Latest / Nonsense with cj little and Jeff Parker / Sovereign AI: The Quest For Independence and The "Holy Trinity"

Transcript

Sovereign AI: The Quest For Independence and The "Holy Trinity" · Nonsense with cj little and Jeff Parker

• 0:00
  Welcome to Nonsense Season 4, Episode 13, our big story.
• 0:02
  I'm Jeff Parker. I'm CJ Little.
• 0:04
  This time, we're looking at why borrowing AI is a risky move for
• 0:07
  most nations. Up next So you.
• 0:09
  Really think that it's risky for them to to borrow.
• 0:11
  AI, by the way, it's the mostly the United States and China that
• 0:14
  you'd be borrowing AI from, and I wouldn't trust either one of
• 0:16
  us. Yeah, that's a good point.
• 0:17
  Who would you trust? Like who would be the best?
• 0:19
  Is like Switzerland? Like do you need like a neutral?
• 0:21
  Oh, I would probably trust Switzerland.
• 0:22
  Yeah, they have great privacy laws.
• 0:24
  Yeah, Switzerland be Iceland out. 22 great choices.
• 0:27
  I would trust the Swedish AI because like what other how
• 0:30
  intent could they have? Well, they want to steal all
• 0:32
  your pickled fish. Please feel free, I'll let them
• 0:35
  write in. Take it off.
• 0:36
  Good luck if you can find that it's yours.
• 0:38
  OK, the new space race Why nations are treating AI and
• 0:41
  compute as critical infrastructure.
• 0:42
  Let's go the. Only pickled fish in my house is
• 0:44
  what my kids are making under their bed without my knowledge.
• 0:46
  I'm not coming over. In 2026, sovereign AI has
• 1:01
  exploded from a niche policy term into $100 billion global
• 1:05
  imperative. The sudden urgency is driven by
• 1:07
  a realization that AI is no longer just software, it's the
• 1:10
  21st century equivalent of the power grid.
• 1:13
  Nations are no longer content being AI takers.
• 1:15
  They are racing to be AI makers to protect their data, their
• 1:19
  culture, and control of their power, water and air traffic.
• 1:22
  At its core, sovereign AI is the radical idea that a nation state
• 1:25
  should own and control the physical and digital foundations
• 1:28
  of its artificial intelligence, rather than renting them from a
• 1:30
  handful of foreign tech giants. To be truly sovereign, a country
• 1:34
  must control the Holy Trinity of the AI, age, domestic data,
• 1:37
  local energy grids, and most crucially, the silicon hardware
• 1:40
  required to run the math. Well, yeah, it's not going to
• 1:43
  run on nothing. We.
• 1:43
  Can start. We can go in anywhere you want.
• 1:45
  You want to talk about the silicon hardware because that's
• 1:46
  going to be impossible to really own.
• 1:48
  That's all going to be made not in your state.
• 1:50
  It's unless you are Taiwan or possibly China or possibly the
• 1:54
  US. Yeah, that doesn't scale out
• 1:56
  horizontally at all, like at all.
• 1:58
  It barely like we don't Yeah, think we're kidding ourselves to
• 2:01
  pretend like we have it in the USI mean, we have something in
• 2:03
  the US We still have Intel fabs here.
• 2:05
  Yeah, we sort of have a budding TSMC fab deployment, tiny, but
• 2:09
  something. I mean, it's better than what we
• 2:11
  had five years ago, which was nothing.
• 2:13
  So. Sure, yeah, sure.
• 2:14
  But like if you are a really like to your point, any other
• 2:17
  country, like maybe India could do it.
• 2:20
  Maybe. I don't know.
• 2:21
  They aren't, but they aren't chips, no.
• 2:23
  So you're going to have some dependencies elsewhere that I
• 2:26
  guess it's a good place to start of like how are you going to
• 2:29
  avoid that? I really.
• 2:30
  Don't know. By the way, a big chunk of why
• 2:31
  we don't want to make chips in scale in the US is the volume of
• 2:35
  pollution it causes. Yeah, this is really
• 2:37
  environmentally unfriendly. Then we're just doing it
• 2:39
  somewhere else, like, oh, we'll pollute those folks, that's OK.
• 2:42
  We are, we're, we're, we're polluting Taiwan.
• 2:44
  That's exactly right. That's terrible.
• 2:45
  It is terrible. We feel bad for using my
• 2:48
  computer. Sorry.
• 2:48
  What a great start here. So we're already down a rabbit
• 2:51
  hole. Yeah, you, you aren't gonna have
• 2:52
  sovereignty over chips. You can't make the chips local.
• 2:54
  Energy grids, You're gonna have to do that.
• 2:56
  Yeah, you need to have tons and tons of power to run.
• 2:59
  A. Well, you will hold done.
• 3:00
  Now you say you're going to have to do that, but so many
• 3:02
  countries do import their primary energy sources external.
• 3:05
  Who's importing energy that you All the countries near Germany
• 3:08
  are importing energy just because they have such a huge
• 3:11
  excess of solar. Power.
• 3:12
  It appears that between midnight and 9:00 AM, Germany is a net
• 3:15
  importer of power. Between 10:00 AM and 4:00 PM,
• 3:18
  they are a net exporter. So midday, that is their solar
• 3:20
  peak and they have a massive solar fleet which apparently
• 3:23
  reached around. You ready for this 116 gigawatts
• 3:27
  a lot of solar power in 2025. There's a lot of.
• 3:30
  Solar power for a country, that is.
• 3:31
  Always cloudy. Yeah.
• 3:32
  I, I, I feel like there's some sort of, I don't know what you'd
• 3:36
  call like a Madoff Ponzi thing going on here.
• 3:38
  This doesn't. I'm no solarologist but this
• 3:41
  doesn't make a lot of sense because like, I don't even think
• 3:43
  most people in Germany could describe what the sun looks.
• 3:45
  Like, well, the panels now are so efficient.
• 3:47
  Oh, you just need light. You don't necessarily need
• 3:49
  direct sunlight. Yeah, that's super interesting.
• 3:51
  You would think there would be some alternate ways to do
• 3:53
  stores, energy stores, but but I don't think they do as much of
• 3:55
  that. I guess it's cheaper for them to
• 3:56
  export it to other countries that to offset whatever coal or
• 4:00
  gas fired plants they're using other places.
• 4:02
  Does that make sense? However, the point of all this
• 4:04
  is you need to be if you're a country who's going to have AI,
• 4:07
  you need to have massive power supplies.
• 4:09
  Oh, yeah, it's gonna drink it all up.
• 4:10
  Yeah. And if you want to be sovereign,
• 4:12
  you need it from sources that you can depend on, right, That
• 4:15
  are that are sovereign to your country, which some.
• 4:16
  Countries right Your your neighbor can't shut off your
• 4:19
  power. Yeah.
• 4:19
  So you need hydroelectric, you need something you can pull out
• 4:21
  of the Earth, you need solar. In this case of of Germany, it
• 4:23
  makes a lot of sense, so. Already we're down to like 3
• 4:25
  countries who can do this. No, I think you've got more
• 4:28
  energy dependence than than three.
• 4:29
  You might not not get all the pieces of the puzzle for
• 4:31
  Sovereign AI right, but you I think you just want to get.
• 4:34
  As many as you can. Yeah, it's you're, it's kind of
• 4:36
  like playing chess. You don't need all the pieces
• 4:38
  to. Win.
• 4:39
  Let's explain why countries should care about sovereign AI.
• 4:41
  Yeah, because what if somebody who has a lot of AI in their
• 4:45
  country suddenly folds their arms and says, you know, OK,
• 4:48
  Netherlands, we're not giving you anymore AI.
• 4:49
  We're cutting you off. And meanwhile you you're using
• 4:52
  AI for doing running all sorts of infrastructure your water
• 4:54
  pumps in. Your I Yeah.
• 4:56
  Your sewage system. No.
• 4:57
  Oh look, I hear you, but I I couldn't imagine any modern
• 5:00
  superpower acting that childish. Well, that is the episode then.
• 5:03
  I mean, yeah, I think that's the real especially as of late where
• 5:06
  we have made these, I'm using we in a very generalized, not you
• 5:10
  and I sort of way where we have have made these decisions of
• 5:13
  like, Oh no, we don't like what you're doing, therefore we're
• 5:15
  not going to give you this thing that you have depended on.
• 5:17
  Yeah, sure, sure, sure. And why would you ever trust us
• 5:19
  ever again behaving. So I gotta believe a bunch of
• 5:21
  folks are worried about this because that if you are powering
• 5:24
  parts of your infrastructure off this because it's more
• 5:26
  efficient, it's more cost effective, whatever the the
• 5:29
  reasons are. And then that gets turned off
• 5:31
  because you've got some grumpy, angry quasi dictator that's
• 5:34
  saying, oh, we shouldn't do these things anymore until you
• 5:36
  give me a big gold building with my name on the side of it.
• 5:38
  And Dutch, that's problematic. Dutch.
• 5:40
  Like that's not good. Now let's go back to the first
• 5:42
  one, which is a country must control its domestic data.
• 5:46
  I would argue that's the most important one.
• 5:47
  That's. Actually something you might
• 5:49
  want to think about. For a couple reasons, right?
• 5:51
  Like one, your data is extremely valuable at scale.
• 5:54
  That's all the keys to the Kingdom, but then also how that
• 5:57
  data is consumed and processed and really whether or not it's
• 6:00
  consumed or processed or, or weighted or ranked is going to
• 6:02
  be super important. I do struggle with this one
• 6:04
  because we've already had a, we'll call it like a cultural
• 6:07
  homogenization. Generally speaking, over the
• 6:09
  last, you know, I, I don't know, 50 years, 100 years, like more,
• 6:13
  we seem to be rolling out into a sort of the same color play DoH,
• 6:17
  right? Like we're all these different
• 6:18
  cultures and they all sort of got smashed together.
• 6:21
  Now imagine you got an AI that you're asking questions and if
• 6:24
  it's loaded or even not just loaded with, but if it it
• 6:26
  prioritizes a given culture or set of cultural answers over a
• 6:30
  another, well that seems like a pretty big nail in the coffin of
• 6:33
  cultural relevance. Or like cultural knowledge just
• 6:35
  get sort of smashed out. So you think when I asked my
• 6:38
  Chinese AI, Yeah. What should I have for dinner
• 6:40
  tonight? It's gonna say, well, obviously
• 6:42
  you should go to Fortune Inn and definitely try the chow Mein.
• 6:45
  It's excellent. No, but I think that OK,
• 6:47
  dinner's a great question. If you ask an AI what should I
• 6:48
  have for dinner tonight, a Western AI might say steak and
• 6:51
  potatoes. In Eastern, AI may say dumpling.
• 6:53
  Because you're thinking the AI is only training on data that
• 6:57
  came from the Starbucks in. Yeah, then it would just tell
• 7:01
  you to eat egg bites. I mean really.
• 7:02
  I mean doesn't. I mean it gets its data from all
• 7:04
  over the world. It does get its data from all of
• 7:06
  the world. Drones and and routers are great
• 7:08
  analog for AI in this case of like meaning you want it to be
• 7:11
  developed on soil, but they're far simpler.
• 7:14
  And and to your point, we don't even have those things figured
• 7:16
  out right. So that's a great.
• 7:17
  This is where it becomes interesting.
• 7:18
  Like like what do you want to build at home?
• 7:20
  What do you not want to build at home?
• 7:21
  If I were a foreign entity, I would not trust the US to run my
• 7:26
  power company, to run my sewage company, to run any of the AI
• 7:30
  systems, to run any important infrastructure.
• 7:32
  Therefore, if I didn't have my own.
• 7:34
  U.S. government or the US companies as well?
• 7:36
  Because the US government can control the US companies and cut
• 7:39
  me off. Yeah.
• 7:39
  And so I don't want to make any of those things dependent on on
• 7:43
  US entities. Now I'm not any more excited
• 7:45
  about making the dependent on Chinese entities.
• 7:47
  Yeah. So if my my AI is not running
• 7:49
  locally, that's not how I'm going to make my infrastructure.
• 7:52
  Work, but even it is running locally, I'm not sure that's
• 7:54
  enough. That's kind of the point of
• 7:55
  sovereign AI. I'd argue Maybe the difference
• 7:57
  is like there's a path here where you where you build it all
• 8:00
  yourself or as many of the components as you can, the the
• 8:03
  silicone, the power, the algorithm, the data.
• 8:06
  But there's another path where maybe you just take if you're
• 8:09
  going to end up on one side or the other, right, one side of
• 8:12
  controlling forces in the world or the other, why not just use
• 8:14
  them both? If it's all just AP is integrate
• 8:17
  your system so that you work great with, you know, open AI,
• 8:21
  but you also work great with deep sea, because odds are if
• 8:24
  one cans you, the other one's not.
• 8:26
  Maybe that's the answer. Also, you can probably get to
• 8:28
  the truth if you have multiple AIS from multiple countries.
• 8:31
  You feed all the answers. In and giving them the same same
• 8:34
  query exactly. Then take the average.
• 8:36
  Then take whatever two of the three are telling you.
• 8:38
  But you're right, why not use just multiple AIS and and the
• 8:41
  odds are you won't get cut off by all three.
• 8:43
  Maybe that's the answer. Maybe that maybe the answer is
• 8:45
  really you just say all of our systems have to have some sort
• 8:48
  of redundancy that allow you to to to eliminate as many of the
• 8:53
  dependencies as possible. And yeah, if you look at the
• 8:55
  powers, if one were to turn you off, if the US were to turn you
• 8:59
  off, it is very unlikely that the other superpower play in
• 9:02
  this game. China would, right?
• 9:03
  They'd probably embrace you with open arms.
• 9:05
  And I think vice versa, right? So maybe that is the answer.
• 9:07
  Maybe. Maybe we just cracked the code
• 9:08
  on this and it's like, well, you can go it yourself.
• 9:11
  Yeah. We're making policy here.
• 9:12
  This is great. Shorter episode than I thought
• 9:13
  it was going. To be, I mean, you can go it
• 9:14
  yourself and and build these things, which I do think is a a
• 9:17
  viable path. Maybe you're not building the
• 9:19
  algo, Maybe you're using an open source algorithm.
• 9:21
  You're not building the algo, you're not building the chip.
• 9:24
  You are going to have to generate the power, which is
• 9:26
  nasty. That's not a pleasurable thing
• 9:28
  to put on your environment. Yeah.
• 9:30
  You already, I mean, you already need power for what they're
• 9:32
  doing. So if you're talking about a
• 9:33
  couple percent overhead, maybe that's something that you could
• 9:35
  keep. It's not a couple percent, it's
• 9:37
  way more than a couple percent. Now it looks like in the United
• 9:40
  States specifically, we think that number could get to 9 to
• 9:44
  12%. Wow, is that right?
• 9:46
  Is that I'd love to see how that's a lot of power usage.
• 9:49
  European Union 7 to 10%, China 4 to 5% Ireland 32% by 2026.
• 9:55
  Because of their tax structure makes you want to build it but
• 9:57
  I'm assuming everything in Ireland.
• 9:58
  McKinsey and Goldman and Sacks suggested in the US data centers
• 10:02
  could account for as much as 40% of all net new electricity
• 10:05
  demand added between now and 2030.
• 10:07
  That's new. That is insane.
• 10:09
  But you're building power that you wouldn't have otherwise
• 10:11
  built probably for those data centers.
• 10:13
  Projections aren't always destiny.
• 10:14
  Between 2010 and 2018, data center workloads rose 600%, but
• 10:18
  energy you stayed almost flat because of massive improvements
• 10:21
  in server efficiency. So that's Moore's law, right?
• 10:23
  I mean, if you're 2010 to 2018 is 4 to 5 Moore's law cycles,
• 10:28
  there's your 600%. Easy P easy.
• 10:30
  So I think it's unclear to know exactly how much load we're
• 10:32
  going to need going forward. I suspect that chip efficiency
• 10:35
  improvement will outpace software improvement in the next
• 10:38
  couple of years. I just don't know when, right?
• 10:40
  At some point your algo is good enough for the masses, right?
• 10:43
  And now your chips just keep getting better because Moore's
• 10:45
  Law won't die until the the sun supernovas.
• 10:48
  So like that's going to go on forever.
• 10:50
  So then the smaller countries will be able to run their own
• 10:53
  AI? Yeah, may not be able to make
• 10:54
  their own models, but they certainly will to run models.
• 10:57
  Yes, on Apple Watch, but coming to the the core sort of be the
• 11:00
  path here. Just make your core model
• 11:02
  another plugin in the API, like use your core model in all
• 11:05
  cases, but then if you're if you get short on power, you can just
• 11:08
  use the open AI model. Oh wait, that doesn't work
• 11:10
  anymore because the US is grumpy.
• 11:11
  You didn't export enough Tulip bulbs right?
• 11:13
  Now go to the Chinese AI. Do they tell you to screw off?
• 11:16
  You're done. No.
• 11:16
  They'll be more. Straight to the Chilean AI.
• 11:18
  Do you think there's not going to be Japanese AI?
• 11:20
  Oh. Of course, of course, I mean,
• 11:21
  the Japanese are trying to much like what they do for for
• 11:24
  bluefin tuna. They're trying to hoard as many
• 11:25
  of these chips as they can so that they can also be a
• 11:28
  competitor in this. The problem with hoarding the
• 11:29
  chip is the it's like hoarding milk.
• 11:31
  It's a great idea, but they're going bad with time.
• 11:34
  Sure, you do not want to keep chips for a very, very long
• 11:37
  time. I mean, think about all the H
• 11:38
  100. That was like V chip.
• 11:40
  Oh, man, that was the most advanced chip for AI.
• 11:43
  And now it's like, sure, no, there's the Blackwell series
• 11:46
  from NVIDIA that yeah, just blown it away.
• 11:48
  Yeah, all those H1 hundreds, you don't want to run those.
• 11:51
  Those are power hungry and they don't give you nearly as much
• 11:53
  output. But that's, I mean, we've had 60
• 11:55
  years to prove that theorem over and over and over again, right?
• 11:59
  I mean, that's seems pretty solid.
• 12:01
  Like, I don't know a lot, but I know that whatever the Blackwell
• 12:03
  hotness is today is going to be half as good in 18 months.
• 12:07
  Because yes, because the Reubens will come out and they'll be
• 12:09
  incredible Reubens. I mean, you're yeah, that's what
• 12:11
  they call the 2026 chip, whatever it is.
• 12:14
  So you don't want to hoard chips, you want to.
• 12:16
  Use Oh, we're breaking for lunch.
• 12:17
  I was excited. For a minute, you don't want to
• 12:19
  hoard chips. No, you want to just have as
• 12:21
  many chips as you need today. I want to hoard reubens and with
• 12:24
  chips on the side too. I don't understand what's the.
• 12:27
  I'm not a huge fan of sauerkraut but like I'll tolerate it in
• 12:29
  the. Reuben, AMD has their own chips.
• 12:32
  They're right on the tail of NVIDIA Intel.
• 12:35
  Are they like the kettle style or the ruffle?
• 12:37
  What I should not use the word chips?
• 12:39
  Oh. Reubens really threw me for a
• 12:40
  loop. Let me say we should really
• 12:41
  record this after lunch. This is the answer.
• 12:44
  You and I, between the two of us, there's almost a complete
• 12:46
  adult sized brain. Almost.
• 12:48
  And we have figured out almost. This is the answer.
• 12:51
  You just need redundancy. Your redundancy should be on the
• 12:53
  things that are going to shut you off.
• 12:55
  And it doesn't mean you shouldn't try to have it
• 12:56
  locally, but you should at least have N + 1 and N + 1 can be, you
• 13:00
  know, your, your US pick whatever your US favorite is
• 13:02
  open AI or anthropic or whatever your, your, your current
• 13:05
  happiness is. And then pick one that is the
• 13:07
  antithesis of that your, your DeepSeek or some other bike
• 13:10
  dance powered interface. And then make your own.
• 13:13
  Yeah, great. You got 3 to choose.
• 13:14
  From all right, so that's the kill switch paranoia.
• 13:16
  Have we dealt with cultural homogenation?
• 13:18
  Because you think if and the everybody's sharing the same AI
• 13:21
  that their cultures are all going to become the same.
• 13:22
  I mean, I'm saying they train on data more than just the data
• 13:26
  that they're getting from the Palo Alto Starbucks.
• 13:28
  Yeah, but you've got the selection process there of who
• 13:31
  of who picked to put that in. I feel like this you start to
• 13:33
  bifurcate the problem here. If if you're talking about your
• 13:35
  infrastructure for the power grid, maybe cultural
• 13:37
  homogenization is less important here.
• 13:40
  You just need it to make sure it keeps the power on.
• 13:42
  Sure, that's probably rooted in physics more than local culture,
• 13:44
  of course, although I guess, but hold on now.
• 13:47
  Hold on. I say that.
• 13:48
  Yeah, go ahead. But also you might want like if
• 13:50
  you're if, if this is the France power infrastructure, you would
• 13:54
  need your France AI model to know that in the month of
• 13:56
  August, nobody's we're not gonna need a lot of power because
• 13:58
  everyone's in Spain. So that seems like something
• 14:01
  that you would want to know that's baked into the culture.
• 14:03
  But I think I can't wait for all the angry French e-mail we're
• 14:05
  going to get from this. By the way, it's just going.
• 14:06
  To be Oh yeah. But I would think the AI would
• 14:08
  figure that out after the first August, you know?
• 14:10
  Well no, I feel like if it was an American based AI it would
• 14:13
  just shame the the European AI people and to tell they need to
• 14:17
  work harder. There is this thing that was
• 14:19
  going around about the Super Bowl and how the AI figured out
• 14:22
  that it needed to up the water pressure during the commercial
• 14:24
  breaks so that they would be able to people would be able to
• 14:27
  flush their toilets. It's.
• 14:27
  Been doing this for years. Yes, this happened a lot long
• 14:30
  before AI existed. Yeah, didn't need AI.
• 14:33
  No, it didn't need AI. The manual.
• 14:34
  Operators all knew they needed. To do this, yeah, of course,
• 14:37
  because they've seen this pattern before in every Super
• 14:39
  Bowl previous, yeah, of course, right.
• 14:41
  I think it's interesting that they don't necessarily know,
• 14:43
  right. Like for some of the things like
• 14:44
  I, I think in in previous years before halftime, like everyone's
• 14:47
  like who's watching the halftime show, everyone's going to pee
• 14:49
  like half people watching the halftime show.
• 14:50
  I think this year you had a bigger turn out for the halftime
• 14:53
  show, so you had less of demand on the infrastructure until
• 14:56
  after the halftime show probably, which I think is
• 14:58
  mildly interesting it had. Been so hyped and made
• 15:01
  political, yeah, that it got such an enormous viewership.
• 15:04
  I'm more worried about cultural homogenization for the stuff
• 15:07
  that I would bucket as like generative AI, sure, I do think
• 15:10
  that stuff could get lost there. But don't you think that's going
• 15:13
  to come from private companies largely anyway?
• 15:15
  So it's going to wherever the private company is located.
• 15:17
  OK, so maybe the country doesn't need to own it, but I do think
• 15:21
  there's value in the country, like sponsoring.
• 15:23
  Like, we're picking on France here, but to my knowledge,
• 15:25
  France still has the media laws that require a certain portion
• 15:29
  of like radio airplay to be. In Canada, just too sure.
• 15:32
  Those and I think like, sure, if Canada didn't do that, I suspect
• 15:35
  they would no longer be speaking French in Quebec.
• 15:38
  They'd probably be speaking Mandarin.
• 15:39
  Oh, they'll never not speak French in Quebec.
• 15:41
  Yeah, I mean, that is probably true.
• 15:42
  You do not know these people if you think that's gonna change.
• 15:45
  They really enjoy no one knowing what the hell they're saying.
• 15:48
  Like they do really seem to enjoy that.
• 15:49
  I also love going to like a hockey game in Montreal and, and
• 15:52
  literally half of it is in French and I have no idea what's
• 15:54
  happening and I'm like, oh sure, what's that crochet like?
• 15:57
  Oh, that's hooking. Oh, cool.
• 15:58
  Now I know. But cultural monetization.
• 16:00
  So yes, I do think there is some value like like you were saying,
• 16:03
  like in Canada or or in France where they sponsor or even make
• 16:06
  laws to regulate some amount of local content that maintains
• 16:11
  that culture, right. Canadian.
• 16:12
  Content. Now should they should they own
• 16:13
  that AI themselves or should they just endorse it via tax
• 16:16
  credits? Or well, I don't really know the
• 16:18
  answer to that, but I feel like you want that to to still exist.
• 16:21
  That seems reasonable. And so 20% of their answers have
• 16:24
  to be wrong about French culture.
• 16:26
  Wrong. I mean, how does that work?
• 16:27
  Slow just to come back angered in Canada. 20% of the answers
• 16:31
  have to be polite. I do think there are very
• 16:32
  cultural related questions that should probably be encompassed
• 16:36
  by that AI. Like for example, if you think
• 16:38
  about what to eat, you just said what should I have for dinner
• 16:40
  tonight? Yeah, sure.
• 16:41
  Your Western AI, the one from Palo Alto is probably tell you
• 16:43
  to go have some Mexican vegan noodles or something.
• 16:46
  And then if if like if you ask the one in Montreal, I should
• 16:49
  tell you poutine because that's all they eat there.
• 16:51
  And if it's the one in France should just tell you like a loaf
• 16:53
  of bread with some butter. There's something there that
• 16:55
  would, I think, get lost the. AI can't ask what's your
• 16:58
  location and then maybe give you an appropriate recommendation.
• 17:01
  But that's the point. It's got to be that has to be
• 17:03
  ingested into it, right? That's got to be like included
• 17:05
  in there. But don't you think that's
• 17:07
  already ingested into it? Don't you think all that
• 17:09
  information is already sucked up into the worst?
• 17:11
  AI Well the problem is Big Vegan wants to stamp out all these
• 17:15
  other cities cultures. This is a huge, huge deal, yes,
• 17:19
  And everyone's talking about sovereign AI and how important
• 17:21
  it is. Every country have their own AI.
• 17:23
  And I'm going to say this is sort of like the concord
• 17:25
  fallacy. This is like every country needs
• 17:27
  to have their own airline. Yes.
• 17:29
  They don't have an national airline.
• 17:30
  They're going up in flames tonight.
• 17:32
  OK, that's never made sense. It never played out to be true.
• 17:35
  Let me make my position clear. Yeah, I do not believe every
• 17:37
  nation should have an airline. However, I do think every
• 17:40
  nation, or at least one, should have the Concorde because I
• 17:43
  desperately missed that plane and never got to fly on it.
• 17:45
  Sure, before they crashed. Sure.
• 17:46
  Yeah, we're trying. But I think you're right.
• 17:48
  I think this is like, there was what, in the 60s and 70s every
• 17:51
  country thought they needed their own national airline.
• 17:53
  And some countries still have residual airlines to this day
• 17:56
  from that moment. I think a lot of them were like,
• 17:58
  we are sinking, we need to lash ourselves onto other national
• 18:01
  airlines that are also sinking and then we can all go down
• 18:04
  together. Sure.
• 18:04
  Like, exactly KLM was Dutch and then they merged with Air France
• 18:08
  because somehow that was going to keep them more buoyant.
• 18:10
  And then they brought in some Chinese money and they brought.
• 18:12
  In some Chinese money, yeah, yeah, They're like, how how can
• 18:14
  we keep this place alive? But I think that's an
• 18:16
  interesting example where you've had consolidation in the airline
• 18:18
  industry. Like if you have passengers,
• 18:20
  United is probably going to come pick them up you don't really
• 18:23
  need. Oh for sure, because it makes
• 18:24
  money. But I think, you know,
• 18:25
  absolutely Air France is an interesting.
• 18:27
  I know we keep picking on France, but France is
• 18:29
  interesting to consider because their territory is so wide.
• 18:32
  You could almost argue that they're at a benefit of of
• 18:34
  operating those airlines at a break even or a loss just to
• 18:37
  service their territory. Yeah, like how else do you get
• 18:39
  to French Polynesia? Let me just back you up here.
• 18:41
  All airlines are operating at break even or loss. touché Sir,
• 18:45
  the line from Warren Buffett was savvy.
• 18:47
  Investors should have shot Orville and Wilbur.
• 18:49
  Down that's really funny did he actually say that?
• 18:54
  Yes, he did. Oh, that's really funny.
• 18:56
  I mean, you could charge more. The problem is it's a race to
• 18:58
  the bottoms completely. Utility completely.
• 19:00
  It's completely utilitarian. There is nothing about it that
• 19:03
  is not a commodity. It's funny you call it the
• 19:04
  Concorde fallacy because to me, we in the US were the ones
• 19:09
  trying to catch up on the Concorde fallacy, right?
• 19:10
  We were trying to build a supersonic jet around the same
• 19:13
  time as the Concorde. We let go of those plans.
• 19:15
  Yeah. The really wasn't even the
• 19:17
  European Union, right? It was really the the Brits and
• 19:19
  the French that had competing plans that then said, oh, we
• 19:21
  should probably do this together, merge them.
• 19:23
  Yeah. Which of course made sense.
• 19:25
  I think the reason that didn't work out well for them is air
• 19:27
  travel just went a different direction.
• 19:29
  It turns out you didn't want to go fast, you wanted to go cheap.
• 19:31
  We assumed supersonic transport was the way to go.
• 19:34
  It just wasn't. They just bet on the wrong
• 19:35
  horse. Is AI the wrong horse?
• 19:37
  I don't think so. I don't know what else you would
• 19:39
  bet on. Because there's so many parts of
• 19:41
  AI that are going to come from other countries anyway.
• 19:43
  I mean, we're going to go back to Waymo.
• 19:44
  There's just no way your nation is going to build your own
• 19:47
  Waymo. No, of course, unless you're a
• 19:48
  giant country. That just doesn't make any
• 19:50
  sense. But I could see a world where
• 19:52
  they say we want to own this fleet, we want to license the
• 19:55
  tech, and they could license the tech from Waymo because at scale
• 19:58
  there's too much risk. And that could.
• 20:00
  Would make sense 'cause if Waymo's not gonna be the only
• 20:01
  provider in town, right, there's gonna be multiples and someone's
• 20:04
  gonna do that deal. Stand by for Rivian.
• 20:05
  They've started putting Lidar into their standard vehicles.
• 20:08
  Yeah, I mean. Their self driving is gonna get
• 20:10
  good fast. So for sure, but like, you've
• 20:12
  got Amazon playing in this space, right?
• 20:14
  Like they've been at at this for a while.
• 20:16
  There are other players in this space.
• 20:17
  My my point is, I think someone's gonna be willing to do
• 20:20
  a deal that a sovereign entity can have some modicum of
• 20:23
  control. Over so yourself driving car is
• 20:24
  gonna be brought to you by AWS. Most likely, yeah.
• 20:27
  Like AWS is a great example of this.
• 20:29
  AWS got pushed back from a lot of governments that didn't want
• 20:32
  stuff stored on US servers. So they said OK, we'll deploy a
• 20:36
  an arm of the cloud in your geography.
• 20:38
  It's still AWS but like for Waymo for infrastructure
• 20:41
  management, I could see a world where you can offset not all but
• 20:44
  some of the risk. I want to run this on my own
• 20:46
  chips that I buy from someone else please.
• 20:48
  I have the chip using my power that I buy from somebody else
• 20:51
  please. I have that relationship with
• 20:52
  software that has been handed over to me.
• 20:54
  Maybe source code escrow. Like you could mitigate a lot of
• 20:57
  these risks with some on Prem solutions like that.
• 21:00
  And I like the idea, except for the fact that these are
• 21:01
  technologies that are evolving so quickly, so quick, that the
• 21:04
  minute you get your chips, the minute you get your power
• 21:07
  source, the minute you get your software, you're already left in
• 21:10
  the dust. Yes and no.
• 21:11
  I mean we're thinking about things here that I don't think
• 21:14
  need the high rate of change. We're talking about like
• 21:16
  infrastructure. Like if this is your water
• 21:17
  management, you probably don't need the latest model to do it.
• 21:20
  If the current model is working, then you could lock at that
• 21:23
  current model and it wouldn't be the end of the world if you
• 21:25
  didn't get the next revision in a month or in three months.
• 21:28
  But it would be much harder to to wrestle with if you lost the
• 21:30
  model you currently have. To me, that's the point.
• 21:32
  I couldn't. Want my water management, even
• 21:34
  in the US, run on AI any less? Why not?
• 21:38
  And my power management, why not?
• 21:39
  AI is just not how I want to run any of these things.
• 21:42
  I want a human making these decisions because if something
• 21:44
  happens with the AI, if I have the AI can suggest things and
• 21:47
  the human at the last minute can say, yes, we'll do that, Sure,
• 21:50
  yeah or no, we won't do that. Yeah, OK.
• 21:51
  I don't want to have any problem from an AI.
• 21:54
  Yeah, making hallucinogenic choices.
• 21:56
  What I think you want is that air gap.
• 21:58
  You want to presume that as a human, yes, there's a human that
• 22:00
  says, OK, this guy just told me to launch nuclear weapons right
• 22:03
  at at this other country. And we probably shouldn't.
• 22:06
  We like probably. Want that human air gap in a lot
• 22:08
  of things though, not just launch of nuclear weapons.
• 22:11
  Yeah, I want that in big power choices.
• 22:13
  You know we're going to power down this power plant because we
• 22:16
  have plenty of power at Blobbity Blue.
• 22:17
  No, I want a human who actually is going to think about that
• 22:20
  order before they push that button.
• 22:21
  I want to tell you that I agree with you.
• 22:23
  Yeah. However, when I look at the
• 22:25
  Waymo it really it really knocks me off that path.
• 22:28
  I know, I know. Just because you're saying with
• 22:29
  the? Whammo.
• 22:30
  It's a smaller number of people and those things are not
• 22:32
  happening synchronously. Well, you, but you are making a
• 22:34
  broad assumption there that your decision set on one one device
• 22:38
  is only going to affect a small number of people.
• 22:40
  If the whammo crashes into a fireball, it at most has only
• 22:43
  harmed or killed a small number of people.
• 22:45
  Sure, sure, if the water. If the AI water system starts
• 22:47
  putting in extra arsenic, you've killed millions of people,
• 22:51
  right? That's bad.
• 22:52
  I didn't know that was a lever on the.
• 22:54
  It's a button. It is a button.
• 22:55
  They've been trying to take it out for years.
• 22:56
  Yeah. It says extra arsenic right on
• 22:58
  it. No one pushes it.
• 22:59
  Wow. It's got a little piece of do
• 23:00
  not push little tape. Do not push.
• 23:02
  Danger. Yeah.
• 23:03
  Death. More death.
• 23:04
  No, I I I I hear you. I do.
• 23:06
  Now back it back up and just for one second and say, what if all
• 23:10
  the power plants are managed by different AIS now and, and very
• 23:14
  granular. So it's just, you know, a few
• 23:16
  blocks and we're managed by this a sure and a few blocks later
• 23:20
  they're managed by a different AI.
• 23:21
  I'm I'm much more OK with. That.
• 23:23
  OK, here's this could be interesting, man.
• 23:24
  We are just solving problems today.
• 23:27
  I don't know if it still works this way because I've seen war
• 23:30
  games and I remember what they did with the Whopper.
• 23:32
  But at least before that, the plan was you had two humans in a
• 23:36
  silo somewhere and they both had to turn to key.
• 23:39
  Yeah. And in order to launch the
• 23:40
  missiles. I'm OK with that.
• 23:41
  And the reason you had that was if one dude went Cray Cray, the
• 23:45
  he or she couldn't just go launching nukes, right?
• 23:48
  You needed two people to go Cray Cray.
• 23:49
  I think the odds of two people going Cray Cray at the same time
• 23:51
  are less, are much less. Same with pilots, right?
• 23:54
  That second pilot is there not just in case the first one has a
• 23:56
  heart attack, also in case the first one goes crazy.
• 23:59
  And there was a yes, there's a pilot in Germany, I think that
• 24:01
  flew into the side of a mountain.
• 24:02
  Or was it, was it Switzerland flew into the side of a mountain
• 24:04
  because it was the first mate who locked the captain out.
• 24:07
  Like terrible, terrible outcome. OK, Jeff, that is the answer.
• 24:10
  So here's what you need. You need your 3 redundant AIS to
• 24:13
  all agree on the decision making process.
• 24:14
  It's a quorum. Now what's going to happen is
• 24:17
  they're going to form a committee, and that committee is
• 24:18
  just going to take months to make any decision.
• 24:20
  But a month in AI my time is a millisecond so it doesn't really
• 24:23
  matter. No, but, but I really, I'm, I'm
• 24:25
  being facetious. I do think a quorum could be an
• 24:27
  interesting approach to this. If you had separate black box
• 24:29
  systems that you fed all the inputs into and they reach a
• 24:31
  quorum, it seems less likely because you wanted to attack the
• 24:35
  system, you'd have to attack all three.
• 24:37
  Yeah, as opposed to just one. And that does seem to give, you
• 24:40
  know, N + 2 would get you probably 3 orders of magnitude
• 24:43
  more security. Yeah.
• 24:43
  OK. We've talked about the the
• 24:46
  chips, yes, we've talked about the algorithms, we've talked
• 24:49
  about the data, Right. You talked about it doesn't even
• 24:51
  make sense to do. I think there's another
• 24:53
  component that I'm in some ways more concerned about than these.
• 24:57
  I I could be, could be wrong in that.
• 24:59
  Yeah. And that is the humans like the
• 25:00
  people that go into this. And I do wonder where they
• 25:04
  should sit from a sovereign perspective.
• 25:06
  The people who who create the AI, the brains, Is that what you
• 25:09
  mean? Yeah, I I'd argue from the the
• 25:11
  edge of the, like the frontier models, like creating the
• 25:13
  things. At some point that's gonna
• 25:14
  become less important to infrastructure management
• 25:16
  because, yeah, you're not gonna need the latest and greatest AI,
• 25:19
  but you still need to know how it works.
• 25:20
  Yeah, like even like you know, as you get down the spectrum of
• 25:23
  of deployment engineer and dev OPS or or like I don't want to
• 25:26
  call it it, but like it plus plus sort of roles.
• 25:29
  Sure, sure. If you don't have those, well,
• 25:31
  how do you how do you manage? I don't want to be exporting all
• 25:34
  of my best brains, my best computer brains to Djibouti
• 25:37
  because they, well, create a huge tax base and they they
• 25:40
  decide, yeah, they're gonna raise tons of money and they're
• 25:42
  gonna put more money in AI than anybody else.
• 25:43
  And suddenly all my brain goes to Djibouti.
• 25:46
  I don't want to do that for. Sure.
• 25:47
  I mean you're talking about export.
• 25:48
  I I would I would argue even more.
• 25:50
  I don't want to not be able to. I want to be able to import as
• 25:53
  many of those brains as possible.
• 25:54
  But now that's real sovereign competition there.
• 25:57
  Yeah, right. Who else is going to want to get
• 25:58
  these folks in? It probably doesn't make the
• 26:00
  world a net better place. My hunch is if you had more
• 26:03
  people focused on building the same thing together, not in one
• 26:06
  place, but in three places, that's probably better than in
• 26:08
  27 places. Sure, I would think.
• 26:10
  Sure. You still want competition.
• 26:11
  Does it make the world a better place if we all are sort of
• 26:14
  locking arms and are interdependent on each other's
• 26:16
  stuff? Oh, for sure.
• 26:17
  I think it actually forces us to get along better because for
• 26:21
  sure if we all are sharing the same information, that is the
• 26:24
  the answer isn't to be more independent, the answer is to be
• 26:27
  more. Codependent, I agree with you on
• 26:28
  this. That, to me, is the best
• 26:30
  optimistic path forward. Yeah, what is best for humanity
• 26:33
  is for us to to be as efficient as possible and remove as much
• 26:36
  duplicity as possible. I'm 100% with.
• 26:38
  You The idea that what is, that we should do what is best for
• 26:40
  the US is a horrible idea. But that's where power kills.
• 26:43
  This whole thing is People get a little taste of power and they
• 26:46
  merely try to start flexing across this.
• 26:48
  And instead of it being everyone's better off, it's
• 26:50
  just, it's just me, me, me, me, me.
• 26:52
  And I do think that is more the world that we are in now.
• 26:55
  And we get wrapped up in the just like you said, right, like
• 26:58
  America first as opposed to what we've done pretty well
• 27:00
  benefiting off a global locking of arms.
• 27:03
  Maybe we should go back to that. Few species have been more
• 27:05
  successful than dogs who do not live by survival of the fittest.
• 27:09
  Yeah, they live by survival of the friendliest.
• 27:11
  Yeah, sure. And that works.
• 27:12
  Yeah. I mean, the dog that snaps other
• 27:13
  dogs, that one doesn't last very long.
• 27:15
  Nobody invites him to parties. You always do have a few
• 27:17
  crazies, but yeah, but the mass collection of dogs, the pack,
• 27:20
  they know which ones are the only once they know, though, how
• 27:22
  to be friendly with everybody and everybody helps everybody
• 27:25
  else. That's why they listen to our
• 27:26
  show. Survival of the friendliness is
• 27:27
  a vastly superior survival strategy.
• 27:29
  Yeah, to survival of the fittest.
• 27:31
  Yeah, I mean, be nice. It's weird how that works.
• 27:33
  Strange. All right, we're out of time.
• 27:34
  We have to get out of here, but quickly before we do.
• 27:36
  Have you seen or read anything good this last week?
• 27:38
  The majority of my reading currently is books to and from
• 27:41
  my children, so I don't want to bore anybody with those,
• 27:44
  although they are they are really really.
• 27:45
  Fun. There's some excellent
• 27:46
  children's there. Are some really great there are
• 27:48
  some really great kids books out there, but no instead I want to
• 27:50
  suggest something that I that I have seen yes and this is sort
• 27:53
  of following with the pattern of my last recommendations that
• 27:56
  were around makers and builders yeah if you have not fallen down
• 27:58
  the winter gotten rabbit hole yet your YouTube feed is
• 28:02
  definitely missing out. Sure this is Martin Mullen.
• 28:04
  He created a viral masterpiece. It's this amazing gigantic 8
• 28:08
  foot tall hand cranked all wooden machine that uses
• 28:12
  something like 2000 steel marbles to play this incredible
• 28:16
  thing. They call it folk Tronica music
• 28:18
  and like the tune I actually love.
• 28:19
  I think it's great, but it looks like this thing out of
• 28:21
  steampunk. It shouldn't work.
• 28:23
  You just look at you like there's no way this thing should
• 28:25
  operate. Sure, the device itself is
• 28:27
  incredible, but the real magic is in is in his sort of journey
• 28:31
  and where he's taking it. Now he's documented the whole
• 28:34
  sort of grueling, brilliant and often hilarious process of
• 28:36
  trying to engineer mechanical perfection.
• 28:38
  He's now going for his next version, which is even and more
• 28:41
  insane. I mean, we're talking about 3D
• 28:42
  printing and trying to come up with like the best 3D printed
• 28:45
  silicone to make the bounce of the ball.
• 28:47
  Sure. Like, it's really nuts.
• 28:49
  And his attention to detail is incredible.
• 28:51
  I don't know if this is ever going to end with the building
• 28:54
  of the next machine. It does seem like it just keeps
• 28:56
  getting farther and farther out. But it's hands down the most
• 28:59
  satisfying engineering saga on the Internet that I found so
• 29:02
  far. It's an interesting mad
• 29:04
  scientist combined with he's an excellent musician.
• 29:07
  Yes, he's actually playing this thing as if it's an instrument
• 29:10
  and. It is the most common complex
• 29:11
  yet simplistic instrument you've ever seen.
• 29:14
  It has like all these levers. It's all mechanical.
• 29:16
  There's nothing in here that is electronic.
• 29:17
  There are no gates, there are no servos.
• 29:19
  Everything is, is all mechanical using gravity in a hand crank
• 29:23
  and you get to solve problems like keeping tempo and how it,
• 29:26
  you know, winds down and how you open up different gates.
• 29:28
  And it's, it's really fascinating.
• 29:30
  It's one of those things I assume everyone has seen because
• 29:32
  I think the first major video on this was like 9 or 10 years ago,
• 29:35
  but many folks haven't. And I do suggest that I'll
• 29:38
  definitely put a link in the in the show notes for folks to
• 29:40
  follow. It's really cool.
• 29:41
  All Steely's except one black marble.
• 29:42
  It's like the one to chase. It's like so you can see some
• 29:45
  movement through the whole like timeline of the machine.
• 29:47
  I guess, yeah. And.
• 29:48
  There's just one that you sort of chase through.
• 29:50
  If you liked the video I posted a couple weeks ago about the the
• 29:53
  marble clock, which is also very cool.
• 29:55
  I feel like it is the distant cousin or or maybe child of this
• 29:59
  project. It's it's really neat.
• 30:01
  Very cool. I'll post a link in the show
• 30:02
  notes for you guys. How about you?
• 30:03
  What have you seen? Marina Xenovich's new
• 30:04
  documentary I'm Chevy Chase and You're Not is a bold, warts and
• 30:08
  all. Look at a comedy legend.
• 30:10
  It tracks Chevy Chase's massive rise from the original star of
• 30:12
  SNL to 80s movie royalty while facing his reputation as one of
• 30:16
  Hollywood's most difficult personalities.
• 30:18
  The film digs deep into a painful childhood to explain the
• 30:21
  biting, defensive wit that became his trademark.
• 30:23
  Yes, family and friends like Dan Aykroyd come on camera, but the
• 30:26
  movie does not look away from the bridges Chase burned,
• 30:29
  including his messy exit from the TV show Community.
• 30:31
  At 82, he remains as blunt as ever, even telling the director
• 30:34
  she isn't bright enough to figure him out.
• 30:36
  Marine is incredibly bright. It's a fascinating, sometimes
• 30:38
  sad portrait of a complex genius who could couldn't always get
• 30:41
  out of his own way. It's not good, it's great.
• 30:45
  I'm Chevy Chase and you're not available now on HBO Max.
• 30:48
  I I absolutely will watch this. This sounds incredible.
• 30:51
  He is much more of a complex person than I think most people
• 30:53
  give him credit for. He is both a horror on the set
• 30:57
  of various TV shows and things he has done, and at the same
• 31:00
  time he is the most beloved, delightful family man.
• 31:02
  There's so many layers to Chevy Chase that this doctor reveals
• 31:05
  it's fantastic. I have to watch this.
• 31:07
  That is the episode. Thank you for joining us for all
• 31:09
  of this nonsense. It's really terrible podcast
• 31:11
  from the awful company this on the web nonsense dot
• 31:13
  productions. I'm CJ Little.
• 31:15
  I'm Jeff Parker. If you like this program, please
• 31:17
  follow down and subscribe and like an Apple Podcast, YouTube
• 31:19
  Music, Amazon Music, Antenna Pod, iHeartRadio, Spotify, My
• 31:23
  Favorite Overcast, or wherever you may get your podcasts from.
• 31:26
  Nonsense. Studios are guarded by Barkley
• 31:28
  the dog. Barkley's a rescue mutt, Free
• 31:29
  and open source. Next time you're tempted to buy
• 31:31
  an inbred dog, a purebred dog, don't get a Barkley.
• 31:34
  You'll be so glad you did. We'll be here every Tuesday and
• 31:36
  Thursday mornings. For more nonsense, please join
• 31:38
  us. What do you want?
• 31:52
  AI to do for you next. Allow me to sleep.
• 31:55
  How would it do that? I don't know lock my the door to
• 31:58
  my kids rooms? Actually come to think about I
• 31:59
  don't need an AII mean it's just duct tape.
• 32:02
  OK, nice. Watch this RIP tear.
• 32:09
  Apply and done. Night, boys, good night.
• 32:12
  Boys.