WEBVTT
00:00:08.329 --> 00:00:15.048
Hello everyone And welcome to the switch statement podcast It's a podcast for investigations into miscellaneous tech topics
00:00:15.981 --> 00:00:19.411
This is episode nine of The Coming Wave by Mustafa Suleiman.
00:00:20.498 --> 00:00:21.667
We'd love to hear from you.
00:00:21.878 --> 00:00:25.748
Send your comments and feedback to team@tkp.co.
00:00:30.559 --> 00:00:31.320
Hey, John.
00:00:31.339 --> 00:00:31.760
How you doing?
00:00:32.848 --> 00:00:33.579
Hello, Matt.
00:00:33.598 --> 00:00:34.939
I'm doing very well.
00:00:34.939 --> 00:00:35.579
How are you doing?
00:00:36.795 --> 00:00:37.744
doing okay.
00:00:37.765 --> 00:00:38.484
Doing okay.
00:00:38.484 --> 00:00:45.273
Just, I've been really acutely experiencing the limitations of AI, recently
00:00:46.268 --> 00:00:47.398
Oh my God, man.
00:00:47.487 --> 00:00:59.427
Well, I'm glad to hear you say that because I'm surrounded by people with AI psychosis that just believe AI can do anything, and they'll show me some vibe coded POS and be like, Look at this app.
00:00:59.704 --> 00:01:06.664
Like, it's amazing." And I just instantly see, like, 900 flaws in it, and I'm just like,"Oh my God."
00:01:06.784 --> 00:01:07.144
flaws
00:01:07.737 --> 00:01:19.400
It's so funny'cause I, I have totally drank the AI Kool-Aid, but at the same time, I feel like I still am super skeptical of, like, so many AI things
00:01:20.239 --> 00:01:23.239
I have, I, I have what I'm, I'm gonna coin a new term.
00:01:23.248 --> 00:01:40.558
I have AI ambicosis, which is I am simultaneously completely bought into the hype and also, like, constantly fucking, like, pissed off about how shitty AI is.
00:01:41.516 --> 00:01:49.027
Dude, it's like, I almost feel like people don't realize that it can't do everything.
00:01:49.037 --> 00:02:04.756
Like, it's so hard to explain'cause AI does have this remarkable thing where it can produce such a close approximation of, like, a real thing that it can, like, convince people that, like, it's actually doing that real thing.
00:02:04.900 --> 00:02:06.219
And I wish I had an example.
00:02:06.239 --> 00:02:09.786
Like, this would be so much easier to explain if I had examples.
00:02:09.786 --> 00:02:16.216
I don't wanna, like, go into super detail about my job or anything, but, uh, uh, like a web app, you know?
00:02:16.216 --> 00:02:25.288
It's like you can vibe code a web app in two seconds, but it's like y- you know, you gotta deploy it, you gotta serve it, you gotta host it, you have to have all these other things.
00:02:25.298 --> 00:02:27.747
It's gotta have security, it's gotta have auth, you know?
00:02:28.277 --> 00:02:29.217
It-- So it's...
00:02:29.647 --> 00:02:41.062
Which, you know, by the way, AI can do all those things, but it's just like you can't instantly be convinced by, like, the two-second vibe code that you have solved any real problems.
00:02:41.534 --> 00:02:41.974
Yeah.
00:02:42.245 --> 00:02:42.664
Yeah.
00:02:42.675 --> 00:02:51.215
No, and I think-- I the greenfield performance is tricking everyone into believing that these are general purpose intelligences,
00:02:51.683 --> 00:02:52.614
Omni use.
00:02:53.807 --> 00:03:07.967
it's funny because I feel like I exist in, a weird space where it feels like either, you are all pro and you're like,"These things can do anything," or you're all anti and you're like,"These things can do n-" And it's like, no, no, I want it to do this.
00:03:07.997 --> 00:03:09.627
Like, I want it to be able to do this
00:03:09.788 --> 00:03:10.248
Yeah
00:03:10.638 --> 00:03:17.195
and I'm trying to get it to do this thing, but I just acknowledge that it absolutely cannot.
00:03:17.195 --> 00:03:19.152
And the question is, like how do you deal with that?
00:03:19.152 --> 00:03:22.973
And, and maybe, maybe it's time to actually talk about the content of this chapter.
00:03:23.737 --> 00:03:28.337
Well, we, we sort of have been, I feel you know,'cause we are talking about AI
00:03:29.685 --> 00:03:34.235
and this, and this is the thing, you know, I think maybe, okay, if the, we have the cleanest segue here.
00:03:34.534 --> 00:03:55.294
You know, he talks about these, this asymmetry of power, and the whole point of this project that I'm working on is, like, a single person should be able to be, like, the CEO of their swarm of coding agents, and they just, articulate a vision, and then the agents just, they just do it all, and it's beautiful and perfect, and they know exactly what you mean.
00:03:55.900 --> 00:03:58.300
Yeah, and there's no intervention necessary
00:03:58.544 --> 00:03:59.824
no intervention necessary.
00:04:00.936 --> 00:04:01.885
asymmetry.
00:04:01.925 --> 00:04:17.776
So, and you know, I do think there's truly an asymmetry between programmers today who are learning to leverage AI agents within their day-to-day workflows versus programmers.
00:04:17.795 --> 00:04:25.105
You know, there's some Luddite programmers who are sort of refusing to adopt AI tooling.
00:04:26.065 --> 00:04:28.355
And, you know, there's always gonna be Luddite programmers.
00:04:28.355 --> 00:04:31.255
Like, I, I actually remember this when web frameworks were coming out.
00:04:31.255 --> 00:04:33.555
Like, I was kind of a Luddite programmer.
00:04:33.555 --> 00:04:35.925
I was like,"Why would you use a web framework?
00:04:35.925 --> 00:04:40.505
Like, I'm just gonna keep using MooTools," which was, like, a JavaScript library.
00:04:41.045 --> 00:04:48.286
D- I just, like, didn't really understand the, the necessity of it, uh, or the, or even the utility of it.
00:04:49.105 --> 00:04:53.846
And, uh, you know, I, I-- my websites were worse.
00:04:55.485 --> 00:05:05.649
So it's kind of this element of, like, y- you know, there is this a huge advantage in adoption of some new technology, but you know, it's a bleeding edge.
00:05:05.838 --> 00:05:08.608
It also has issues, which you were discussing.
00:05:14.550 --> 00:05:17.350
he actually starts this chapter in a point that...
00:05:17.788 --> 00:05:32.622
And, and I guess he's commenting on actual, you know, actual events that had happened in the war in Ukraine, but it was pretty prescient, like that, that turned out to be very indicative of the direction that everything was gonna be moving.
00:05:32.672 --> 00:05:47.422
He was talking about drones, and I didn't, I don't know, I hadn't heard this, that, you know, it, it was very widely reported that Russia had this really long, line of like tanks and trucks and what have you, this 40-kilometer line of, of armored vehicles,
00:05:47.805 --> 00:05:48.204
Right
00:05:48.451 --> 00:05:50.132
and, and then they were all just stuck.
00:05:50.322 --> 00:06:17.596
What I hadn't heard was that there was this ragtag, uh, group of I don't know what the right word is, but like group of soldiers, militia, who in cover of darkness, like went and were able to use drones to disable a couple of vehicles at the f- head of the line, and then it's like, all right, doesn't matter how long the line of armored vehicles is, they're not going anywhere because all of the, the ground around was unpassable.
00:06:17.617 --> 00:06:17.997
So,
00:06:18.384 --> 00:06:30.995
Yeah, this is an, uh, an element where, and, uh, this is sort of the classic thing where it's like if you're invading a country that's sort of fighting for its own existence, it's gonna use all the advantages it has at its disposal.
00:06:31.005 --> 00:06:36.355
Whereas if you're the Russian military, it's like these guys are barely...
00:06:36.514 --> 00:06:38.324
You know, they don't even wanna be there.
00:06:38.752 --> 00:06:39.153
Right
00:06:39.264 --> 00:06:41.644
know, they're, like, doing the bare minimum.
00:06:41.673 --> 00:06:51.043
Obviously they're, they wanna survive, uh, but, you know, they're just, following orders, doing the conventional whatever they've historically been doing.
00:06:51.802 --> 00:07:02.432
And so yeah, I mean, this is this perfect example of asymmetric warfare where Ukraine fighting for its own existence is very willing to adopt whatever new technology it can.
00:07:03.012 --> 00:07:08.942
Also, Ukrainians are geniuses, so they're very aware of, like, new technologies and drones and things like that.
00:07:09.303 --> 00:07:15.689
and so yeah, I, I do think it's a good example'cause it's this very real world thing where we all kinda witness this live.
00:07:15.769 --> 00:07:20.559
It's like the Russian military supposedly was this mighty behemoth, and
00:07:20.562 --> 00:07:32.853
The second most powerful military, like I, before this war, I think most people would be like,"Yep, r- you know, Russia's got the second most powerful military." You know, because there's always this like kind of ambiguity, like Russia, China.
00:07:33.523 --> 00:07:38.202
Um, but it's like are they even the second most pow- capable military?
00:07:38.233 --> 00:07:41.423
I mean, y- sorry, are they even the third most capable military?
00:07:41.432 --> 00:07:46.783
Like at this point it's like they, feels like they fell like several, several ranks
00:07:47.182 --> 00:07:48.882
No, it was very embarrassing.
00:07:48.992 --> 00:08:02.252
And I think, uh, you know, Putin, who prides himself on image and just the optics of everything, like, man, he must have not enjoyed that situation at all.
00:08:02.992 --> 00:08:04.442
And I feel like it continues
00:08:05.144 --> 00:08:11.735
Trump is doing a pretty good job of, uh, of fighting Putin to who- to see who could fall down
00:08:11.879 --> 00:08:12.718
Ugh, God
00:08:12.735 --> 00:08:16.524
the military power, firepower list as fast as possible
00:08:16.773 --> 00:08:19.163
Yeah, Trump's handling of Iran has been just incredibly embarrassing, which is kinda sad.
00:08:19.163 --> 00:08:33.023
I mean, I, uh, it's weird because I feel like whenever the United States does something like this on the global stage or what have you, like, I am hopeful.
00:08:33.062 --> 00:08:46.490
You know, even, even if I dislike the current administration or whatever, like, I'm hopeful that we'll just do the right thing and, like, I felt the same way when Bush was president actually,'cause we were doing all this military stuff, and I was just...
00:08:46.880 --> 00:08:57.750
Yeah, I, I was hopeful that the United States would sort of comport itself well, and that obviously ended up being just utterly, utterly false during the Bush era.
00:08:58.461 --> 00:08:59.520
But with Trump, I had hope.
00:08:59.520 --> 00:08:59.801
Like
00:09:00.864 --> 00:09:06.264
by, like, by con- c- by contrast, I feel like the operation in Venezuela,
00:09:06.985 --> 00:09:07.375
Right.
00:09:07.475 --> 00:09:08.144
Exactly.
00:09:08.744 --> 00:09:12.274
that was one where I was like,"Okay, well, clearly someone did their homework.
00:09:12.754 --> 00:09:22.594
They figured out the, you know, the reality of the situation on the ground, the feasibility." It was this incredibly targeted...
00:09:22.793 --> 00:09:25.974
It was like s- it was like an hour or something, you know?
00:09:25.974 --> 00:09:26.514
It was crazy.
00:09:26.524 --> 00:09:28.094
It was like you blinked and you missed it.
00:09:28.144 --> 00:09:30.774
Uh, it was already over by the time anyone knew it was happening.
00:09:30.774 --> 00:09:39.884
So, um, whereas, like, when everything in Iran started happening, I was kind of like, listen, if this...
00:09:39.894 --> 00:09:41.274
If they could pull this off...
00:09:41.283 --> 00:09:49.953
A- and, like, the other thing is, like, it's, it's important to remember that the Iranian regime is, like, this incredibly oppressive dictatorship.
00:09:49.953 --> 00:09:57.864
So it's like if they were able to get, like, overthrow the government, like, for the people of Iran, like, that would be a good.
00:09:58.364 --> 00:10:07.443
But they clearly didn't do their homework, and, like, they were just too optimistic about their, their chances, and now we're in this fucking quagmire.
00:10:07.443 --> 00:10:11.734
It's just, like, so stupid and a waste of all of American taxpayer dollars
00:10:12.078 --> 00:10:17.999
It, it's also just such a-- I mean, like you're saying, it's like they just misplayed their hand.
00:10:18.009 --> 00:10:20.917
And I feel like we know about Iran.
00:10:21.047 --> 00:10:22.547
Like, we understand Iran.
00:10:22.557 --> 00:10:32.927
We-- There are people in intelligence who know that the IRGC is, like, incredibly entrenched in their administration, and it's incredibly difficult to topple that.
00:10:33.297 --> 00:10:45.716
Unlike Venezuela, which, you know, Maduro was also a piece of trash and sort of corrupt, and his administration was filled with people just taking whatever oil riches Venezuela has.
00:10:46.147 --> 00:10:53.258
Uh, the, the average Venezuelan should be doing well based on the opulence of the country's natural resources.
00:10:53.908 --> 00:10:56.658
And so, you know, Maduro was not that popular.
00:10:56.668 --> 00:11:06.498
And so in that situation, taking him out, I think, you know, the right move i-in the way that they did it, I think was the right move.
00:11:07.398 --> 00:11:17.018
Iran, on the other hand, you have to be much more sort of careful and surgical and, uh,'cause you can't just take out the head.
00:11:17.018 --> 00:11:26.038
I mean, the Ayatollah, sure, they, he has a lot of power, but it's like there's this entire power structure that they've worked for decades to put in place.
00:11:26.962 --> 00:11:27.423
Yeah
00:11:27.817 --> 00:11:28.107
yeah.
00:11:28.177 --> 00:11:35.817
Anyway, uh, maybe we shouldn't wade into, like, this type of territory, but just frustrating to me.
00:11:36.932 --> 00:11:37.982
No, I, I feel you.
00:11:37.982 --> 00:11:53.450
But, but, um, but anyway, so getting back to the core point, Ukraine is able to use$1,400 drones from DJI to disable a$10 million tank a-
00:11:53.773 --> 00:12:12.836
He makes this point about transfer of power, and this is sort of what new technology can do a lot of the time is, like, introduce this asymmetry where these traditional historical power centers, you know, and the things that they did to remain in power are no longer as effective.
00:12:12.936 --> 00:12:23.541
There's some new thing that's way more effective, and if you're this entrenched power structure, it's gonna be difficult for you to leverage that'cause the-- you have this inertia in some particular direction
00:12:24.206 --> 00:12:39.225
It's so weird because it feels like it's moving in both directions where the power of the individual is being expanded by this, but also like, it's also a massive centralization of power in the hands of all of the AI companies.
00:12:39.316 --> 00:12:39.855
You know what I mean?
00:12:39.855 --> 00:12:45.289
So it's like, it, it is kind of confusing, like I'm not sure, uh, you know, net-net how to feel.
00:12:45.309 --> 00:12:54.921
Is this net empowering to the individual, or is this net empowering to the entrenched companies, uh, you know, and, and governments?
00:12:55.085 --> 00:13:00.279
I hate to say it, but I think this is gonna be mostly net powering to entrenched companies.
00:13:00.279 --> 00:13:05.687
'Cause I think for a long time, the harness is gonna be what is truly...
00:13:06.118 --> 00:13:08.217
you know, has, has the true utility.
00:13:08.217 --> 00:13:10.248
It's like AIs are incredible.
00:13:10.368 --> 00:13:17.736
They can, you know, they can return responses that are, like, on the level of human intelligence.
00:13:18.181 --> 00:13:22.962
But it's just hard to accomplish anything with that until you have a harness.
00:13:23.052 --> 00:13:26.043
And a harness is, is still really, really hard to build.
00:13:26.241 --> 00:13:43.620
Uh, and so I think, you know, at least for a while, there- you're gonna have these power players that know how to build harnesses, have, you know, the bags of tools and, uh, whatever necessary in order to create truly useful harnesses.
00:13:43.966 --> 00:13:51.206
And, you know, they will they'll be the ones getting most of the, you know, reward, I guess, from the AI revolution.
00:13:51.385 --> 00:13:52.086
Which I hate to...
00:13:52.456 --> 00:13:52.865
I don't know.
00:13:52.895 --> 00:13:56.405
Like, I wish it was different,'cause I think it's actually different with models.
00:13:56.895 --> 00:14:07.518
I think you- today you have these frontier models and, you know, companies are spending so much money on these frontier models, building up infrastructure and training and running inference.
00:14:08.118 --> 00:14:17.398
I, I actually think in the future, frontier models won't be nearly as relevant because the average open source model will be, like, as good or better
00:14:18.860 --> 00:14:19.551
Exactly.
00:14:19.571 --> 00:14:26.490
No, that's what I was just about to say is that we have all these incredibly good open weight models.
00:14:26.500 --> 00:14:28.640
Not to mention that distillation is always gonna
00:14:28.850 --> 00:14:29.220
Right
00:14:29.921 --> 00:14:45.341
a problem where it's like, yeah, you spend all this money to create this frontier caliber model, and then it just takes a month or two for one of the, you know, another lab, an open lab to, to create a model with just as good performance.
00:14:45.760 --> 00:14:58.041
Uh, which, which actually scares me because in a way it's a disincentive from training those higher capability models because it's like, why am I gonna do this if another company's just gonna come along and steal our lunch?
00:14:58.524 --> 00:15:03.794
Luckily, these companies don't operate on logic, so they won't, they won't
00:15:03.870 --> 00:15:19.677
it is interesting because i- in a way, like it's, it's damned if they do, damned if they don't, because they, then they have a bunch of other domestic AI companies that are nipping at their heels, so if they don't continue to advance, then they're gonna get left in the dust, so
00:15:20.104 --> 00:15:20.663
Absolutely.
00:15:20.663 --> 00:15:25.573
There's a lot of zero-sum games around that and like CapEx and things like that, so
00:15:31.471 --> 00:15:52.716
so there was a quote in here that I really wanted to call out, and this is kind of what I was alluding to when I was talking about this agent swarm thing that I've been building, which is he says, There's no obvious reason why a single operator with enough wherewithal could not control a swarm of thousands of drones." Which like, that's terrifying just by itself.
00:15:53.087 --> 00:16:06.250
but I think it's true if you take, take a step back and you think about it in the context of like if you have someone who, who nails the like wiss하시는земcontrarved absor1치는rau amin conti jeopardy, amendsงาม�ρ 燕 世 了 解 到 sudden
00:16:19.927 --> 00:16:30.927
Yeah Totally.
00:16:31.097 --> 00:16:31.427
Yeah.
00:16:31.567 --> 00:16:34.357
First, first single individual billion-dollar company.
00:16:34.368 --> 00:16:36.418
I feel like a lot of people are talking about this.
00:16:36.427 --> 00:16:37.967
I think it's gonna happen.
00:16:38.418 --> 00:16:41.577
so yeah, there's asymmetry, which is one of the four things.
00:16:41.658 --> 00:16:45.268
Did we ever say that this chapter was called Four Features of the Coming Wave?
00:16:45.508 --> 00:16:46.238
I don't know if we said
00:16:46.462 --> 00:16:48.822
I don't think we did, but thank you, yes, for doing that
00:16:49.518 --> 00:16:51.788
So the first feature was asymmetry.
00:16:51.947 --> 00:17:09.561
The next feature is hyper-evolution, which I sort of like this one, and he's touched on this a couple times, but he had a really cool statement during this section where he said,"Imagine the industrial era where our physical world changed drastically." Or I guess that was my own commentary.
00:17:09.582 --> 00:17:16.721
So, you know, you have the industrial era where people were building roads, people were building factories, fabricating things, and our...
00:17:16.976 --> 00:17:17.645
Yeah, exactly.
00:17:17.645 --> 00:17:24.955
Like our, our-- The physical world around us was just utterly every year, it was like some crazy new...
00:17:25.016 --> 00:17:31.705
Like a telegraph wire would be there, and you're like,"What is that?" And it, you know, confers some crazy new ability on society.
00:17:31.887 --> 00:17:46.057
So he's saying the next 40 years will see both the world of atoms rendered into bits, and new levels of complexity and fidelity in the world of bits rendered back into atoms, which is crazy if you think about it.
00:17:46.067 --> 00:17:50.483
Like I, I found that statement to be kind of scary in a way
00:17:51.041 --> 00:17:58.941
he's calling this out, this thing that I feel like I've noticed where software feels like it's going super, super fast.
00:17:58.941 --> 00:18:01.131
It's getting better all the time at a super rapid pace.
00:18:01.131 --> 00:18:02.760
You know, you've got Moore's Law and what have you.
00:18:03.070 --> 00:18:14.570
Whereas it feels like the manufacture of physical i- items and objects is still, like, not that different from, uh, maybe at the end of the Industrial Revolution.
00:18:15.059 --> 00:18:15.460
Yeah
00:18:15.692 --> 00:18:27.959
but we're seeing-- we're starting to see inklings, and I think the 3D printer revolution is this crude first step at, like, these more flexible modes of, of manufacture.
00:18:28.309 --> 00:18:28.829
Totally
00:18:29.175 --> 00:18:42.109
And I think, you know, if, if we take that into the r- realm of biology, and I think we've talked about this before, but it's like if you can start to manufacture things at the, at atom level precision, it's like that's when you can do anything.
00:18:42.119 --> 00:18:44.420
You know, you can make metamaterials, you can do what have you.
00:18:44.799 --> 00:18:49.369
The-- I guess the name of the game there is gonna be the speed of manufacture.
00:18:49.369 --> 00:18:52.549
Like, d- are you gonna need to, like, grow this thing?
00:18:52.559 --> 00:18:56.940
Like, is it, you know, it's gonna take a year, and then you grow a car, you know, or something?
00:18:57.336 --> 00:18:57.676
but
00:18:57.804 --> 00:19:04.297
Yeah, it's, it, it's one of those things where I feel like there's a confluence of, you know, just technology.
00:19:04.336 --> 00:19:06.721
but not just technology, like ergonomics almost.
00:19:06.938 --> 00:19:19.400
I watch some YouTube people where it's like they literally just have some machining, tool in their garage that can, like, make a piece of stainless steel.
00:19:19.759 --> 00:19:20.619
You know, they're like...
00:19:20.910 --> 00:19:24.980
Like, I just watched a video where a guy was, like, exploding the skin off apples.
00:19:26.009 --> 00:19:36.826
A- and he was, you know, which he was doing it with pressure, where it's like you build up pressure in this chamber and then you, like, release it or whatever, and so the moisture in the apple superheats and the skin flies off.
00:19:38.205 --> 00:19:44.695
he had to build this, like, pressure chamber, and it's like you're building a pressure chamber, you need to use, like, incredibly strong materials.
00:19:45.145 --> 00:19:46.836
And he just did it in his garage.
00:19:47.076 --> 00:19:50.256
Like, he was literally just, like, laser printing metal.
00:19:50.364 --> 00:19:59.961
And, you know, obviously not everyone can afford that sort of thing, but I feel like that is the exact type of thing that is becoming easier.
00:19:59.990 --> 00:20:01.480
It's becoming more ergonomic.
00:20:01.480 --> 00:20:10.830
It's easier to just buy a laser printer and then, like, hook a computer up to it with CAD and, like, print some, you know, model of a pressure chamber that you just made.
00:20:11.520 --> 00:20:13.624
and, that's the trajectory of things.
00:20:14.021 --> 00:20:25.337
and I do think that that slowly leads to what Mustafa's talking about, where it's like people are just gonna be working on quantum computers at the atomic scale, fabricating things instantly.
00:20:25.498 --> 00:20:26.607
Like, it's gonna be crazy.
00:20:27.449 --> 00:20:31.528
Yeah, YouTuber makers are getting into, like, wild territory.
00:20:31.528 --> 00:20:38.354
There's one that's, has been doing his own, semiconductor fabrication, like, in his, in his apartment.
00:20:38.480 --> 00:20:43.510
Is he doing, is he dropping the little tin balls and, like, exploding them with lasers or whatever?
00:20:43.523 --> 00:20:48.663
No, I don't think he's, he's making features quite, quite so small, but it is, it is crazy.
00:20:48.683 --> 00:21:16.077
But, but, talking about this endless acceleration, this is something that has worried me a little bit before because There's a, there's I think a pitfall you can make where it always seems, and maybe I've relayed this, this kind of thought exercise to you before, where it's like it can always seem beneficial to improve the tools that you're using to make the next thing.
00:21:16.528 --> 00:21:25.657
And like you can spend 100% of your time just working on like just producing a tool that produces better tools and like it, it, you know, refines itself.
00:21:25.988 --> 00:21:28.036
And it's like, you're not actually doing anything.
00:21:28.056 --> 00:21:34.692
You know, you're only-- You know, it feels like you're doing something because, you know, you actually are having an effect, but like it's all self-referential.
00:21:34.732 --> 00:21:35.393
You know, you're only...
00:21:35.403 --> 00:21:50.893
And this is, you know, when I'm working on this agent harness in a, in a way that I, I sometimes I need to take a s- take a step back and be like,"Wait a minute," like I'm not ac- not actually doing anything with the, with this tool that I've built, and the whole point was that I, it should be able to like do things for me, and then all it's doing is building itself.
00:21:51.844 --> 00:21:53.663
so I, I do worry about this
00:21:54.124 --> 00:21:56.423
Well, but that's what AGI is, you know?
00:21:56.423 --> 00:22:05.005
It's like you h- you create some sort of self-improving atom and it eventually becomes a god, you know?
00:22:05.015 --> 00:22:07.015
So maybe, maybe that's what you're achieving.
00:22:08.180 --> 00:22:10.660
Well, yeah, not not, not just yet.
00:22:10.670 --> 00:22:14.759
Definitely still have these things making, doing very stupid, stupid things.
00:22:14.914 --> 00:22:21.474
You gotta get past the, like, delete this code then immediately re-add it issues, and then, and then you'll be close.
00:22:22.055 --> 00:22:32.334
I'm wondering actually if we should stop here,'cause, uh, we- we've only covered two of the things, but we're at almost a half hour And uh, feels like a pretty good nugget
00:22:32.528 --> 00:22:39.460
All right, but yeah, let's, let's stop there'cause two is half of four, we can cover the next two features.
00:22:39.460 --> 00:22:40.960
I'm not gonna, not gonna say what they are.
00:22:41.329 --> 00:22:45.772
Gonna have to tease, features of the, uh, coming wave for next week
00:22:46.692 --> 00:22:47.471
Sounds good.
00:22:47.971 --> 00:22:48.422
See you,
00:22:48.531 --> 00:22:48.561
Matt
00:22:48.605 --> 00:22:48.964
All right.
00:22:49.095 --> 00:22:49.424
See you
00:00:08.329 --> 00:00:15.048
Hello everyone And welcome to the switch statement podcast It's a podcast for investigations into miscellaneous tech topics
00:00:15.981 --> 00:00:19.411
This is episode nine of The Coming Wave by Mustafa Suleiman.
00:00:20.498 --> 00:00:21.667
We'd love to hear from you.
00:00:21.878 --> 00:00:25.748
Send your comments and feedback to team@tkp.co.
00:00:30.559 --> 00:00:31.320
Hey, John.
00:00:31.339 --> 00:00:31.760
How you doing?
00:00:32.848 --> 00:00:33.579
Hello, Matt.
00:00:33.598 --> 00:00:34.939
I'm doing very well.
00:00:34.939 --> 00:00:35.579
How are you doing?
00:00:36.795 --> 00:00:37.744
doing okay.
00:00:37.765 --> 00:00:38.484
Doing okay.
00:00:38.484 --> 00:00:45.273
Just, I've been really acutely experiencing the limitations of AI, recently
00:00:46.268 --> 00:00:47.398
Oh my God, man.
00:00:47.487 --> 00:00:59.427
Well, I'm glad to hear you say that because I'm surrounded by people with AI psychosis that just believe AI can do anything, and they'll show me some vibe coded POS and be like, Look at this app.
00:00:59.704 --> 00:01:06.664
Like, it's amazing." And I just instantly see, like, 900 flaws in it, and I'm just like,"Oh my God."
00:01:06.784 --> 00:01:07.144
flaws
00:01:07.737 --> 00:01:19.400
It's so funny'cause I, I have totally drank the AI Kool-Aid, but at the same time, I feel like I still am super skeptical of, like, so many AI things
00:01:20.239 --> 00:01:23.239
I have, I, I have what I'm, I'm gonna coin a new term.
00:01:23.248 --> 00:01:40.558
I have AI ambicosis, which is I am simultaneously completely bought into the hype and also, like, constantly fucking, like, pissed off about how shitty AI is.
00:01:41.516 --> 00:01:49.027
Dude, it's like, I almost feel like people don't realize that it can't do everything.
00:01:49.037 --> 00:02:04.756
Like, it's so hard to explain'cause AI does have this remarkable thing where it can produce such a close approximation of, like, a real thing that it can, like, convince people that, like, it's actually doing that real thing.
00:02:04.900 --> 00:02:06.219
And I wish I had an example.
00:02:06.239 --> 00:02:09.786
Like, this would be so much easier to explain if I had examples.
00:02:09.786 --> 00:02:16.216
I don't wanna, like, go into super detail about my job or anything, but, uh, uh, like a web app, you know?
00:02:16.216 --> 00:02:25.288
It's like you can vibe code a web app in two seconds, but it's like y- you know, you gotta deploy it, you gotta serve it, you gotta host it, you have to have all these other things.
00:02:25.298 --> 00:02:27.747
It's gotta have security, it's gotta have auth, you know?
00:02:28.277 --> 00:02:29.217
It-- So it's...
00:02:29.647 --> 00:02:41.062
Which, you know, by the way, AI can do all those things, but it's just like you can't instantly be convinced by, like, the two-second vibe code that you have solved any real problems.
00:02:41.534 --> 00:02:41.974
Yeah.
00:02:42.245 --> 00:02:42.664
Yeah.
00:02:42.675 --> 00:02:51.215
No, and I think-- I the greenfield performance is tricking everyone into believing that these are general purpose intelligences,
00:02:51.683 --> 00:02:52.614
Omni use.
00:02:53.807 --> 00:03:07.967
it's funny because I feel like I exist in, a weird space where it feels like either, you are all pro and you're like,"These things can do anything," or you're all anti and you're like,"These things can do n-" And it's like, no, no, I want it to do this.
00:03:07.997 --> 00:03:09.627
Like, I want it to be able to do this
00:03:09.788 --> 00:03:10.248
Yeah
00:03:10.638 --> 00:03:17.195
and I'm trying to get it to do this thing, but I just acknowledge that it absolutely cannot.
00:03:17.195 --> 00:03:19.152
And the question is, like how do you deal with that?
00:03:19.152 --> 00:03:22.973
And, and maybe, maybe it's time to actually talk about the content of this chapter.
00:03:23.737 --> 00:03:28.337
Well, we, we sort of have been, I feel you know,'cause we are talking about AI
00:03:29.685 --> 00:03:34.235
and this, and this is the thing, you know, I think maybe, okay, if the, we have the cleanest segue here.
00:03:34.534 --> 00:03:55.294
You know, he talks about these, this asymmetry of power, and the whole point of this project that I'm working on is, like, a single person should be able to be, like, the CEO of their swarm of coding agents, and they just, articulate a vision, and then the agents just, they just do it all, and it's beautiful and perfect, and they know exactly what you mean.
00:03:55.900 --> 00:03:58.300
Yeah, and there's no intervention necessary
00:03:58.544 --> 00:03:59.824
no intervention necessary.
00:04:00.936 --> 00:04:01.885
asymmetry.
00:04:01.925 --> 00:04:17.776
So, and you know, I do think there's truly an asymmetry between programmers today who are learning to leverage AI agents within their day-to-day workflows versus programmers.
00:04:17.795 --> 00:04:25.105
You know, there's some Luddite programmers who are sort of refusing to adopt AI tooling.
00:04:26.065 --> 00:04:28.355
And, you know, there's always gonna be Luddite programmers.
00:04:28.355 --> 00:04:31.255
Like, I, I actually remember this when web frameworks were coming out.
00:04:31.255 --> 00:04:33.555
Like, I was kind of a Luddite programmer.
00:04:33.555 --> 00:04:35.925
I was like,"Why would you use a web framework?
00:04:35.925 --> 00:04:40.505
Like, I'm just gonna keep using MooTools," which was, like, a JavaScript library.
00:04:41.045 --> 00:04:48.286
D- I just, like, didn't really understand the, the necessity of it, uh, or the, or even the utility of it.
00:04:49.105 --> 00:04:53.846
And, uh, you know, I, I-- my websites were worse.
00:04:55.485 --> 00:05:05.649
So it's kind of this element of, like, y- you know, there is this a huge advantage in adoption of some new technology, but you know, it's a bleeding edge.
00:05:05.838 --> 00:05:08.608
It also has issues, which you were discussing.
00:05:14.550 --> 00:05:17.350
he actually starts this chapter in a point that...
00:05:17.788 --> 00:05:32.622
And, and I guess he's commenting on actual, you know, actual events that had happened in the war in Ukraine, but it was pretty prescient, like that, that turned out to be very indicative of the direction that everything was gonna be moving.
00:05:32.672 --> 00:05:47.422
He was talking about drones, and I didn't, I don't know, I hadn't heard this, that, you know, it, it was very widely reported that Russia had this really long, line of like tanks and trucks and what have you, this 40-kilometer line of, of armored vehicles,
00:05:47.805 --> 00:05:48.204
Right
00:05:48.451 --> 00:05:50.132
and, and then they were all just stuck.
00:05:50.322 --> 00:06:17.596
What I hadn't heard was that there was this ragtag, uh, group of I don't know what the right word is, but like group of soldiers, militia, who in cover of darkness, like went and were able to use drones to disable a couple of vehicles at the f- head of the line, and then it's like, all right, doesn't matter how long the line of armored vehicles is, they're not going anywhere because all of the, the ground around was unpassable.
00:06:17.617 --> 00:06:17.997
So,
00:06:18.384 --> 00:06:30.995
Yeah, this is an, uh, an element where, and, uh, this is sort of the classic thing where it's like if you're invading a country that's sort of fighting for its own existence, it's gonna use all the advantages it has at its disposal.
00:06:31.005 --> 00:06:36.355
Whereas if you're the Russian military, it's like these guys are barely...
00:06:36.514 --> 00:06:38.324
You know, they don't even wanna be there.
00:06:38.752 --> 00:06:39.153
Right
00:06:39.264 --> 00:06:41.644
know, they're, like, doing the bare minimum.
00:06:41.673 --> 00:06:51.043
Obviously they're, they wanna survive, uh, but, you know, they're just, following orders, doing the conventional whatever they've historically been doing.
00:06:51.802 --> 00:07:02.432
And so yeah, I mean, this is this perfect example of asymmetric warfare where Ukraine fighting for its own existence is very willing to adopt whatever new technology it can.
00:07:03.012 --> 00:07:08.942
Also, Ukrainians are geniuses, so they're very aware of, like, new technologies and drones and things like that.
00:07:09.303 --> 00:07:15.689
and so yeah, I, I do think it's a good example'cause it's this very real world thing where we all kinda witness this live.
00:07:15.769 --> 00:07:20.559
It's like the Russian military supposedly was this mighty behemoth, and
00:07:20.562 --> 00:07:32.853
The second most powerful military, like I, before this war, I think most people would be like,"Yep, r- you know, Russia's got the second most powerful military." You know, because there's always this like kind of ambiguity, like Russia, China.
00:07:33.523 --> 00:07:38.202
Um, but it's like are they even the second most pow- capable military?
00:07:38.233 --> 00:07:41.423
I mean, y- sorry, are they even the third most capable military?
00:07:41.432 --> 00:07:46.783
Like at this point it's like they, feels like they fell like several, several ranks
00:07:47.182 --> 00:07:48.882
No, it was very embarrassing.
00:07:48.992 --> 00:08:02.252
And I think, uh, you know, Putin, who prides himself on image and just the optics of everything, like, man, he must have not enjoyed that situation at all.
00:08:02.992 --> 00:08:04.442
And I feel like it continues
00:08:05.144 --> 00:08:11.735
Trump is doing a pretty good job of, uh, of fighting Putin to who- to see who could fall down
00:08:11.879 --> 00:08:12.718
Ugh, God
00:08:12.735 --> 00:08:16.524
the military power, firepower list as fast as possible
00:08:16.773 --> 00:08:19.163
Yeah, Trump's handling of Iran has been just incredibly embarrassing, which is kinda sad.
00:08:19.163 --> 00:08:33.023
I mean, I, uh, it's weird because I feel like whenever the United States does something like this on the global stage or what have you, like, I am hopeful.
00:08:33.062 --> 00:08:46.490
You know, even, even if I dislike the current administration or whatever, like, I'm hopeful that we'll just do the right thing and, like, I felt the same way when Bush was president actually,'cause we were doing all this military stuff, and I was just...
00:08:46.880 --> 00:08:57.750
Yeah, I, I was hopeful that the United States would sort of comport itself well, and that obviously ended up being just utterly, utterly false during the Bush era.
00:08:58.461 --> 00:08:59.520
But with Trump, I had hope.
00:08:59.520 --> 00:08:59.801
Like
00:09:00.864 --> 00:09:06.264
by, like, by con- c- by contrast, I feel like the operation in Venezuela,
00:09:06.985 --> 00:09:07.375
Right.
00:09:07.475 --> 00:09:08.144
Exactly.
00:09:08.744 --> 00:09:12.274
that was one where I was like,"Okay, well, clearly someone did their homework.
00:09:12.754 --> 00:09:22.594
They figured out the, you know, the reality of the situation on the ground, the feasibility." It was this incredibly targeted...
00:09:22.793 --> 00:09:25.974
It was like s- it was like an hour or something, you know?
00:09:25.974 --> 00:09:26.514
It was crazy.
00:09:26.524 --> 00:09:28.094
It was like you blinked and you missed it.
00:09:28.144 --> 00:09:30.774
Uh, it was already over by the time anyone knew it was happening.
00:09:30.774 --> 00:09:39.884
So, um, whereas, like, when everything in Iran started happening, I was kind of like, listen, if this...
00:09:39.894 --> 00:09:41.274
If they could pull this off...
00:09:41.283 --> 00:09:49.953
A- and, like, the other thing is, like, it's, it's important to remember that the Iranian regime is, like, this incredibly oppressive dictatorship.
00:09:49.953 --> 00:09:57.864
So it's like if they were able to get, like, overthrow the government, like, for the people of Iran, like, that would be a good.
00:09:58.364 --> 00:10:07.443
But they clearly didn't do their homework, and, like, they were just too optimistic about their, their chances, and now we're in this fucking quagmire.
00:10:07.443 --> 00:10:11.734
It's just, like, so stupid and a waste of all of American taxpayer dollars
00:10:12.078 --> 00:10:17.999
It, it's also just such a-- I mean, like you're saying, it's like they just misplayed their hand.
00:10:18.009 --> 00:10:20.917
And I feel like we know about Iran.
00:10:21.047 --> 00:10:22.547
Like, we understand Iran.
00:10:22.557 --> 00:10:32.927
We-- There are people in intelligence who know that the IRGC is, like, incredibly entrenched in their administration, and it's incredibly difficult to topple that.
00:10:33.297 --> 00:10:45.716
Unlike Venezuela, which, you know, Maduro was also a piece of trash and sort of corrupt, and his administration was filled with people just taking whatever oil riches Venezuela has.
00:10:46.147 --> 00:10:53.258
Uh, the, the average Venezuelan should be doing well based on the opulence of the country's natural resources.
00:10:53.908 --> 00:10:56.658
And so, you know, Maduro was not that popular.
00:10:56.668 --> 00:11:06.498
And so in that situation, taking him out, I think, you know, the right move i-in the way that they did it, I think was the right move.
00:11:07.398 --> 00:11:17.018
Iran, on the other hand, you have to be much more sort of careful and surgical and, uh,'cause you can't just take out the head.
00:11:17.018 --> 00:11:26.038
I mean, the Ayatollah, sure, they, he has a lot of power, but it's like there's this entire power structure that they've worked for decades to put in place.
00:11:26.962 --> 00:11:27.423
Yeah
00:11:27.817 --> 00:11:28.107
yeah.
00:11:28.177 --> 00:11:35.817
Anyway, uh, maybe we shouldn't wade into, like, this type of territory, but just frustrating to me.
00:11:36.932 --> 00:11:37.982
No, I, I feel you.
00:11:37.982 --> 00:11:53.450
But, but, um, but anyway, so getting back to the core point, Ukraine is able to use$1,400 drones from DJI to disable a$10 million tank a-
00:11:53.773 --> 00:12:12.836
He makes this point about transfer of power, and this is sort of what new technology can do a lot of the time is, like, introduce this asymmetry where these traditional historical power centers, you know, and the things that they did to remain in power are no longer as effective.
00:12:12.936 --> 00:12:23.541
There's some new thing that's way more effective, and if you're this entrenched power structure, it's gonna be difficult for you to leverage that'cause the-- you have this inertia in some particular direction
00:12:24.206 --> 00:12:39.225
It's so weird because it feels like it's moving in both directions where the power of the individual is being expanded by this, but also like, it's also a massive centralization of power in the hands of all of the AI companies.
00:12:39.316 --> 00:12:39.855
You know what I mean?
00:12:39.855 --> 00:12:45.289
So it's like, it, it is kind of confusing, like I'm not sure, uh, you know, net-net how to feel.
00:12:45.309 --> 00:12:54.921
Is this net empowering to the individual, or is this net empowering to the entrenched companies, uh, you know, and, and governments?
00:12:55.085 --> 00:13:00.279
I hate to say it, but I think this is gonna be mostly net powering to entrenched companies.
00:13:00.279 --> 00:13:05.687
'Cause I think for a long time, the harness is gonna be what is truly...
00:13:06.118 --> 00:13:08.217
you know, has, has the true utility.
00:13:08.217 --> 00:13:10.248
It's like AIs are incredible.
00:13:10.368 --> 00:13:17.736
They can, you know, they can return responses that are, like, on the level of human intelligence.
00:13:18.181 --> 00:13:22.962
But it's just hard to accomplish anything with that until you have a harness.
00:13:23.052 --> 00:13:26.043
And a harness is, is still really, really hard to build.
00:13:26.241 --> 00:13:43.620
Uh, and so I think, you know, at least for a while, there- you're gonna have these power players that know how to build harnesses, have, you know, the bags of tools and, uh, whatever necessary in order to create truly useful harnesses.
00:13:43.966 --> 00:13:51.206
And, you know, they will they'll be the ones getting most of the, you know, reward, I guess, from the AI revolution.
00:13:51.385 --> 00:13:52.086
Which I hate to...
00:13:52.456 --> 00:13:52.865
I don't know.
00:13:52.895 --> 00:13:56.405
Like, I wish it was different,'cause I think it's actually different with models.
00:13:56.895 --> 00:14:07.518
I think you- today you have these frontier models and, you know, companies are spending so much money on these frontier models, building up infrastructure and training and running inference.
00:14:08.118 --> 00:14:17.398
I, I actually think in the future, frontier models won't be nearly as relevant because the average open source model will be, like, as good or better
00:14:18.860 --> 00:14:19.551
Exactly.
00:14:19.571 --> 00:14:26.490
No, that's what I was just about to say is that we have all these incredibly good open weight models.
00:14:26.500 --> 00:14:28.640
Not to mention that distillation is always gonna
00:14:28.850 --> 00:14:29.220
Right
00:14:29.921 --> 00:14:45.341
a problem where it's like, yeah, you spend all this money to create this frontier caliber model, and then it just takes a month or two for one of the, you know, another lab, an open lab to, to create a model with just as good performance.
00:14:45.760 --> 00:14:58.041
Uh, which, which actually scares me because in a way it's a disincentive from training those higher capability models because it's like, why am I gonna do this if another company's just gonna come along and steal our lunch?
00:14:58.524 --> 00:15:03.794
Luckily, these companies don't operate on logic, so they won't, they won't
00:15:03.870 --> 00:15:19.677
it is interesting because i- in a way, like it's, it's damned if they do, damned if they don't, because they, then they have a bunch of other domestic AI companies that are nipping at their heels, so if they don't continue to advance, then they're gonna get left in the dust, so
00:15:20.104 --> 00:15:20.663
Absolutely.
00:15:20.663 --> 00:15:25.573
There's a lot of zero-sum games around that and like CapEx and things like that, so
00:15:31.471 --> 00:15:52.716
so there was a quote in here that I really wanted to call out, and this is kind of what I was alluding to when I was talking about this agent swarm thing that I've been building, which is he says, There's no obvious reason why a single operator with enough wherewithal could not control a swarm of thousands of drones." Which like, that's terrifying just by itself.
00:15:53.087 --> 00:16:06.250
but I think it's true if you take, take a step back and you think about it in the context of like if you have someone who, who nails the like wiss하시는земcontrarved absor1치는rau amin conti jeopardy, amendsงาม�ρ 燕 世 了 解 到 sudden
00:16:19.927 --> 00:16:30.927
Yeah Totally.
00:16:31.097 --> 00:16:31.427
Yeah.
00:16:31.567 --> 00:16:34.357
First, first single individual billion-dollar company.
00:16:34.368 --> 00:16:36.418
I feel like a lot of people are talking about this.
00:16:36.427 --> 00:16:37.967
I think it's gonna happen.
00:16:38.418 --> 00:16:41.577
so yeah, there's asymmetry, which is one of the four things.
00:16:41.658 --> 00:16:45.268
Did we ever say that this chapter was called Four Features of the Coming Wave?
00:16:45.508 --> 00:16:46.238
I don't know if we said
00:16:46.462 --> 00:16:48.822
I don't think we did, but thank you, yes, for doing that
00:16:49.518 --> 00:16:51.788
So the first feature was asymmetry.
00:16:51.947 --> 00:17:09.561
The next feature is hyper-evolution, which I sort of like this one, and he's touched on this a couple times, but he had a really cool statement during this section where he said,"Imagine the industrial era where our physical world changed drastically." Or I guess that was my own commentary.
00:17:09.582 --> 00:17:16.721
So, you know, you have the industrial era where people were building roads, people were building factories, fabricating things, and our...
00:17:16.976 --> 00:17:17.645
Yeah, exactly.
00:17:17.645 --> 00:17:24.955
Like our, our-- The physical world around us was just utterly every year, it was like some crazy new...
00:17:25.016 --> 00:17:31.705
Like a telegraph wire would be there, and you're like,"What is that?" And it, you know, confers some crazy new ability on society.
00:17:31.887 --> 00:17:46.057
So he's saying the next 40 years will see both the world of atoms rendered into bits, and new levels of complexity and fidelity in the world of bits rendered back into atoms, which is crazy if you think about it.
00:17:46.067 --> 00:17:50.483
Like I, I found that statement to be kind of scary in a way
00:17:51.041 --> 00:17:58.941
he's calling this out, this thing that I feel like I've noticed where software feels like it's going super, super fast.
00:17:58.941 --> 00:18:01.131
It's getting better all the time at a super rapid pace.
00:18:01.131 --> 00:18:02.760
You know, you've got Moore's Law and what have you.
00:18:03.070 --> 00:18:14.570
Whereas it feels like the manufacture of physical i- items and objects is still, like, not that different from, uh, maybe at the end of the Industrial Revolution.
00:18:15.059 --> 00:18:15.460
Yeah
00:18:15.692 --> 00:18:27.959
but we're seeing-- we're starting to see inklings, and I think the 3D printer revolution is this crude first step at, like, these more flexible modes of, of manufacture.
00:18:28.309 --> 00:18:28.829
Totally
00:18:29.175 --> 00:18:42.109
And I think, you know, if, if we take that into the r- realm of biology, and I think we've talked about this before, but it's like if you can start to manufacture things at the, at atom level precision, it's like that's when you can do anything.
00:18:42.119 --> 00:18:44.420
You know, you can make metamaterials, you can do what have you.
00:18:44.799 --> 00:18:49.369
The-- I guess the name of the game there is gonna be the speed of manufacture.
00:18:49.369 --> 00:18:52.549
Like, d- are you gonna need to, like, grow this thing?
00:18:52.559 --> 00:18:56.940
Like, is it, you know, it's gonna take a year, and then you grow a car, you know, or something?
00:18:57.336 --> 00:18:57.676
but
00:18:57.804 --> 00:19:04.297
Yeah, it's, it, it's one of those things where I feel like there's a confluence of, you know, just technology.
00:19:04.336 --> 00:19:06.721
but not just technology, like ergonomics almost.
00:19:06.938 --> 00:19:19.400
I watch some YouTube people where it's like they literally just have some machining, tool in their garage that can, like, make a piece of stainless steel.
00:19:19.759 --> 00:19:20.619
You know, they're like...
00:19:20.910 --> 00:19:24.980
Like, I just watched a video where a guy was, like, exploding the skin off apples.
00:19:26.009 --> 00:19:36.826
A- and he was, you know, which he was doing it with pressure, where it's like you build up pressure in this chamber and then you, like, release it or whatever, and so the moisture in the apple superheats and the skin flies off.
00:19:38.205 --> 00:19:44.695
he had to build this, like, pressure chamber, and it's like you're building a pressure chamber, you need to use, like, incredibly strong materials.
00:19:45.145 --> 00:19:46.836
And he just did it in his garage.
00:19:47.076 --> 00:19:50.256
Like, he was literally just, like, laser printing metal.
00:19:50.364 --> 00:19:59.961
And, you know, obviously not everyone can afford that sort of thing, but I feel like that is the exact type of thing that is becoming easier.
00:19:59.990 --> 00:20:01.480
It's becoming more ergonomic.
00:20:01.480 --> 00:20:10.830
It's easier to just buy a laser printer and then, like, hook a computer up to it with CAD and, like, print some, you know, model of a pressure chamber that you just made.
00:20:11.520 --> 00:20:13.624
and, that's the trajectory of things.
00:20:14.021 --> 00:20:25.337
and I do think that that slowly leads to what Mustafa's talking about, where it's like people are just gonna be working on quantum computers at the atomic scale, fabricating things instantly.
00:20:25.498 --> 00:20:26.607
Like, it's gonna be crazy.
00:20:27.449 --> 00:20:31.528
Yeah, YouTuber makers are getting into, like, wild territory.
00:20:31.528 --> 00:20:38.354
There's one that's, has been doing his own, semiconductor fabrication, like, in his, in his apartment.
00:20:38.480 --> 00:20:43.510
Is he doing, is he dropping the little tin balls and, like, exploding them with lasers or whatever?
00:20:43.523 --> 00:20:48.663
No, I don't think he's, he's making features quite, quite so small, but it is, it is crazy.
00:20:48.683 --> 00:21:16.077
But, but, talking about this endless acceleration, this is something that has worried me a little bit before because There's a, there's I think a pitfall you can make where it always seems, and maybe I've relayed this, this kind of thought exercise to you before, where it's like it can always seem beneficial to improve the tools that you're using to make the next thing.
00:21:16.528 --> 00:21:25.657
And like you can spend 100% of your time just working on like just producing a tool that produces better tools and like it, it, you know, refines itself.
00:21:25.988 --> 00:21:28.036
And it's like, you're not actually doing anything.
00:21:28.056 --> 00:21:34.692
You know, you're only-- You know, it feels like you're doing something because, you know, you actually are having an effect, but like it's all self-referential.
00:21:34.732 --> 00:21:35.393
You know, you're only...
00:21:35.403 --> 00:21:50.893
And this is, you know, when I'm working on this agent harness in a, in a way that I, I sometimes I need to take a s- take a step back and be like,"Wait a minute," like I'm not ac- not actually doing anything with the, with this tool that I've built, and the whole point was that I, it should be able to like do things for me, and then all it's doing is building itself.
00:21:51.844 --> 00:21:53.663
so I, I do worry about this
00:21:54.124 --> 00:21:56.423
Well, but that's what AGI is, you know?
00:21:56.423 --> 00:22:05.005
It's like you h- you create some sort of self-improving atom and it eventually becomes a god, you know?
00:22:05.015 --> 00:22:07.015
So maybe, maybe that's what you're achieving.
00:22:08.180 --> 00:22:10.660
Well, yeah, not not, not just yet.
00:22:10.670 --> 00:22:14.759
Definitely still have these things making, doing very stupid, stupid things.
00:22:14.914 --> 00:22:21.474
You gotta get past the, like, delete this code then immediately re-add it issues, and then, and then you'll be close.
00:22:22.055 --> 00:22:32.334
I'm wondering actually if we should stop here,'cause, uh, we- we've only covered two of the things, but we're at almost a half hour And uh, feels like a pretty good nugget
00:22:32.528 --> 00:22:39.460
All right, but yeah, let's, let's stop there'cause two is half of four, we can cover the next two features.
00:22:39.460 --> 00:22:40.960
I'm not gonna, not gonna say what they are.
00:22:41.329 --> 00:22:45.772
Gonna have to tease, features of the, uh, coming wave for next week
00:22:46.692 --> 00:22:47.471
Sounds good.
00:22:47.971 --> 00:22:48.422
See you,
00:22:48.531 --> 00:22:48.561
Matt
00:22:48.605 --> 00:22:48.964
All right.
00:22:49.095 --> 00:22:49.424
See you
