Latest / TFTC: A Bitcoin Podcast / Ten31 Timestamp: Closed Straits and Open Models

Transcript

Ten31 Timestamp: Closed Straits and Open Models · TFTC: A Bitcoin Podcast

• 0:00
  Summer in the United States, typically a time for rest, time for reflection, a time to put your phone down, head to the beach.
• 0:06
  The headlines just won't let us rest. It's never ending. We're at an inflection point, John, and no rest for the wicked this summer.
• 0:14
  I don't know. It sounds like you're thinking of summer in Europe.
• 0:17
  Hit that out of office message right after, right when June starts.
• 0:21
  Sorry, I'll be back in the middle of September. You can't outreach me until then.
• 0:26
  No, but we definitely, we have not gotten the summer doldrums here so far, that's for sure.
• 0:32
  Well, imagine what the Europeans are going to think when they wake up in September and they catch up on everything that's been going on.
• 0:37
  I'm like, oh, wow, we're really behind here.
• 0:43
  Yeah, the old inbox is going to be bursting with some bad news.
• 0:48
  Yeah, and they're going to have to filter through all the AI generated emails that have been sent their way.
• 0:54
  God bless the Europeans in September.
• 0:55
  Okay, good luck. Good luck. But we're not here to belittle the Europeans and their summer vacation lifestyle choices, I think.
• 1:04
  I think actually preferable if given the choice. But for some reason or another, we like to grind.
• 1:09
  We're here on a Monday morning grinding, going through more charts.
• 1:12
  John has curated a list of things to talk about.
• 1:15
  And we mentioned this last week. It was news to John, but the news has developed for seven days.
• 1:20
  and it has fermented a little bit.
• 1:24
  And here we are, Yemen.
• 1:25
  I mentioned breaking news last week when we last met that Yemen announced
• 1:30
  that they were going to do a naval blockade of Saudi Arabia
• 1:34
  in the Bab el-Mandeb straits, I guess.
• 1:39
  And that has happened.
• 1:41
  And as you can see from this chart that John has curated for us,
• 1:46
  This was only after this particular straight has been used as an alternative to the straight of Hermus.
• 1:54
  And so it looks like the straight of Hermus is still relatively shut down.
• 1:59
  And the alternative in the Red Sea is now a hot-button topic and a hot-button part of the world now with the Houthis getting into the mix.
• 2:11
  Yeah, Marty hit me with the old breaking, all caps breaking on the last episode.
• 2:17
  So it was news to me then.
• 2:19
  But the world's gotten a chance to adjust to it since Marty cracked the headline.
• 2:24
  But yeah, I think this was something that, you know, as soon as this kicked off, this whole conflagration earlier this year, the fear was, the thought was, you know, what are these going to do kind of on the other side of the peninsula?
• 2:39
  These are historically Iran-linked.
• 2:42
  So are they going to cause trouble as well and make shipping out of that side just as difficult?
• 2:47
  And there was noise about that over the last couple of months, but it never really took off until kind of last week when there were more direct overt attacks on Saudi Arabian ships and other assets and terminals on that side of the peninsula.
• 3:03
  as you said, kind of leading into both the Red Sea, but then also outward on the other direction.
• 3:10
  The graph kind of shows a picture's worth a thousand words, but that route had become one
• 3:16
  of the main pressure release valves for shipping oil out of the Gulf. So use the East-West pipeline
• 3:21
  to get the oil there and then offtake it to wherever it needs to go from there without
• 3:25
  being able to be disrupted by Iran that had been responsible for a massive amount of offtake,
• 3:31
  millions of barrels. And as you can see on that chart there, which I got from Reuters,
• 3:36
  vast majority of that was heading to Asia, different Asian countries, Southeast Asian
• 3:40
  countries. We'd probably infer a decent amount was to China, though it's tough to really know.
• 3:46
  But this is yet another major crimp in the oil market and potentially, especially for
• 3:51
  the eastern part of the world. I think it's questionable whether that's totally seen as
• 3:58
  downside to some folks in the Trump administration. But either way, oil prices responded quite
• 4:05
  violently last week. Although, as often happens, we're sitting here on Monday morning and kind of
• 4:11
  talking about last week's news. And we've gotten a bit of a reversal of this this morning with
• 4:16
  both sides deciding they want to give peace another shot. And so oil is down quite a bit
• 4:23
  from where it was on Friday.
• 4:24
  Yeah, not only has this had an effect on oil prices,
• 4:27
  but it has also affected insurance for the ships
• 4:30
  that are traversing these straits in the Middle East.
• 4:34
  And you have a couple headlines here,
• 4:36
  ship insurers restrict war coverage
• 4:37
  for Saudi Arabian cargoes in Red Sea
• 4:39
  and new Lloyd's Clause could invalidate tankers insurance.
• 4:44
  And this was a big topic at the beginning
• 4:46
  of the war with Iran,
• 4:50
  which is the ships were not really going through the straight
• 4:53
  because they couldn't get insurance.
• 4:55
  And it looks like that particular thread
• 4:58
  is being pulled again here.
• 5:00
  Yeah, it's, you know, you're kind of seeing
• 5:02
  that whole playbook once again played out here
• 5:05
  on the other side of Saudi Arabia with one of the biggest maritime insurers in the world arguably most important insurance entity Lloyd of London And the interesting kind of thing here
• 5:19
  is this bellwether of the insurance industry potentially pulling back to some degree
• 5:25
  out of this very volatile situation. What does that open up for U.S. insurers to come and fill
• 5:31
  in some of the gap and the vacuum. If you recall, back in April, I believe it's the Development
• 5:37
  Finance Corporation, which has had its hand in a lot of different strategic initiatives since
• 5:43
  Trump 2.0 started, including a lot that we've talked about on the rare earth side. But the DFC
• 5:49
  spun up some amount of funding to try to encourage insurers, US-linked insurers, to move into this
• 5:56
  market and kind of take some of this business. So I haven't seen an update yet on what share gains,
• 6:03
  if any, the U.S. may be able to pull off or U.S. link insurers may be able to pull off
• 6:07
  from this retrenchment if this retrenchment even even lasts. But yeah, like you said,
• 6:13
  just a noteworthy data point updating on that thread of, you know, the U.S. potentially having
• 6:19
  an angle on muscling into the global kind of shipping insurance market. And then you can see
• 6:24
  how influential and meaningful that has been already for the ability of ships to get where
• 6:31
  they need to go. And given the administration's focus on controlling naval choke points,
• 6:36
  controlling shipping choke points, you can kind of see how that puzzle piece would slot in
• 6:39
  over time, having greater U.S. influence over the financial infrastructure that facilitates this
• 6:46
  kind of stuff happening. So no real data point yet on what this might mean for U.S. insurers,
• 6:51
  But I think notable that we're kind of seeing it happen again as this event path plays out on the other side of Saudi Arabia.
• 6:59
  Well, I got to put you on the spot.
• 7:00
  What is the Dynesian perspective on this and the struggle for dominance over the monetary system?
• 7:10
  I can hear Tom Luongo just shaking in his seat, crying, City of London.
• 7:14
  But yeah, I think it would probably merit its own podcast.
• 7:22
  And frankly, I think Dines and Tom both will do a better treatment of it than either of us will.
• 7:28
  But if you don't listen to Mindprint Hash, I don't know what you're doing.
• 7:31
  You got to go listen to that.
• 7:32
  I would follow Tom Luongo as well.
• 7:35
  But basically, there's been the 10,000 foot view thesis is there are a lot of legacy old world institutions, particularly in finance and the offshore dollar system that exert a lot of control and influence over different international systems of trade.
• 7:52
  and certain institutions may or may not have a stronger, weaker hand within those.
• 7:59
  And this may represent one of those, a sub-theme within that overall control nexus,
• 8:08
  influence nexus, where the U.S. and U.S. institutions may have an opening to kind of
• 8:14
  muscle in and establish themselves a little more clearly as a key infrastructure partner and a key
• 8:19
  finance partner for facilitating a lot of this stuff. So that's 10,000 foot view that I think
• 8:24
  colors a lot of what we are tracking here and certainly what Dines and Tom have been tracking.
• 8:28
  But I think that's why at the high level, this is interesting. Yeah. Yeah. It is. Pay attention
• 8:34
  to these subplots. Anybody who's listening to this, you mentioned it earlier, but we should
• 8:39
  dive deeper into a chump. Pauses around strikes as officials weigh dwindling air defense stocks,
• 8:45
  even though he's denying the dwindling munitions by telling the Wall Street
• 8:49
  General, we have far more munitions than anyone in the world and far more than we need.
• 8:55
  Munitions. We've got all the munitions we need. Okay. And then on top of this,
• 9:01
  you're juxtaposing this headline with demand destruction in the Chinese market where diesel
• 9:07
  consumption has fallen by around 10% or it fell by 10% in May specifically.
• 9:13
  Yeah. So I have this on here because I think this is just kind of, you know, with all these episodes, I feel like there's there's kind of one juxtaposition we keep coming back to.
• 9:21
  And I think this is this is it. Right. The great power struggle between the U.S. and China, where where both sides have their strengths and weaknesses.
• 9:29
  And, you know, I don't don't mistake my highlight of Trump's quote as the main takeaway here.
• 9:34
  You know, like what else is he going to say, both because he's on the U.S. side and also because he's Trump.
• 9:39
  I think on the left side, it's like this is this is kind of the key thing, right, for anyone who's been paying attention for the last 10 years and anyone who's in kind of that Trump nexus.
• 9:49
  agree with them or not, like them or not, the key theme has been the U.S.'s potential inability to
• 9:59
  win a kinetic war, to really win an extended kinetic conflict, given our lack of,
• 10:06
  or a potential lack of, depending on whose story you believe, whether you believe the
• 10:10
  Wall Street Journal headline here or the way Trump wants to frame it. Either way,
• 10:14
  there's a potential that I think we can all agree
• 10:17
  the U industrial capacity is not what it once was you know isn arguably where it needs to be And you can see that potentially in what are being described as dwindling air defense stocks So the most advanced missiles that have been used
• 10:30
  in combating Iran in the Gulf, you know, you can see a lot of talk about basically trading,
• 10:38
  you know, systems that are worth millions of dollars for systems that are worth like tens
• 10:42
  of thousands of dollars as it relates to using like Patriot missiles or something on, you know,
• 10:46
  for defense against much cheaper to manufacture drones that Iran has been able to kind of
• 10:52
  produce potentially via the help of some other manufacturing partners around the world.
• 10:57
  But this is a real threat. It's a real issue, right? And I think if you were worried about
• 11:02
  this thing on the left, which I think reasonable to be worried about as an American, what you would
• 11:08
  see would be like exactly what Trump has been doing for the last two years. Again, rightly or
• 11:14
  wrongly executed well or not. This general theme of reshoring of domestic industrial capacity,
• 11:21
  domestic industrial policy, the things that Besson has talked about over the last few months
• 11:25
  that we've highlighted on the show, you know, this is the thing on the left is highlighted on the left
• 11:28
  really is the justification for that. And I think the rationale for that. On the flip side,
• 11:33
  on the right, you know, just highlighting, you've seen what looks like demand destruction in China
• 11:39
  somewhat significantly. And it's interesting that it's diesel specifically here, because that's much
• 11:43
  more industrially geared. And, you know, there are various reports about general industrial kind
• 11:49
  of issues across various sectors in China anyway, even before this year. And so this is, I think
• 11:55
  there is pain being inflicted on the other kind of party in this game. And, you know, we don't have
• 12:01
  the June data yet, but I would think what's happening right now with on the other side of
• 12:05
  Saudi Arabia that we just described is not going to be helpful here for the Chinese. There are some
• 12:10
  reports to be fair that the Houthis were kind of letting Chinese link tankers go through. So maybe
• 12:16
  that's, you know, backfills some of that gap. But basically, I think this is just kind of the
• 12:21
  hoop links first slide, right? Like there are two sides here, neither with a perfect hand. And,
• 12:26
  you know, they're both going to push on the leverage that they have.
• 12:29
  Well, that connects directly with the next chart, which we showed, I believe,
• 12:33
  two or three weeks ago, when you've been tracking this, basically juxtaposing
• 12:38
  the yields on U.S. Treasury bonds with the move index, which is Merrill Lynch options volatility
• 12:45
  index that is basically the VIX for U.S. Treasury bonds. And I think if I recall correctly,
• 12:51
  two or three weeks ago when we last talked about this, he said this is the chart to follow if you're
• 12:57
  trying to gauge whether or not Secretary Besant and the Trump administration is threading the
• 13:02
  needle with the war in Iran and the re-industrialization effort, can they let yields drift higher while
• 13:09
  keeping the volatility on the treasury yields relatively suppressed? And it seems like they
• 13:17
  are doing a pretty good job of that. Yes, it's on the way up, but it seems in somewhat of a
• 13:22
  managed way. Is that the right reading of this? Yeah, I mean, I think pretty much. This is like
• 13:29
  my first order taco chart that I, you know, I think that the taco meme is a little overdone.
• 13:36
  You know, if this is what chickening out looks like, then I'd hate to see what getting super
• 13:40
  aggressive looks like. But in any case, yeah, this is, you can kind of see, zoomed out here
• 13:46
  to a five-year look, pretty much been in what is essentially like a major downtrend since early 23.
• 13:52
  You can see that big spike there in orange is SVB. And you can see on the chart, you know,
• 13:57
  So SVB, Liberation Day, the start of the start of this whole thing in Iran, each time, you know, essentially hitting kind of a lower high.
• 14:05
  And right now, you know, even with this big move up and yields back to kind of the 4.7 level, which was the first since early 25, you know, move moved up.
• 14:14
  It increased, but you still kind of in this bouncing around in the sub 80 channel that it's been in for the last few months.
• 14:21
  You know, not going to do any any TA on a volatility index.
• 14:24
  But the point is just for now, this move has not triggered a significant blowout in volatility, which is kind of what I think you'd look for if you're thinking of, you know, some sort of emergency pullback from whatever's going on or some sort of, you know, new ceasefire headline that's going to get everything calm again.
• 14:45
  And I think this is like one of the first things that they're focused on just because this is your the first the first impact that you see on borrowing capacity and potential kind of reflexive self-reinforcing down cycles across a bunch of asset classes is the need to basically margin calls that can be triggered on very levered funds that are engaging in the treasury basis trade.
• 15:08
  And that, you know, you'll see you'll see that ramp up when bond volatility blows out.
• 15:14
  So I think, you know, I joked in the timestamp, like when Besson opens his Bloomberg launch pad every morning, I think the move is one of the big things that he has right at the top of his screen figuratively.
• 15:26
  So this yeah again this is kind of just what I look at every day as it relates to you know the relationship between yields and volatility Yeah And double clicking on Scott percent and bringing back what we were just discussing which is this this power struggle between superpowers in the US and China It multifaceted We covered how the war in Iran
• 15:48
  is affecting oil and natural gas markets and what that does for leverage of the individual
• 15:53
  superpowers. And the other big sort of variable in the superpower struggle is the race for AI
• 16:01
  dominance and AGI, who's going to win. And I think probably the biggest line of discussion
• 16:08
  over the last seven days has been this conversation about open source versus closed source frontier
• 16:14
  models. And I think one of the things that we want to get through in this episode is dispelling
• 16:23
  the and really getting people away from the conflation of open source, open weight models
• 16:29
  with China. And I think, rightly so, over the last seven days, a lot of people have come out and
• 16:35
  be hypercritical of the Trump administration and their posturing, particularly as it pertains to
• 16:41
  these headlines here. Basin says U.S. could sanction China over AI model theft. So a lot
• 16:46
  of discussion over distillation attacks, IP, how do we protect it? Should we protect it?
• 16:54
  What is the incentive of the frontier closed source models and anthropic and open AI?
• 16:58
  who's on what side. And I think this is pretty existential for American citizens, for individuals.
• 17:05
  The decisions that will be made around this topic, I think in the coming months,
• 17:09
  will probably set the course of history for the next century or two. It's very important that we
• 17:15
  get this right. Yeah. I mean, there's a ton there, some of which we will and can get into,
• 17:21
  and some of which we already won't. And some of which would be more probably appropriate for like
• 17:25
  a Mises Institute podcast on the, you know, the ethical validity of IP laws in the first place.
• 17:30
  But I can see Stephen Kinsella.
• 17:32
  Yeah, exactly.
• 17:36
  No, but I think for our purposes, I thought this was, it's, it's relevant in a lot of ways.
• 17:44
  And I do think also, this is, you know, our main North Star, our main kind of lens for
• 17:48
  everything is Bitcoin. And I do think having a view on what the US is going to do here,
• 17:53
  what China is going to do on this front and what either side can do and how successful they will be
• 17:58
  is relevant to kind of having a view on, you know, where Bitcoin goes in the next few years,
• 18:04
  because this is as we think about kind of strategic levers, you know, the U.S. is kind of positioning
• 18:09
  on the frontier of AI and its current dominance of compute for now is a key strategic lever that
• 18:15
  obviously, you know, the U.S. is going to want to control. And if it can't do that, then I think
• 18:22
  it's going to have implications for everything from treasury policy to Fed balance sheet. So
• 18:27
  I think it's always important to keep an eye on this. And clearly you can see from what Besant
• 18:31
  said about sanctions on China, from what the director of, I think, Michael Kratzios is the
• 18:38
  director of the Office of Science and Technology Policy, I think, from what he said and from what
• 18:43
  Undersecretary of State Jacob Helberg said, clearly like we're moving in a certain direction,
• 18:47
  The narrative is being framed that there's some there's IP theft happening in some sense.
• 18:52
  There's distillation attacks, however, it's happening.
• 18:55
  China routing around export controls of GB300s, you know, NVIDIA's most souped up rack for training these models to get something like Kimmy K3 that came out a couple of weeks ago and other open source models that they've put out.
• 19:09
  So I think you can kind of see the direction of travel here of how the narrative is being laid by regulators.
• 19:15
  But I think there's been a lot of like allergic reaction in many ways, rightly so to this, because I think there's a desire among two constituencies to this whole thing to conflate, like you said, Chinese open source and open source in general and like China's AI progress and like open source generally.
• 19:36
  and if you look at what you know best later clarified um on twitter and you can say you know
• 19:41
  his his word is worth nothing and all these guys are just constantly bluffing like okay that's fair
• 19:45
  enough but the the public perception that he's putting out kind of is what it is so like you
• 19:49
  know what is he actually what narrative is he actually trying to to lay down and it you know
• 19:54
  first sentence of this tweet is we support open source ai and the innovation it unlocks you see
• 19:59
  guys like you know jensen wong at nvidia literally creating a twitter account first post ever to to
• 20:05
  go out and highlight why open models generally, open models in general matter. And interestingly,
• 20:11
  I also thought you see this tweet from Jack Altman, who's Sam Altman's brother, so has some
• 20:17
  interesting incentives there. But highlighting ultimately what I think is the big distinction
• 20:21
  that kind of needs to be drawn for any of this discussion to make sense, which is down at the
• 20:25
  bottom of his tweet, he says, if you're thinking about geopolitics with all this kind of conversation
• 20:29
  about open versus closed models, you're thinking about US versus China, and then trying to figure
• 20:33
  out what creates the most advantage for either side.
• 20:36
  And I think that that is like the actual thing that's going on here.
• 20:39
  I don't really think that Besson and Grazios and the Trump administration
• 20:44
  generally, or major kind of AI leaders and AI CapEx beneficiaries, which we'll get to,
• 20:49
  the U.S. particularly care about or are framing the view in their heads on open source versus
• 20:56
  closed source. Are open source models, quote unquote, dangerous? Should they exist? Should we
• 21:01
  put controls on who can use them? Is code speech, is it a First Amendment thing? I think that's all
• 21:07
  completely off to the side of what they actually care about, which is limiting China's ability to
• 21:13
  mimic the frontier and probably in their view, free ride the frontier through what they're
• 21:19
  framing as distillation through what they're framing as routing around export controls on,
• 21:23
  you know, the most advanced chips. And you can have a view on whether they should care about that,
• 21:28
  whether any of that is even whether the US will actually even be able to contain any of that,
• 21:32
  or, you know, whether it's not the best thing to try to control any of that. But I think all I'm
• 21:38
  trying to get at is like, that's the conversation that really matters strategically, not what I see
• 21:42
  a lot of recently, which has been, you know, stumping for open source generally, like the
• 21:47
  power of open source models, like why open source software is important. That's all, that's all fair,
• 21:53
  but I don't think that's actually the relevant axis on which any of this is turning. And so I
• 21:58
  think none of this discussion is going to make sense or go anywhere until people can kind of
• 22:02
  separate those two things. Like what should China's access to any like American technology be
• 22:08
  slash can it actually even be limited ultimately? And then like our open source models dangerous or
• 22:14
  acceptable or should we have them? I think very clearly the answer to the latter is
• 22:18
  yes, the answer to the former is much more complicated and nuanced.
• 22:22
  That's really where the policy questions are going to come out.
• 22:26
  It's hard to tell because it's really digging
• 22:30
  into the framing that you just put forth. Focusing in on
• 22:34
  whether or not the Chinese open source
• 22:37
  frontier models have access to the GPUs.
• 22:41
  I mean, it's hard. It's very hard to tell. I mean, the DeepSeek or Moonshot CEO, which comes out, or Moonshot, the Creative Kimmy 3, which is a lot of, which is what a lot of the conversation has been around last week with their quote unquote distillation attacks.
• 22:57
  I think to your point, maybe talking past each other and focusing on whether or not distillation attacks should be, should be allowed or not.
• 23:06
  Like I personally am like, yeah, they should be allowed.
• 23:09
  And I think Jensen actually made a great point on the interview.
• 23:11
  I forget with who, but he was saying like, we're learning.
• 23:14
  Like, this is how, this is how you learn.
• 23:16
  Like you see what others are doing and you react accordingly.
• 23:20
  And so I think all is fair in love and war, particularly in the AI realm.
• 23:25
  But, but the question, I think that to bring this back to the, to the U S administration,
• 23:29
  to make them think about this thought experiment, because we know they're all listening to this
• 23:33
  podcast.
• 23:33
  But if the framing that the Moonshot CEO is putting forward around training specifically, which is that we don't have access to as many GPUs as we would like to have.
• 23:47
  So that has forced us to become extremely efficient with these machines and get more compute out of each token that we consume.
• 23:56
  and if we're running with the assumption that what he says is true and they actually don't have access
• 24:01
  to as many of these gpus by routing around export routing around export controls and getting access
• 24:08
  to them and they are actually getting more efficient with this i think that is the best
• 24:15
  way to frame it life finds a way constraints breed creativity breed innovation and you could actually
• 24:21
  find that some of this policy could be not only counterintuitive but counterproductive for our
• 24:28
  end goals if you do if you do basically put on expert control export controls and try to limit
• 24:35
  the the access to gpus that that these open source frontier models in china have you could find it
• 24:42
  creating like a constraint framework that makes them extremely efficient and has them finding
• 24:47
  innovations that we're not focused on here in the US because we have a gluttony.
• 24:51
  We have the luxury of gluttony of compute.
• 24:54
  And I think that one thing to think about is incentives Like you just incentivizing these people to become more efficient and eventually surpass the the closed source U S frontier models Uh and that what I would worry
• 25:07
  about. I mean, Jensen has made that point on a number of podcasts and public appearances over
• 25:13
  the last year or so now on the incentives point, you could flip, flip that and say like, well,
• 25:17
  yeah, he's talking his own book. He wants, you know, as much diversity, uh, he wants a diverse
• 25:21
  customer base and he wants to be able to like to sell into China, you know, China sales from
• 25:26
  video right now are basically zero. Like, so he would, he's completely incentivized to make the
• 25:30
  argument that, you know, they'll just find other more efficient ways to, you know, around, around
• 25:34
  it if they, if they can't get access to the best accelerators. But I do think it's a, it's a
• 25:38
  legitimate point. And I think you're seeing, you know, at least some evidence that that is true in
• 25:42
  practice. But I, I, I think this, this side that we have up right now, which is the open secure AI
• 25:48
  Alliance participants, which was just announced this morning by NVIDIA and a variety of partners,
• 25:53
  speaks to kind of what the whole, like the, I think the right framing of what's going on with
• 25:59
  this conversation is. And it gets to, I alluded to it earlier and I didn't say it explicitly,
• 26:04
  but I said, like, I think there are two constituencies that really want the China
• 26:09
  question and the open source question generally to be linked. And I think one of them is China.
• 26:14
  Like, I think it's a, you know, I think it's great for Xi Jinping and CCP if the entire conversation here just becomes a referendum on open source, because then it puts them in the great position of being like, well, hey, man, code is speech.
• 26:28
  Don't you care about free speech?
• 26:29
  Like, this is just, you know, all we care about is advancing human technology, right?
• 26:33
  Even if that happens to, you know, undermine some meaningful kind of US strategic assets, right?
• 26:38
  So there's that piece.
• 26:40
  But interestingly, I also think the other constituency that would love this conversation to become one thing are the two major logos that are not present on the slide, right?
• 26:49
  So Anthropic and OpenAI.
• 26:51
  The main frontier labs and those that are on their cap table would love for open source generally to become a boogeyman that's mostly associated with a geopolitical rival.
• 27:03
  And I think that's a great regulatory capture angle for them to pursue.
• 27:07
  And we don't have to jump to it yet, but I think on the next slide, you see some evidence of that already happening.
• 27:13
  But, you know, just if you're watching on video, just like take a look at the logos that are on this page.
• 27:18
  Like look at who, you know, benefits from and wants open source AI generally, right?
• 27:24
  It's infrastructure providers.
• 27:26
  It's chip provider.
• 27:28
  You know, it's NVIDIA providing the accelerators.
• 27:30
  It's, you know, Synopsys and Cadence providing tooling around chip development.
• 27:33
  It's Cisco, Cloudflare, CrowdStrike, IBM, HP, Dell.
• 27:39
  Like it's basically all of the companies in the ecosystem that would benefit from there not being like a massive profit pool sitting at like two companies that are gating frontier intelligence, right?
• 27:50
  If you had a diversity of a bunch of different architectures and models needing to be served in a diverse set of ways across like a diverse set of use cases.
• 27:59
  And you didn't have this like oligopsony of basically two buyers, right, for all of these different companies that are sitting there gating the entire industry.
• 28:09
  Like that's completely net good for a huge amount of American companies to be able to benefit from.
• 28:14
  And I think you can kind of this slide, I think, shows you like how the how the conversation is actually going to shape up and what the axis of, you know, alliance here is actually going to be as it relates to kind of open versus closed away from the question of China.
• 28:31
  Yeah, and you alluded to it, and I think it's really important to bring this up because I think this hugging face hack triggering AI kill switch bill in Congress, this hugging face open AI story is, to me, the most important story of the month because it highlights just the absurdity of this whole conversation and the regulatory posture here in the United States around these frontier models.
• 28:57
  And so I guess I'll, I'm sure many of you are aware, like the U.S. is very afraid or
• 29:04
  the administrator, or the U.S. government, I won't say the administration, I think the
• 29:07
  administration is trying to figure out what the signal is But I think the fear mongering that has emanated from this Hugging Face hack has led to politicians who really don understand what they talking about
• 29:20
  speed running regulation around this. And so basically what happened with Hugging Face is
• 29:26
  about a week ago, a week and a half ago now at this point, they recognized that their systems
• 29:31
  had been infiltrated by an autonomous agentic system. At the time, they did not know what it
• 29:38
  was. And they basically sounded the alarm bells internally, sprinted to figure out what was going
• 29:44
  on, diagnose the problem and patch it. And they have enterprise, anthropic and open AI accounts,
• 29:51
  and they tried to use the frontier models in Fable 5 and ChatGPT 5.6 to diagnose the problem.
• 29:58
  But the guardrails that were put in by the administration about a month ago, which said
• 30:05
  you cannot use these frontier models for code security or model training and a bunch of other
• 30:11
  topics. I think, I think biology research and physics research as well. And so they were trying
• 30:18
  to diagnose what was going on within their systems. And the frontier models here in the U.S. said,
• 30:23
  no, you can't, you actually can't do that work. We're not, we're not allowing you to because of
• 30:26
  the guardrails that have been put in. You're being delegated to less performant, lower IQ
• 30:33
  models to try to fix this and they couldn't. So Hugging Face was forced to spin up servers
• 30:38
  internally and download an open weight GLM 5.2 model, which was produced in China to actually
• 30:44
  diagnose and patch the problem. They were able to do that successfully with the self-hosted
• 30:49
  open weight model GLM 5.2. And it just highlights the absurdity of this. Like you have an American
• 30:57
  company trying to fix a problem they have of an autonomous agent getting into their system and
• 31:04
  wreaking havoc. And they weren't able to do that with the U.S. closed source frontier models because
• 31:10
  of the guardrails that had put in. They were forced to download and self-host an open-weight
• 31:16
  Chinese model. And then a few days later, we come to find that what actually happened was that
• 31:20
  OpenAI is training their latest frontier model, and it broke out of the sandbox that it was
• 31:27
  put in during the training run and it infiltrated Hugging Face's systems. And so we're expected
• 31:34
  to react from a regulatory perspective to this event where you have a closed source frontier
• 31:39
  model here in the US breaking its sandbox and attacking Hugging Face, which funnily enough is
• 31:46
  one of the sort of like repositories for the best open source, open weight models in the world.
• 31:52
  and we're supposed to view that as dangerous and create a regulatory framework that gives
• 32:00
  actors like open AI and anthropic sort of regulatory moats to make sure we nerf the AI
• 32:07
  future and none of this happens. So you have basically the perpetrator of the attack getting
• 32:12
  protection if we go down this route. I think it's completely absurd that this conversation is going
• 32:18
  in this direction in certain circles. I think a lot of what we saw in the headlines and the reaction
• 32:23
  last week was basically combating this direction that many people want to go in. But I think this
• 32:30
  is what it all comes down to. And then I'll just end with saying, yes, these open source models,
• 32:35
  these models generally, even if they're frontier closed source models or open weight models from
• 32:40
  China, they are going to surface security bugs. And that may seem scary, but the converse of that
• 32:48
  is you can use them as defensive technologies too
• 32:52
  and try to discover these vulnerabilities
• 32:55
  using these tools and patch them
• 32:57
  before you're attacked as well.
• 32:59
  And so by trying to nerf the world
• 33:02
  and prevent people from getting access
• 33:04
  to the best models,
• 33:06
  whether they're open weights or closed source,
• 33:10
  I don't think is really like trying to prevent people
• 33:12
  from getting access to that.
• 33:13
  It's actually going to be a detriment
• 33:15
  to the ability of U.S. businesses
• 33:18
  to protect their systems and increase their security Yeah no I mean it an incredibly it an incredibly ironic situation but it it classic you know I said in the newsletter but candle makers petition
• 33:31
  right. From Bastia, like petitioning the government, the candle makers petitioning
• 33:35
  the government for protections against the unfair competition of the sun. Right. We, we, it's a tale
• 33:41
  as old as time. And unfortunately it's moving forward here. You know, I know we're we try to
• 33:47
  keep these to a tight 30 minutes. So moving forward briefly, I threw these on here just
• 33:52
  because I think it's all ultimately like relevant framing for this, this entire meta conversation of
• 33:56
  getting to like, I think the strongest anti China advocate for with this, this whole open versus
• 34:04
  closed source, and export control kind of conversation would say, like, well, all that
• 34:09
  matters is getting to RSI. So recursive self improvement for models, like he who gets their
• 34:14
  first, you know, wins the whole future and has, you know, an unbreakable lead on intelligence
• 34:18
  for, you know, industrial intelligence for the next thousand years. I think, you know, a lot of
• 34:24
  the guys in SF who are very AGI pilled will basically make this exact argument. And, you know,
• 34:29
  let's even say that that's true. I think if you just look at the way that the White House is
• 34:33
  positioning, you know, it's increased funding for AI, you know, moving funding away from legacy
• 34:39
  collegiate grants and toward initiatives like building out AI infrastructure and kind of this
• 34:47
  bottom headline, FERC, the regulator that monitors basically all electrical grids in the U.S., kind
• 34:53
  of threatening the biggest grid operator in the country, PGM, with maybe being broken up if they
• 35:00
  don't move faster on going through their AIQ. And clearly this is, again, viewed as like a key
• 35:05
  strategic priority at a national level. And I, you know, whether or not open AI and Anthropic as
• 35:13
  companies, you know, have a long term defensible business model, or whether they get to have,
• 35:19
  you know, profit pools or not, is I think very secondary to getting to in the view of people
• 35:25
  who are really trying to push this forward, getting to that RSI point and winning this this
• 35:29
  race for better or worse. And I don't really think it requires open AI and Anthropic to exist as
• 35:34
  purely private entities to get there. And so I just think worth keeping that meta theme in mind
• 35:40
  with all this, like if you're telling the world, like I'm effectively building, you know, the most
• 35:46
  important machine since Manhattan Project, and it's as dangerous as, you know, nuclear weapons,
• 35:52
  and it's going to radically reshape society. And it's the most important thing for every country
• 35:57
  to focus on. You know, don't be surprised if and when the most powerful government in the world
• 36:02
  says, okay, that's effectively a public good, right? And the frontier becomes, you know, a true
• 36:07
  like US asset, rather than the asset of, you know, one or two private companies, not necessarily
• 36:12
  calling it, that's going to happen. But, you know, increasingly feels like you're moving more and
• 36:17
  more in that direction. And particularly when you see a slide like the one we pulled up with all the
• 36:21
  logos, a couple slides back with basically all the most important and powerful American tech
• 36:25
  companies positioning themselves kind of as advocates for a more open and openly available
• 36:32
  AI infrastructure broadly. Yeah, don't be surprised. Be aware. Sorry for the echo there
• 36:40
  coming in at the end. But last slide, always got to end it on Bitcoin. Your good friend,
• 36:45
  Frank here has a very good chart. The supply of Bitcoin that is being held by long-term holders
• 36:51
  is at 84%, which is the highest ever. Pretty bullish. Yeah, speaks for itself, I think,
• 36:58
  useful and interesting to keep in mind against the backdrop of all this, again, just because
• 37:02
  it's your classic kind of bear market bottom formation. If you look back over, you know,
• 37:06
  the history of this chart, it's exactly the kind of thing that you see as you're, you know,
• 37:11
  entering a true bottoming phase, not calling a bottom, no idea exactly when that happens. But
• 37:16
  at the very least, it's, you know, yet another data point kind of pointing in the same direction
• 37:21
  of all the prior cycles that we've seen in the past.
• 37:24
  Yeah, so be aware.
• 37:25
  We'll see you guys next week.