
32:05
a water wheel, and you would mechanically attach to whatever you're doing to
32:11
that big, that big mechanical source.
32:13
There's a-- If, if anybody's seen, you know, the old, uh, the old,
32:16
um, Chitty Chitty Bang Bang, right?
32:19
With, where he's got his, he's got his windmill, and inside the windmill,
32:22
he's got just a ton of stuff that's mechanically hooked up to that windmill
32:27
and all these gears, and he, he's kind of got ways of attaching new machines
32:30
as he builds around his workshop, right?
32:32
when people first started getting electric motors, the idea was, "Well,
32:36
we're just gonna get a, a, an electric motor and put it on that big drive
32:40
shaft at, at the top of the factory, and, and that's gonna be better." Well,
32:44
it, it wasn't really better, right?
32:45
Sometime-- I mean, occasionally, there, there, there were, there were places
32:48
where, um, they couldn't get a boiler room or they couldn't get water power for
32:52
whatever reason, um, and, and the electric motor was better, uh, better than nothing.
32:57
But the real gains came when instead of having one big source of mechanical
33:02
power that you then vector to, to each of your workstations, you
33:06
just put an electric engine at each workstation and then ultimately on,
33:10
on each machine, each power tool.
33:12
And that was the real, the real gains.
33:15
And so we're gonna have to, we're gonna have to do the same thing, right?
33:18
Our corporations, the way our corporations are working right now, they have a
33:22
management system that is effectively that big mechanical drive shaft.
33:27
And so we need to redesign the corporation.
33:29
We need to redesign human organization around distributed judgment and command,
33:33
not around departmental handoffs built for a pre-AI factory, right?
33:37
So, so there's a bunch of people that are trying to figure out how to do this.
33:41
I'm not-- I don't, I don't know that anyone quite has, you know, seems
33:44
to have come out with a, "Hey, this is the secret sauce. This is the way
33:47
to do it." you know, and this, this is a new information technology.
33:51
The, the corporation was originally built around paper.
33:54
It was, it was the printing press.
33:55
It was the ability to create identical sheets of paper and spread them
33:59
around what you were doing that allowed us to do the corporation.
34:02
And now AI is, is dramatically changing that.
34:06
You know, one of the things that, that AI makes possible is that, um,
34:09
you can have the same information.
34:12
It looks like, uh, my friend Robert has jumped off, but this is what, this
34:16
is what Runcible is, is working on.
34:18
Um, you can have the same information presented in multiple different ways.
34:22
Um, and, and that's, that's a problem that AI can solve, right?
34:25
It can bring a bunch of different information or information systems
34:28
that categorize it a little bit differently and bring it into one
34:31
information-- one kind of user interface, and you can standardize that across
34:35
all, all the things you're doing.
34:37
Another thing that you could do is you could have every single person have
34:42
their own idiosyncratic user interface.
34:45
Now within limits, right?
34:46
The, the, the corporation is gonna have requirements, but you could
34:49
have the AI optimize the user interface for each individual person.
34:54
And I think certainly for people that are using AI, when they've got the power,
34:58
they've got the, the, the, the tool that can make them new stuff, right?
35:02
They're for sure gonna do that, right?
35:04
It's the same kind of thing, you know, I, I really wonder when, uh, when
35:08
Figure gets its, um, gets its robots into production, like, what are you not
35:13
gonna be able to do with that robot?
35:14
Like, you buy a robot that can, that can make stuff, are, are you gonna need
35:18
construction workers anymore, right?
35:20
When you, when you do a remodel of your house or a remodel of, of, of your factory
35:25
or your office, if you've got one of those robots, you'd be crazy not to have the
35:29
robot just work on it at nighttime, right?
35:31
Like, this is gonna create all… Like, just having that thing, having
35:35
that capacity in there is going to dramatically reshape things, so All right.
35:41
So now we're gonna talk for, for the next extended little bit, we're
35:44
gonna talk about one of my favorite subjects, which is farming, right?
35:47
And of course, we'll start with my, uh, you know, the, the, the obligatory, those
35:51
of you who have heard me rant on this subject will just have to put up with it.
35:54
Right now, we farm in a way that is, I consider really bad, right?
35:58
Standard industrial agriculture runs on a petrochemical stack.
36:01
We have synthetic fertilizers, we have herbicides, we have
36:03
insecticides, we have fungicides, and there's just a bunch of hazards.
36:07
There's, you know, acute toxicity.
36:09
presumably, most of you will have seen on, on, on Twitter at one point, you know,
36:12
the, the pictures of people out, you know, in the fields with hazmat suits on, right?
36:18
I know parenthetically that, that people are considering whether or not to,
36:21
uh, to allow amnesty for farm workers right now, and we really should ask
36:25
what, what is going on in those, uh, in those working conditions and, and, you
36:30
know, do we, do we have modern slaves?
36:32
And is that, is that really how we wanna do it?
36:34
Um, and then there's also, you know, long-term effects on soil
36:36
microbiome, pollinators, residual chemistry in food and water that,
36:40
you know, we're still figuring out.
36:41
Um, so I, I have a friend of mine who has-- owns land that was formerly a,
36:45
a peanut farm, and you use a lot of Roundup in, in, in peanut farming.
36:49
And so for peanuts, you know, what you want is, you know, three to six
36:53
inches of sand and then clay, right?
36:56
Or at least you can do peanuts on that land, and so people do.
36:59
but that setup, when you're spraying Roundup on the fields,
37:03
uh, the Roundup goes down into the clay, and it just sits there.
37:07
A Roundup, you know, glyphosate, you know, breaks down in, in an oxygen-based
37:11
environment, in an aerobic environment.
37:13
But if it gets into the clay, the clay is anaerobic, uh, it doesn't break
37:16
down, and so it just concentrates year after year after year.
37:19
And so you get, you know, you get clay samples that have three or four times the
37:25
legal d- limit for what you're supposed to have in soil that, that, that, you know,
37:29
is, is around humans and, and animals.
37:32
And, you know, nobody, nobody thought of that.
37:34
Nobody knew that.
37:35
Nobody had done that analysis ahead of time and then, "Oh, well, maybe on this
37:39
particular land, we shouldn't use this chemical that's in everything," you know.
37:42
And, and, I mean, glyphosate is, is a, is a crazy common chemical, right?
37:48
So-- But there's some really interesting new technologies that
37:51
are coming out that I think are gonna, are, are gonna change that.
37:54
So the one, the one that, that's, that I think is, is most applicable to the stuff
37:57
we're talking about today is, is light.
37:59
So, um, using lasers, uh, to… They can do a bunch of different things.
38:04
One of the things they can do is they can suppress surface pathogens.
38:06
So you have, uh, you have, uh, blight, or you have some fungus, or you have
38:11
something that's kinda growing on the plant that you're trying to harvest.
38:15
You can use lasers, um, you can use UV to, to deal with those, those fungus, those
38:22
fungus without using a chemical fungicide, without using a chemical herbicide, right?
38:26
If you have laser weed eating, you burn the weed's growing point
38:30
with no herbicide residue, right?
38:33
Um, and then you have spectrum recipes, so, so different wavelengths.
38:35
You know, everyone's heard of, like, red light therapy for humans now.
38:38
Well, for plants, there is red light and blue, far-red, green,
38:42
controlled UV, and all of these kind of activate different things.
38:45
They shape growth and defense chemistry.
38:47
Light is, is food and signal.
38:49
Now, of course, light cannot It can make nutrient use more efficient.
38:53
It cannot-- Like, if there's no nitrogen or no phosphorus or no potassium in
38:56
the, in the plant, it can't create those things ex nihilo, right?
39:00
Those, those things are, are, are, you know, you gotta
39:02
have those atoms in the soil.
39:04
What light can amplify is a living polyculture.
39:06
Many species on one acre, not a single crop drowning in synthetic inputs.
39:10
Um, there's a guy who I greatly admire named Joel Salatin in
39:14
Virginia, and he has talked about this idea of polyculture, right?
39:18
So having many plants, um, and on, and on Salatin's farm, animals
39:22
sharing the same ground, so the acre does several jobs at once.
39:25
Um, and this is really interesting.
39:27
This allows you in many ways to get away from, uh, from chemical monoculture.
39:31
Um, one of the, the, the famous examples that Salatin uses that
39:35
I'm, I'm, I'm taking, right?
39:36
But he came up with this.
39:38
Um, at the… The old prairie was, was very interesting.
39:43
The old prairie was, you know, just grass.
39:45
It was, it was, it was, you know, very big, right?
39:48
The prairie grass would grow higher than a man's head.
39:51
and it had a bunch of different species in it.
39:54
And basically, they thought, well, prairie grass is just this, is this
39:57
magical grass of some, some genetic variant that was extinguished 'cause
40:00
we cut down the prairie grass and it didn't grow, uh, it didn't grow back.
40:03
And what they found was they, they… By the way, they still have, I think it's
40:06
the University of Wisconsin still has like a, like a quarter acre or a half acre of,
40:10
uh, of prairie grass that they preserved.
40:13
And it's got like fifty different species in it.
40:17
Um, and we, we have been able, you know, to, to replicate when you introduce more
40:22
types of grass, especially many more.
40:24
So, so, uh, Cedar Creek, uh, did this thing where they had sixteen
40:28
species of grass, and the grass grew two and a half times, two point
40:32
seven times larger and more dense.
40:35
Um, and so, so if you, if you have grass that is just one species,
40:39
just one type of grass, it, it can't, it can't grow very big.
40:43
Uh, mixed roots and canopies build biomass and soil architecture in a way
40:47
that a single variety cannot match.
40:48
Yeah.
40:49
Another thing that different plants do is they put different
40:51
nutrients back into the soil.
40:53
Uh, legumes fix nitrogen, deep roots mine minerals, covers and
40:56
companions feed the microbiome.
40:57
And, and you've heard, you know, uh, George Washington Carver, crop
41:00
rotation, you know, that's where, that's when we started raising peanuts.
41:03
Um- But the, the, the thing that is interesting about polyculture is
41:09
what if you could have all of those plants in the field at the same time?
41:13
Especially if you pair that with, you know, light prescription.
41:16
So you have spectrum, you have dose and timing, and you can get this
41:19
vigor without having to, to spray the fields with petrochemical mixes that
41:24
you need a hazmat suit for, right?
41:26
Um, now the reason why we don't do this, the reason why we haven't done
41:29
this is that a combine cannot pick three crops, uh, at one time, right?
41:34
The combine, you gotta tune it for whatever crop it's picking.
41:37
There's a certain pressure that it can use, and you have to have one thing
41:40
that, you know, uh, that if you're using a spray boom or whatever, it, it,
41:43
it needs every plant to be the same.
41:46
And, and in fact, you know, that's kinda how, how it works.
41:48
If you have, you know, potatoes or whatever that are, that are too soft or
41:51
sometimes too hard, those are rejected.
41:53
There's something wrong with them.
41:54
We wanna sort those out.
41:56
But more and more, we're having, you know, articulated robotic hands and,
42:00
and they, and they change that, right?
42:01
So you can, you can have something that's integrated.
42:04
It can see, classify, and handle individual plants.
42:07
And, and with this, with AI and robots, polyculture becomes something
42:11
that can reach industrial scale.
42:13
the, the, the example that I'm giving give is, uh, Kyber Labs.
42:16
So, uh, at the end of last year, uh, Kyber Labs posted a back-drivable
42:21
robotic hand spinning a nut onto a bolt at extreme speed.
42:24
So they had the, the nut and the bolt and, and the hand was very, very
42:28
quickly, like it, it was blurred, um, turning the, the, the nut on.
42:32
And what they did was somebody would stick their finger in into this very fast-moving
42:37
machine, and the AI would stop it, right?
42:40
Immediately.
42:40
It wouldn't hurt your hand.
42:42
I don't really understand why this is true, but for whatever reason, he got this
42:46
bad interview with some, some of these mainstream, uh, mainstream interviewers.
42:50
It was a couple of women, and they decided that, that his product was, like, about
42:55
sex toys or something of this nature.
42:57
And so there's some really weird marketing around him or re-weird
43:01
kind of search results around him.
43:02
Um, the founder kinda came out on X. He was like, "This is not what we're
43:06
about. We don't know why they talked about this. This is not what we talked
43:09
about." This is very frustrating to him.
43:11
Uh, it's a great innovation.
43:13
Um, and, and the real payoff that I see is… So, but, you know,
43:16
obviously, if you search for it, just kinda, uh, be, be advised.
43:19
Um, some, some, so he-- They got some interviewers that were, that were weird,
43:23
bad modern women, though I repeat myself.
43:26
so the real payoff that I see is as a general purpose harvester, you know,
43:30
for soft fruits and for different fruits at industrial speeds, okay?
43:35
So, um, and then, then there's a, a related concept called permaculture,
43:39
which takes polyculture even further.
43:40
So you stack layers, canopy, understory, shrubs, herbs, and ground cover, um,
43:45
and it's designed as a food forest that feeds itself over the years.
43:48
And, and, and again, the opportunity that I want you to see is with
43:51
robotic harvesting that can see and, you know, with, with AI that can
43:54
see and pick plant by plant, you are no longer forced into monocultural
43:58
rows for the sake of the machine.
43:59
You can still get incredibly efficient yields, dense layered production
44:03
without using hazmat chemistry.
44:05
Okay?
44:06
And, you know, we might call this a, a plant ledger, a living record of every
44:10
plant that stood on that square foot for the last three years and, and a plan,
44:14
a projection of every plant that will stand there for the next three, right?
44:17
If you pair that light prescription with that ledger, you have rotation,
44:21
companion planting, and the, and the, the light spectrum, the wavelengths,
44:24
that becomes a plan score, not, not a guess after the spray truck leaves.
44:27
Because right now, again, we do things at the level of, you know, a, a sometimes a
44:32
fifty-acre field or a thirty-acre field.
44:34
There are still people that do, you know, ten-acre fields, um, but,
44:37
but less and less of them because, you know, monocultures is becoming
44:40
less and less efficient, and the margins are, are worse and worse.
44:43
There's, there's way more co-quality control issues that, that, that
44:46
you-- than you used to have.
44:48
Um, for management and for compliance, you know, whether it's buyer regulatory
44:52
or insurer, I think this is an incredible use case for some kind
44:55
of blockchain or crypto technology.
44:57
You know, it's an auditable history of what grew where,
45:00
what was treated, and when.
45:01
And, and then again, we, we, we are reducing this to the same
45:05
three basic skill sets, right?
45:07
So you have AI-- little drones that are going out and scouting.
45:11
They've got cameras on there that are feeding back into an AI system
45:14
that's classifying what it's seeing, and then there's doing analysis and
45:17
prescribing what do we need to do.
45:19
Then if, if what it needs to do is, is, is, you know, strike with a laser at
45:24
a, at a bug or at a, at a weed or at something that needs to be cauterized,
45:29
um, you know, this is, this is the same loop as that defensive swarm,
45:33
that, that defensive drone swarm that we were talking about last week.
45:36
Um, it's defending a field faster than the farmers of today can imagine.
45:39
I mean, you know, most farmers basically if, if you get blight, if you get
45:42
some kind of, of, you know, fungal or some kind of infection in your
45:46
field, the whole field is done, right?
45:48
There's, there's no coming back from that.
45:50
Well, if you could have drones that are flying around looking, and every
45:55
time they, they see something going bad or, or, or, you know, fungus starting
45:59
to grow, they cauterize the area.
46:01
Or every time they see a weed, they, they zap it, right?
46:03
Every time they see a bug or, you know, um, you know, some, some kind
46:07
of, of insect, some kind of pest coming in, they, they can kill the bugs as
46:11
they're flying into the field, right?
46:14
Well, then you don't need to just leave these chemicals that have
46:17
to last for days or weeks, right?
46:19
Um, 'cause I mean, that-that's one of the real problems with these chemicals.
46:22
They're kind of designed to be forever chemicals.
46:24
If they were easy to break down, then you'd have to spray them all the time.
46:27
So you wanna-- what you want is to be able to spray and then, you know, then
46:31
have it last so that anything that flies in for the next six weeks or, or
46:35
more dies that you don't like, okay?
46:38
And, and again, I think, I think this is gonna dramatically,
46:41
uh, reshape agriculture.
46:43
So, uh, one of the, one of the, the, the companies in this space is called Carbon.
46:48
Just so you know that this is not just something that I came with--
46:50
came up with out of my own mind.
46:51
Um, I-- for the record, I was talking about this before any of
46:55
these companies existed or had announced any of these things.
46:58
I saw it coming, gentlemen, uh, ladies and gentlemen.
47:01
So in February of twenty twenty-five, Carbon Robotics announced, uh, scale
47:04
numbers for its laser weeder line.
47:06
It's a, it's a tractor-pulled module with cameras, classifiers, and CO2
47:10
layers-- lasers trained on more than a hundred and fifty million plants.
47:13
So Carbon says this tech has been used on two hundred and fifty thousand
47:16
acres in fifteen countries on a hundred plus crop types with more than fifteen
47:20
billion weeds eliminated, which is great marketing copy, and, and again, the system
47:24
is, is doing what I was talking about.
47:25
It can, it can look at, it can take a picture, it can identify what is
47:28
the crop versus what is the weed.
47:30
It can feed, uh, fire at the weed, and, you know, and, and that might
47:33
be, you know, mere centimeters, inches or centimeters from the plants that
47:38
you're needing to protect, and it only hits what you want it to hit.
47:41
And then it, it logs the shot forever.
47:43
Like, that-- whatever-- every single time you zap something, the system
47:47
knows that it, it zapped, and it knows, you know, basically on what
47:50
square foot it did it, So then Cornell University, uh, which is the land-grant
47:54
university in New York, it's got, uh, you know, Cornell Agri-Agritech.
47:58
Um, it runs comparative field trials on farm technology.
48:01
So in June of twenty twenty-five, the Cornell Chronicle reported their,
48:04
their trial of commercial laser weeders against common East Coast weeds.
48:09
and so what they found was that lasers can match or beat common herbicides
48:13
on several upright broadleaf weeds.
48:15
It's weaker on some grasses and sprawling plants.
48:18
But again, the thing is, is that y-you can record every single kill, right?
48:23
You can record every single laser strike, and, and you can preserve that so you
48:27
know-- you, you, you don't just know that all the weeds were taken care of.
48:31
You know how many weeds were taken care of.
48:32
You can see the whole thing.
48:34
And, and, and I'll note parenthetically, um, this is the
48:37
defender's wartime loop, right?
48:39
You scout, you, you classify, and then you strike.
48:42
And, uh, this-- what this is gonna mean is that the ROTC and the Grange
48:45
kids are o-once again gonna be the same kids using effectively the same tech.
48:50
Trick Robotics over, you know, in California, there's a guy named
48:53
Adam Stager, uh, who is-- again, these, these are tractor units.
48:57
Um, they're using UV, UV light and bug vacuums on some of California's
49:01
largest strawberry fields.
49:03
So they've, they've treated tens of thousands of acres, and
49:05
they're, you know, they're, they're doubling their fleet every year.
49:08
So they-- basically, they go out at night and with no chemicals, they have
49:12
UV, they kill the pathogens and pests, and then they suck it out of the field.
49:15
Okay?
49:16
I think that the next step, th-this is the thing that I'm gonna predict,
49:19
is that, that this is gonna leave the, um, we-- this is gonna leave
49:24
the, the, the tractor thing.
49:26
There might be still some things that, that run on tractors.
49:27
Obviously, for whatever reason, if you need to vacuum stuff out, right?
49:31
That'll probably be more efficient on a tractor.
49:33
But I think for lasers, that can be put on instead of having, uh, you know,
49:38
a laser array in a box that you're dragging across a field, have one
49:42
laser per drone and have, have a cloud of drones moving across the field.
49:46
And one of the things that will allow is if one of those breaks, then you
49:50
fix the one, the one drone, right?
49:53
Rather than having many of these models, um, when, when one laser
49:57
breaks on the array, you have to take the whole array out and get it fixed.
50:00
It's a very expensive process.
50:02
uh, uh, you know, one of, one of my ideas here for this would be, uh,
50:05
something called Farm OS, right?
50:08
So You have, you have a scout which, which scout drones which map stress,
50:13
it, it map insects, weeds, disease, and then you have the Farm OS,
50:16
which writes a light prescription.
50:18
You know, if you're using the, the, the, the wavelengths against the
50:20
plant ledger, it, it figures out what spectrum, what dose, what the
50:23
time, what the tool, and what square foot we're doing this on, right?
50:27
What's been planted here for the last three years?
50:29
What are we planning to plant here?
50:31
Um, and, and then sometimes 'cause sometimes the, the, the remedy might
50:35
be, oh, well, we need to plant.
50:37
We need to add to this field.
50:39
We need to plant new things, um, which is really interesting.
50:43
It's, it's really interesting that, that, that can be a solution because different
50:46
plants kind of create immune resistance against monocultural blights, right?
50:52
And then, and then you send out the drones, which, which just do it.
50:55
And, and, and again, I'm, I'm kind of describing this without describe--
50:58
without talking about the human role.
51:00
This is all going to work much better when you have a human in the loop.
51:03
I assume at some point we'll get, you know, kinda automated hydroponic
51:07
farms where everything is in an enclosed, you know, shipping container
51:09
style thing and, and the-- and, and AI can probably get to the point
51:12
where it can run that pretty well.
51:14
Um, but for actually dealing with the real world, I don't think you're
51:18
gonna be able to replace, certainly not anytime soon, and again,
51:22
certainly not cost-effectively, right?
51:24
You might be able to have, you know, Peter Thiel might be able to spend, you
51:28
know, a billion dollars or something on an AI that watches his field
51:32
and makes all the decisions, right?
51:35
But it's not gonna be cost-effective for a really long time to have a-- just to
51:40
have all of the decisions made by the AI.
51:44
There's gonna be a need for humans in the loop to, to make decisions and say, "Okay,
51:48
do this, do that." the thing that I… Hopefully, it's coming through that this
51:53
is still the same three skill sets, right?
51:56
This is code, this is content, this is command.
51:59
So you're setting up the logical requirements.
52:01
You're, you're figuring out how to… You know, one of the, one of the things about,
52:06
about content for this is when you're dealing with, with user interface, you're
52:09
dealing with making things comfortable.
52:11
You're also trying to make it comfortable for the animals, right?
52:13
You're also trying to figure out how not to stress the animals.
52:16
'Cause if you stress the animals out, then you're gonna get, you know,
52:18
bad, bad outcomes from that, right?
52:21
And, you know, right now our bottleneck is toxic chemical
52:24
inputs and the cheap disposable labor needed, uh, to apply them.
52:28
And, and I think as these technologies come in, we're gonna
52:32
need more and more holistic humans.
52:33
We're gonna need more and more gentleman farmers, more guys that are, are able
52:37
to think through how do we… And, and I think that's gonna be true even for, uh,
52:41
the guys that stay on the chemical kinda track Because they're gonna need, you
52:46
know, the-- i-in ten years, you're not gonna have-- you're not gonna use cheap
52:51
foreign labor that's gonna get cancer, right, from, from your chemicals, and
52:55
you're just gonna hope that they go back to their home country and don't sue you.
52:58
Um, or the-- I, I think we're gonna move to robots in all of
53:01
these, all of these industries.
53:03
And again, once you've got robots, once you've got an AI system that's,
53:07
that's looking at the field, well, then there's no reason, there's no need to
53:12
have it be a monoculture at that point.
53:13
You've already made the leap.
53:15
You've already made the investment.
53:16
It's gonna be way easier to convert to a polyculture or maybe
53:20
even a permaculture, uh, framing.
53:22
And so, you know, that, that, that's my white-pilled take on this, and,
53:26
you know, I mentioned Peter Thiel.
53:27
He, he invested heavily in, uh, Halter, which is a New Zealand company that puts
53:32
solar-powered GPS collars on cattle.
53:34
Um, and, and again, th-this is-- they're able to use virtual fences.
53:38
They're able to track them.
53:40
Um, so the, the, the, the founder said, you know, the idea is here, you're
53:43
gonna be able to manage your farm from a phone without dogs, horses,
53:46
motorbikes, or helicopters as the default.
53:48
And, you know, I mean, I th-- I think one of the things that I would, I
53:52
would point out there is, is that you're still gonna need drones.
53:56
You're gonna need drones to… If you, if you've got an animal that's sick
53:59
or that's got something wrong with it, you're gonna need a drone to go out there
54:01
and, you know, deliver the medicine.
54:03
You're gonna need the drone to go out there and give, you
54:05
know, kind of the supplements.
54:06
You're also gonna need in, in most all of Latin America, for sure, and
54:09
m-- and much of the, the American Midwest, uh, you're gonna need, you
54:13
know, attack drones to defend, uh, your herds from, from dogs, from
54:17
predators, from, you know, cougar and, and, um, and wolves in, in many places.
54:22
And even, even in some, you know, for especially for smaller livestock,
54:25
from coyotes and things like that.
54:27
And, and again, that is a skill set that has profound, uh, profound
54:33
military applications The final, the final major, um, vignette that
54:38
I'll share with you tonight or today is, is the, uh, Morpho in Rio.
54:43
Um, in January, this is about two years ago.
54:46
They, they, they demonstrated seeding drones at Mirante do, do Pedral in,
54:51
um, Botafogo with a pilot aimed at the Posse Forest in Campo Grande.
54:55
So the, the CEO of Morpho is, uh, Gregory Maitre, and he partnered with
55:01
the city under Re-Refloresta Rio.
55:04
So the AI planting-- the AI would come up with planting plans.
55:07
It would look at the topography and say, "Here is where we could plant trees,"
55:11
where there's some level of, like a little natural protection, something
55:15
where the water's gonna gather so the, the sa- the sapling can come out.
55:18
and they're, they're able to do that.
55:20
So they were able to seed at least twenty native species across succession
55:24
statio- stages, and they called it a smart plantation, which decides which
55:28
seeds go where in what mix and quantity.
55:30
And the drone was able to fire about a hundred and eighty capsules a minute.
55:34
Um, so a two-person team and a drone claimed roughly a hundred times faster
55:38
than traditional planting, right?
55:40
So Maitre said, "The work begins much earlier with the reconnaissance of the
55:43
area, with the identification of the best species, and ends much later with
55:47
the long-term monitoring. The drone is the central part of all this work."
55:51
And of course, they're using artificial intelligence to monitor
55:53
and track seedlings and s- and seeds and not just to seed the drop
55:59
Mm-hmm.
56:00
So again, you know, these are civilian jobs.
56:03
They have massive implications, right?
56:05
Builder, solo marketer, laser weeder, UV strawberry fleet,
56:08
cattle caller, seed drone forest.
56:11
Um, hopefully, you see how these types of force multipliers are im-- Like, it's
56:15
impossible to do this in the civilian context without acquiring skills that
56:19
are relevant in, you know, what, uh, what Roman Helmut guy famously called the
56:23
natural meritocracy of extreme violence.
56:25
Um, and, and that's gonna mean that the most valuable profession in peace
56:29
and the most dangerous operator in war are going to look, uh, similar.
56:33
They're gonna understand the world more closely than those two
56:37
groups have in four hundred years.
56:39
Um, and I think that's going to, um, that's going to have massive…
56:42
And, and, and I can't stress this, like, there, there, there is the
56:45
possibility for massively positive changes to our world from this.
56:50
so, um, as I said in the beginning, humans are not obsolete.
56:53
AI is going to reshape every industry: software, marketing,
56:56
medicine, manufacturing, uh, the field and the forest, right?
56:59
The various types of ag.
57:00
And the-- it's gonna do that through these core skill sets: code,
57:04
content, command, which basically all have military, uh, applications.
57:08
And again, the, the social and economic implications of that are
57:11
huge, maybe greater than what I'm putting in front of you, um, today.
57:15
So next time, uh, we'll, we'll talk about, uh, where the safe professions
57:19
are, which I actually think is the wrong question, but we'll get into
57:23
it, and then we'll talk about next time the, the economics of a Prime.
57:26
What are the economics of somebody that has all three skill sets
57:30
and, and what, what would you do?
57:32
'Cause I think it's a very different, uh, track than, you know, the, the,
57:36
the kind of the technical university, the research university that we have
57:39
based our, our education systems around for the last four hundred years.
57:42
So that concludes my prepared remarks.
57:45
Um, so thank you so much for everybody that, uh, that's joined us for the free
57:48
portion and, uh, let's move into the Q&A.