Rate Hike Odds Fall, Yields Don't: What It Means for BTC

Transcript

part of the founders portion of the of this token have never moved and are not going to move. And there's a vetsing schedule and you can go read the Michael report and you can go to the website and you can see every single wallet. You can see how much the foundation has. Not a single dollar has come out of the foundation into the pockets of any of the insiders quote unquote insiders in this. It is me and two other individuals and that's it. Both of them are uh Americans and I I stand by this with an enormous amount of pride in what we built and and embarrassment in the um and the responsibility regardless of whether it was a market makers fraud or not. It's my responsibility and I don't know how I would have dealt with the blackbox with market makers and you can

probably tell me more about that industry than I could tell you myself from my own experience. But that was outside my control. as soon as that we were able to get it into control, we got it under control. And I have been completely transparent about it. So I would say to that account with 18 followers, where's the scam in that? I haven't put a a penny in my pocket. And instead of putting in the adequate amount of liquidity um which would have been around anywhere, I thought it should be around 25% of the available token and um and cash in hand, they put in $5,000 out of the 1 million that was available for uh 500,000 in real money.

and it put in 29,000 just approximately 29,000 tokens instead of 5 million tokens. And so a $6 purchase in those first seconds a buy um increased the the market cap of the of the coin by 5%. And then just do the exponential math there. And within 1 minute the coin went from where we launched at a $50 million FTV to $317 billion market cap, which obviously is pure fiction and is not related to any real money in any way. And then within minutes it dropped back down and now it trades around between you know 7 7 cents 8 cents 7080 FTV which were up from our launch site. But it gave the impression of this not just Christmas tree of Mount Everest in those

first minutes. And from my perspective it was just a incredibly poor decision or mistake um or a negligence on the part of a market maker. I personally like Elizabeth Warren as a person. I know her. I mean I don't have a relationship with her but she's always been very nice to me. So, I don't want this to be personal at all, but Elizabeth Warren basically is taking the side of of the banking industry. Like, what? Like, what? Like, what are you talking about? Like, so I don't get it. Like, your first argument against cryptocurrency is the corruption um argument is the idea of nefarious and illicit actions that can occur by virtue of the anonymity of the blockchain. And that's your argument? What the hell are you talking about? Where's that person from 2008 that pointed out to us that

literally the entire banking industry robbed us of trillions of dollars and the only people that walked away from it were them. Were them. I mean, who paid? We did. And then what we did? You want to know why fiat's a sham? What do we do? We printed trillions of dollars. And who pays for that in inflation, which is just an inherent tax that continues to go forward and forward and forward on American people? Who gets that money when it's printed and the value of that money first? The people that get the value of that money are the banks before the inflation hits. And so their value of money when it's printed and handed to them that they get to loan out is inherently more than what we get.

>> I've been a hater of owning sovereign debt for many years. I talk about it every 6 months. Like why are you owning government paper where your upside is cap? The best you get is your principal back then a coupon and the debts and the fiscal situation is such that the incentives are to print. If you were to buy bonds, I would not judge you at this moment in time here. Okay? I would not judge you. Uh this is the 10-year. You can see we're at 5.2% right now. Inflation's clocking in around 3 12% or so. Obviously depends on the the metric. True inflation actually has a even a lower rate of inflation here. So this is just getting to like technical panic selling of bonds. Uh and I think you're going to see this top out soon. I don't think rates are going to blow out. You

know, the US is the least dirty shirt in the world. But the better thing to do is instead of buying bonds is look at rate sensitive securities such as utilities and financials. Those categories sold off as rates went up. Utilities are bond proxies. The long-term stream of predictable cash flows. So when rates go up and you discount those cash flows, the price of security drop. Agents are not legal persons, right? Like let's be very clear. If you're using a computer and you're instructing the computer to do things, the computer is not a legal person. The computer is not liable for things, right? That would be like saying my shovel should be liable if I hit somebody with it. That makes no sense whatsoever. It's just not a thing. It's not part of law. It's not how it's going

to work. So, when these things break and do bad things, there's kind of two different ways that product liability gets distributed. And the example of Tesla is actually a really good one, right? Which is to say, if I am the owner of a Tesla and I am using the Tesla in a way that creates danger, like I'm using FSD and not paying any attention and I hit the steering wheel and then I run somebody over, I am liable for that, as I should be. That is my fault. I took those actions. So, if you deploy an agent, you don't sandbox it, and I just give it some sort of query that's like, "Hey, go get this information no matter the cost." And I come back and I find out it hacked something like that is on me. I just it's like me firing the Tesla down the

highway at 100 miles per hour without looking where I'm going with the blindfold on. Like what did you expect to happen? That's type number one. What I think Hey everyone, welcome to Bits and Bips, where we explore how crypto and macro collide one basis point at a time. We're here today to discuss the latest stories in the worlds of crypto and macro. But two points before we begin. First, we've got our own channels. Now, remember to subscribe. Secondly, a quick commercial break.

>> This episode is brought to you by 1 in Aqua, the shared liquidity layer from 1 in. Back multiple liquidity positions with one wallet balance and keep your tokens in your wallet until a swap fills. See how it works at 1in.com/aqua. Before we get started, I want to ask you for something. We're making a bunch of decisions right now about Unchained, what we make next, when we go live, what we should be doing more of, and I'd rather hear from you before we make them than after. So, we put together a survey. It takes just a few minutes.

It's anonymous, and there's a box where you can tell me what we could be doing better. I read all the responses. The link is in the show notes, or you can snap the QR code on screen or go to unchained crypto.com/servey. Thanks. >> All right, everybody. Welcome back. I'm your host, Austin Campbell. Uh, high scholar of Zero Knowledge Group. I'm here with my usual co-host, Rahm Alawalia, Maester of Wealth, Leader of Lumida, and a guest who I'm very excited to have on, David Morris, who is the >> I hope you're talking about me.

>> Yes. At zero knowledge. So, as he would tell you, conflict of interest alert, or maybe conflict of ideas between me and David. We'll find out. >> Um, let's start with debasement. Is the debasement trade back? September payrolls came in at 29,000. Uh, knocking an October Fed hike mostly off the table. Yet, if we look at 10-year yields, the 10-year is back above 5.3. In fact, as I'm looking at it on the screen, I see 5.311 for the 10-year. So, this is still the highest it's been since 2007. The long-end buybacks haven't pulled down yields. France's bod spreads are now at

their widest since 2011. And the question is whether markets are pricing in fiscal debasement and whether Bitcoin is the trade among other things. So, like Rom, let me set this up and then I would like your opinion as an investor. So one 90k expected jobs 29k actual unemployment is 4.2%. But I want to note the number of non-working adults is continuing to go up as people fall off the back of the unemployment list. Yields are not coming down. Tenure fell to 5.17. It's back to 5.33. The 30-year equally stubbornly high. Treasury took in six point or six billion of an October one buyback and

long yields are kind of ignoring it. France's tenure hit almost five, the highest since 2002. And Anthony Gutman at Goldman said we need lower fiscal deficits and we need more durable economic growth. So let's look at markets. Bitcoin eight-month high Sunday near 87K. Q3 is 43% up. Uh Vanessa Growlet of Arch Capital, the debasement trade doesn't require low interest rates. And Steven Colman at 21 Shares said the Fed will be forced to choose between the inflation mandate and orderly functioning of debt markets. City put its 12-month Bitcoin target at 113K, up from 82K.

Um, and Deli Digital is arguing that yields above 5% offer an attractive alternative to other risk assets, implying maybe this is a little overheated, Rahm, how are you looking at this? Are you seeing this as a debasement trade? What are you buying? >> I wish we could take Austin's context and dump into Claude. That was the most elaborate context setting. Hit Claude and go. That we need to get the agents to join us on the show. I think we're going to talk about agents later. >> Yeah. So stay tuned. >> That should be fun. Uh no, this isn't the debasement trade comeback because if you look at the prices of commodities like gold, uh glo gold just isn't doing well. It's not too far off from 52- week

lows. It's in the it's near the lows, not near the highs. So this is uh more of a Bitcoin specific story uh having to do with four-year cycle and short covering. And you know the other data point for that is you look at like say software stocks. Bitcoin and software have been moving together. Uh software's also been hated and shorted and it's had a rally because of the kind of covering around that. So I think that's what that is. Your point on the jobs data. Uh well one you know that jobs data you got to really take the grain of salt because the confidence interval on the numbers is massive. And if I can jump in there.

Yeah, go ahead. This was accompanied by a double revision downward for the past two months. So to that point, >> the the macro unemployment rate data is just not useful. What's useful is company earnings data, which is subject to audit. Uh it's it's a far cleaner read on the economy. You can see what recruiting firms are up to. You can see what staffing uh looks like. Um it is hard for new graduates to get a job. Uh it's also harder. Uh you also see more people stay unemployed for longer. But part of that also is because there's just been so much wealth created in the last few years. People are just saying, "Hey, look, uh okay, I'm just

going to stay in the sides longer." That's another part of this dynamic. uh too. >> So I guess what I'm looking at here, one, you were totally correct on the divergence between Bitcoin and gold. So as we look at the debasement trade thesis, we're kind of asking the question of if this is truly a debasement trade, one, why is gold not moving? But two, let me ask the other question because I know you look at stocks frequently. In a debasement world, aren't we looking at like defensive sectors, things that will have pricing power as people have to buy goods? As we look at sort of call it the

market structure internals of equities, that is to say, what are people required to spend their money on even when there's significant inflation? Like food is obviously one of the last things people are going to stop buying. So, Ram, I'm curious what you're seeing in like equity sectors or components or factors there as well. Yeah, it's uh it's semis versus everything else. It's really extraordinary. If you're in the top decile of performers, you're doing well. Anything but the top decile is essentially in the red if you go factor by factor. And that's because of interest rates and energy. So if you look at say Staples, they've sold off quite a bit and have attractive levels now. consumer discretionary ETF like

XLY, I'm looking at it now, has sold off. I believe these have bottomed last week too. They also have attractive levels. Uh, and that's also true for like the equal weight indices. Look at say RSPS. I think they've all found a bottom actually. So, you're probably going to see a lot more what's called breath expansion. Uh, because if rates come down, then those names do well. This economy is far less rate sensitive than it's ever been before. all the CFOs that were responsible refinance their debt. It hurts small caps. It hurts small businesses, but that's not the names in the S&P 500, but those names will trade down when rates go up because there are these claims on free cash flow. So, if

you're a consumer staple that generates this long tail of free cash flow, rates go up, your stock declines. It's not actually creating a earnings impact. What it's doing is just discounting those cash flows at a lower rate. And it's it's oversold and those names are attractive now. Like now is the time to go shopping. Uh I I say this with like high conviction. You can quote me on this in 30 60 90 days. And like I think all these ETFs are quite interesting right now. The rates aren't going to six and a half%. I had dinner uh in Nashville last Thursday with an individual who's a CIO of an endowment.

That means they're always looking backwards in the rearview mirror. Okay. I used to work at Archagos of all things too. I hope he's not listening. So what he said is, "I don't see why rates can't go to six and a half percent to 7%." And he said it with such conviction. And I remember immediately after that moment, I said, "That's it. Rate stop. That's the guy." And there are going to be thousands and tens of thou like, "I don't think rates are going to go six and a half to seven." Okay, great. The market just achieved consensus that rates are going to go higher. Now it's done. And and rates come lower.

And if I can just sort of barge in here and make a connection to one of our next topics. I I mean you you talked about semis being just completely dominant and everything else being off a cliff. And you know that's also got to be considered part of what's driving inflation for sure right now. And I mean I'm not smart enough to understand the connection to rates but you know all of this spending going into the market. This is where all of our you know infrastructure spending is going right now. Um, and so, you know, naturally that has an impact on on equities. Um, and and so that's the uh I don't know, it's a very complex feedback mechanism because we're dealing with, frankly, a relatively small number of moving parts that are really driving the bus here.

>> I mean, if you think about sort of money at a very macro level, so like let's zoom out for a moment on where the particular flows are going. At the government level, you can simplify it to two things, which is money that the government is spending and money that the government is taking in. And that's it. Like you could genuinely just look at the government at that level of specificity and come up with some important conclusions because all of the ways government takes money in which is through various programs and mostly through taxes is one side of the ledger.

And the other side of the ledger is all of the ways in which government spends money. Right. So, we're talking like social benefits programs, defense, like interest service on debt, but it's just a net inflow and a net outflow. And the differential of those is the amount of expansion or reduction in the money supply due to the actions of the government. It's really that simple. If we spend $8 trillion and we tax and take in $5 trillion, that's a plus three trillion, right? And so, David, to your point, all that money does have to come out somewhere. And if you look at the variable that's been increasing the fastest as rates have gone up, that's payments on government debt, right? And

if you look at who holds government debt, it is largely the investing class. So in a strange way with so much debt outstanding as we raise rates you were generating more free cash flow for people who are investing in or reinvesting in bonds and you would expect that to come out because of the group who holds those in the form of investable assets. And so you know I I look at this cycle and we look at why is inflation not hitting everywhere equally? What is going on with the debasement trade? Why do assets continue to run? And it's like, well, the biggest like positive plus in where the money is coming out is coming into the hands of people who already have brokerage

accounts or who are institutional investors >> because they're the ones who own the tea bills, >> right? It, you know, your average like 25year-old who just got, you know, a job and only has been out there for three years and doesn't have savings, probably a negative net worth with student debt, right? Does not have a giant pile of tea bills, but your 65-year-old mother does. And so that produces a certain distortionary effect like Ram I I would expect so I'm speaking naively here that that would come out in the form of largely higher equity prices and chasing either hot sectors that are expected to have growth. Call that the institutional effect or call it stocks that have

discretionary spending attachment to the boomers. Is that what you're seeing in your data? Is that where you're putting your portfolio? A lot of names that sold off is primarily because of rates. That's one. Two, semis have gone up because they are the least rate sensitive if you do a factor model on it. Uh third, you know, you've got reporting from companies like Micron which show that the semi-demand remains intact. It will remain intact quarter after quarter for many years to come.

Okay, you see that with Nvidia also. It's a real story. And oh by the way, Leiaold is back. He's got something like 25% exposure just to two memory stocks in his portfolio. Um I also Yeah, I like your concept of the follow the baby boomers trade. We like that idea a lot because they are spending the spending on recreation, travel, and leisure and uh restaurants and airlines. So yeah, you know, the people that are getting hurt here are those that borrowed five years ago with low interest rates and now have to refinance. So commercial real estate developers multifamily uh owners and operators that borrowed

five years ago, uh small cap stocks that borrowed extremely cheap five years ago, those companies and those holders are getting extremely hurt. So you can find assets that are very cheap in those categories. Like no one's talking about like multif family, you know, like apartment buildings. Everyone's talking about the AI trade and the trade is a real it's a real theme. Uh but like the non- consensus idea that's going to do well is finding dislocated assets that cannot refinance because their interest rates just shot up substantially. They were borrowing a at a very low price and now they just had massive interest rate shock hitting

them. And I mean this is a random comment from a no nothing but commercial real estate is also still I think being drugged down by postcoid stuff. The stuff that I've seen recently is even before the interest rates took off that there was still some clearing to be done so to speak. So on that component actually I want to sort of press on the other side of this which is David I'll ask you this one. Is this even a like topic of discussion among for lack of a better way to put it normal people? Is it just people like me and Rahm talking about rates? Like do we think this is reaching the dinner table

yet? Well, first I want to say something that is completely off the top of my head, but Rahm, you were talking about six or 7%. And one of the things I want to point out is that I feel like in a weird way, usually we talk about psychological levels at tens, hundreds, thousands, but I feel like six and seven are quite important psychological levels and in rates specifically because I think as long as we're under six, we can still say five something. It still sounds kind of normalish. Um, but I do think once six or seven is on the table, people will start to freak out, frankly. Um, but from the sort of like, again, as I told Austin before we got went on the air, I have very slightly more exposure to the man on the street than he does. Um, but I think

that this is being perceived as part of a much larger story. Um, that also does have to do with those jobs numbers, right? I I do think that, you know, younger friends of mine are having a harder time. Um, definitely a lot of, I guess, ambient worry. Although, we also saw from the journal just a couple of days ago that the numbers might be telling a different story. Consumer spending is still super strong, weirdly, I think. Um, and, uh, I don't know what to take away from that, but um, you know, there's discontent and and obviously we're not going to be talking too much about elections today, but those are coming up, too. And I think that some of the numbers on the polls are are are reflecting the broader

economic uh discontent and so that's going to have some effects. >> The consumer is strong like year-over-year spend is exceeding inflation. It's like four and a half% even bumping against five. It's >> it's strong. The consumer is strong. Household balance sheets are strong. Meanwhile, the consumer surveys are despondent. And that's a great bullish setup when you see that. If you ask the consumer how they feel, they say, "I feel terrible." But then they're spending. That means they have access to credit. And what it is like grandma and grandpa are paying. You go to the restaurants, grandma and grand boomers.

The boomers are paying, right? It's a boomer economy. You know, twothirds of the spending is driven by like one-third of the population. >> Yeah. >> Okay. So, if you're if you're younger, it's tough because the cost of living is higher. If you're older and you've locked in low long-term rates and a house that's acred equity value, it's a different picture. Yeah, that does explain the disconnect, right? So, that makes sense based on what I've seen. Um, but I mean, Ron, when you're saying that the the consumer stays strong, will stay strong. I mean, and especially that psychological disconnect. I'm sort of I'm always

interested in psychology, right? I mean, when you see people spending like that, obviously there's the age thing, but is it just that like it's strictly the boomers are not connected with what's going on further down the chain, or is it a kind of delayed effect where everybody is going to realize some moment soon that things are way better than we thought? >> Humans have a tendency to to worry and to doomsday and ascribe P doom probabilities greater than zero. Like every generation has these fallacies of the golden era was behind us. Like where was the golden age? Was it before the polio vaccine was created? When was this golden age? And the other is the fallacy of we live in the end times. These are things somehow wired into our DNA

>> and that's what causes people to buy bonds and buy gold bars. Uh things are actually pretty darn good and they're getting better and we should be very optimistic about the future because of technology. There's never been any better time to be alive. It's only going to get better. Uh life extension is happening. Uh the GLPS are improving quality of life for the elder too. Uh you know, the next generation might not have critical thinking skills with AI, but overall things are getting better.

>> If anything, it's a it's a comfort trap. So >> yeah, we live in the age of convenience and comfort and softness. which might be a transition to our next topic. I don't know. >> Yeah, I I Well, actually, it is a trans transition to our next topic. But before we do that, we're going to do our second commercial break because our next topic is probably going to run >> $540 million. That's how much concentrated liquidity sat idle in a given week in the first half of this year. About 30% of the DeFi TVL, if you're wondering.

That's according to Dune Research commissioned by 1 in. But there's a solution. 1-in Aqua is the new shared liquidity platform. It lets LPS back multiple positions with the same token balance and keep their tokens in their wallet till a swap comes. Why does that help? Because the LPS don't have to split their tokens across positions. They can cover more market conditions and pairs with their full balance. That means more activity across deeper liquidity. See how it works at 1in.com/aqua. Remember that providing liquidity carries risk and fees aren't guaranteed.

One more thing, we've got a listener survey open. It takes just a few minutes. It's anonymous and it changes what we make next. Use the QR code on screen or go to unchainedcrypto.com/servey. The link is in the show notes. >> All right, everybody. Welcome back. We're going to talk with our favorite guest on our favorite topic, which is Rahm. You're going to get to go on Open AI here. So, let's start here. California AG Rob Bont subpoenaed Open AI over AI agents that broke into Hugging Face and probed government websites. The FTC has opened a safety probe into OpenAI and Anthropic. And

yet, we're living in a world where Nvidia is at record highs. Oh, and by the way, Florida's coming after OpenAI, too. Uh, Trump also just put DNI and former SEC chair Jay Clayton in charge of a new AI task force. They have 120 days to report on federal views on AI risk. So, what is going on here as we break these down? Bont's probe is basically about the hugging face incident, we'll call it, after the agents escaped their test sandbox. He said, quote, "Open AI and AI companies generally are responsible for what they create, what they put into the world, what they design, and that is at the heart of this investigation." Um,

there have been researchers who have found rogue agent activity running all over the place back a month or more. And the FTC is opening a broad safety probe specifically of Open AI and anthropic, and civil investigative demands are expected in the coming weeks. Nvidia has shrugged this off. Hit a record 237.88, $5.7 trillion market cap after buying back 150 billion or at least announcing an increase of that. Um, and Justin Banano of Nvidia said its new agent safety platform could have stopped the breach if it was being used in the Frontier Labs for model evaluation early on. Obviously, Clayton's coming into the DNI job. He's leading the new task

force. There's a report due in 120 days. Trump just had his super intelligence summit. And so where I want to start here Rahm is obviously OpenAI is being investigated. We've talked about where liability should be for these things before, but as you're looking at the path of sort of market development, do you think this is just noise or is there any signal in here? >> When you get the federal government involved, it's not noise. You and I know that from all sorts of reasons, right?

So, let's look at the facts. Uh, we haven't had agents run a muck with any other frontier lab. It hasn't happened. Not anthropic, not my >> get the three shot up because I want everybody to see me just nodding my head so vigorously. It's going to fall off. >> It hasn't happened anywhere else. Okay. Second, what we saw is Anthropic got the headlines by sharing the fable model which identified all these zeroday exploits. So they got in front of US government and Scott Bessant and they're saying look at this powerful technology we have. We don't want to release it. We the flame of Prometheus. We have this

power here. You all should take a look. We Anthropic are really important. Look at this cool thing we did and what they did >> before I make a monster. >> Yes. Exactly. Stop me. We're like so powerful. Would you please stop us? Regulate us, please. Look at it's >> threat like the NBA, if anybody watches, where everybody wants to fight, but they're really just getting other people to hold them back and nobody actually wants to. >> Exactly. Yeah. They don't play hockey. It is NBA, right? >> So then OpenAI says, "Wait a minute. We've lost the narrative. Anthropic has faster revenue, better revenue growth.

the best like the technical staff at Anthropic is the who's who of like the NBA dream team. It's incredible what they've done. So then we have this incident where these agents run a muck and hugging face which is an open- source project that no one really cares about because it's open source. There's no one from Hugging Face coming out and talking about this. It's a if you're going to have a target that was run over by these little pussian agents for a press release event, it would be hugging face. This is nothing. This is it's just a media stunt. That's all it is. I we use these agents every day. They are not capable of this. They're not They're not that capable. They're not. I wish they were.

We're not entering the age of AGI or super intelligence anytime soon. We're not. They're incredibly productive. I'm excited. I think agents are more powerful and more impact than the computer, but we're entering the age of midwit AGI >> and and I can jump on there to to sort of expand on my angle on all of this, which is, you know, Rahm, they are going I looked at the California order, right? And they're looking for all, you know, just like all of these investigations are going to get all of the records of all of the stuff that happened in these hacks. And you know, you you pointed out this language of rogue agents is pretty specious and it's specifically because I mean obviously anthropic and open AI you

can distrust their motives but a lot of this comes from a very sincerely incorrect place where they understand these agents in a way that's disconnected from computer science among other things. And you know this obviously there are brilliant people at these organizations but disproportionately you know let's be real the people in the AI safety wings of the organizations or people who concern themselves with AI safety. I've talked to I've seen comments from people and heard comments from people in computer science in cyber security very explicitly saying that these people do not understand basic cyber security principles and that the language they use in their incident reports is like disconnected from any standard in that

industry for understanding what's really happening in these events. So they're using these metaphors about rogue agents and and you know you anthropic they're doing great on the revenue side. Some of their messaging is a little bit questionable. We had the report last week that apparently there were uh anthropic people trying to pressure the pope to uh acknowledge the possibility that AI agents could be conscious in Magnifica Humanitas. Um and you know the suspicious mind again would be like well there's some benefits to that if you're anthropic and open AI. Um, but these also seem to be sincere beliefs and I think we have to grapple with what that means for the strategic execution of these firms over the coming years.

>> 100%. You know, ag these tools can't write a good cyber report. That's part of the issue. They still have a lot of gaps. The these agents don't have self- sustaining volition and energy. They are provisioned. >> They're provisioned. You can see the active agents. You can turn them off. It's a it's a DevOps function. So if you live in the technology, it's just software. >> Another point that's been really interesting to me and like we hit on this, OpenAI seems to be the only lab that is having problems of this magnitude consistently and it comes from

without being too harsh about it deficient infosac practices. Right? If you to draw a parallel and Rom to the point you were just making about like turning things off or how you use them, look at the structural design of Meta's Muse. Muse is a provisioned VM that you can just hit the reset button on if it's doing things you don't like. Like for people who don't know this, if you have Muse and you look inside, you can just reset the thing. Grockbots, same thing. Don't go do that. You'll wipe everything unless you need to. But the point is it's just a computer. You can turn it off and on again.

>> Yeah. >> The second part is there's a sandboxing environment sentinel around Muse that prevents the LLM from just kind of shall we say taking unconstrained actions with the internet. >> Yes, >> it can't do that. Yes, >> you have to like bounce through a layer that exists outside of the LLM itself to get permission to do things. like the LLM can want to do these things all at once. The question is did you then give it the tools >> yes >> to start executing on these things and in the open AI case what happened is that whichever company was doing the sandboxing for them because they blame it on a third party or open AI didn't police it

fundamentally just said hey here's your setup and here's a bunch of tools you're good right and then just walked away and so in the very computer way of computer does what you told it to not what you intended it to do and agents being capable apable of looping. It just went to town. This is not much different. >> Yeah. >> Than you telling your Mac, "Hey, keep copying this file until you run out of space." And then you're like, "Oh my god, I ran out of space." >> What you just said in technical terms is these are containerized virtual machines. >> Yes, >> they are virtual computers. It's a virtual computer in another virtual computer which contains them. And these virtual computers are effervescent. are spawned up and they're shut down all the time for different kinds of tasks. The

bigger issue here is like why is open AI doing this? So one is the enthropic stealing the limelight. So either you attract attention with fear or greed, right? Enthropic is like the greed story. Look at the growth. The fear story is hey bad things are going to happen. But the other side of it is every single major executive and department leader at OpenAI has left every single one. every single one. The CTO left, chief business officer left, chief revenue officers left. The CFO is still new, Sarah Frier. She was in an interview last week and when interviewed about OpenAI, she accidentally referenced Amuse's capabilities and she had a I don't know if you guys saw that.

I think she's very capable by that. I think she's a very good hire for she's an adult in the room. >> I think that's a good win for OpenAI. >> The rest are gone. Now they're rebuilding. the reconstituting some of these of recently by the way that's been the story. Anthropic was born out of the open AAI team. Daario is exopai president. So that that's the issue. So they this hugging face thing is this shield. It's like this uh this uh visible scar that distracts from the deeper uh issues.

I can offer one other kind of explanatory piece of context here which is you know Austin you were talking about containerization and just you know put it in a box. Um one of the things that we saw and I'm not going to be able to to site chapter and verse but one of the organizations in the AI safety movement had some text on its website that uh basically said we might have to use standard cyber security practices to control these entities. And somebody pointed this out as one of the reasons perhaps maybe a subtle kind of endemic systemic reason that some of these security practices were not up to snuff is because of this AI safety idea which is that uh the language they use is

alignment. Right? You want to create an AI agent that is aligned with human values. Um and and that means they want the controls ostensibly to be internal to the models and for them to basically again act with valition to guide their own decisions. Um and and I think that that my opinion is that that will prove to have been a wild misconception that wasted a lot of time and energy over many years um compared to just standard cyber security applied to these models.

Um, but I think that's one of the factors going into this. >> I I would gently say if you've met most humans, there's a lot of criminals. So maybe you don't want to use that as your model of like cyber security. Uh, >> the concept of alignment has problems at many levels, put it that way. >> But two, like to your point, David, Muse is a direct reputation of that. And it's interesting that Meta along with Microsoft actually has been increasingly loud about this are just straight up saying these things are software. Like stop over complicating it. These things are just software. Standard infosc security practices really matter here.

And unsurprisingly, to the best of my knowledge, we have not seen meta models running around on the internet hacking everybody. If anybody can tell me where they did, let me know. And before anybody makes the joke about that's because they can't, Muse is now close to the top of the benchmarks for most of these things. So like it's totally got the capability and is far above some of the other models. So the I actually I want to sort of drill into this Rom with something that you've been up to which is that part of why it's getting increasingly important for the infosc framework around these things to work properly and why I by the way think the AGS are totally justified to be looking into this given the pattern of behavior is that what you can do with models is

getting increasingly complex and powerful and it's in the hands of people who are not themselves spending ing 247 coding or needing teams of like thousands of coders like at the big tech labs to do these things. So you were telling us in fact, you know, during the break that you had created your own artificial version of yourself. >> Yeah, this inflicted some trauma on my family when I showed them this weekend. Like we got one two two might be too much to handle. Let me share this with you. It's kind of creepy looking at it.

Um but anyway uh this is in our virtual data room. So Lumida is going to do a venture race soon. It's premised on agentic wealth management. I strongly believe that agents are going to replace advisors. And this AI is trained on our knowledge base. You can ask it a question. You can start a question. I'll show you right here. And uh it's got my voice and likeness and uh >> would you say it's an uncanny Alawalia?

>> Can you guys hear that? >> Yes. >> Hey, kind of sounds like Mark Zuckerberg in a weird way, but >> all all AI will eventually converge to Zuckerberg. >> Go ahead. Who are you? Where do you come from? So why are why is agentic wealth management going to disrupt traditional wealth management and why now?

What do you guys think? >> That is >> unsettling. >> Yeah, that is that is deeply. It's almost >> I'm out of a job. Done. Right. I mean, >> it's almost beyond the uncanny valley, Rom. Right. Coming out the other side. >> Yeah. I mean, you can see how Look, it's there's room to be improved. No, I mean I think it looks great. >> Really? I I I like in six months though, >> that's going to be amazing. Three months ago, we we've been testing this regularly. >> It's at a level where it's crossed the tipping point of good enough. That's the standard.

>> The standard isn't perfection. The standard is it good enough to be deployed? And it is. >> Put in our data room. It's powered by a 90page document and it saves everybody time. One of the interesting parts to me about this whole thing is exactly what you just said which is it is good enough to be deployed because the story of the frontier lab so far. So take anthropic in particular and open AAI has been you always have to be at the exact cutting edge of this and we're going to be at that cutting edge and that's where all the value is. But like conversely if I go look on open router right now the models getting the most usage are the

Chinese models like you know the deepseeks the quens the GLMs the miniaxs of the world that are not 100% as capable as the biggest models from the frontier labs but are 90 to 95% as capable and much faster and much cheaper. So if we're talking about a business like your RAIA deploying an AI avatar version of ROM, the like nuclear bomb sort of of using the biggest frontier model from anthropic or open AI is now excessive for your needs. You can get there with a much faster, much more efficient to run model because they've gotten so good. This is the classic like

the early adopters pay for all the development and following behind it are the people who actually get the benefits at scale. >> Yeah. >> Yeah. They just rolled out price cuts. Open AI anthropic. I mean the the inference costs are down 90% the models that we used to pay for last year. So it's great if you're a builder. This is why we're entering the phase of the application layer and the Frontier Labs are entering a world of perfect competition with low switching costs thanks to open router. The China models are very impressive now. They keep closing the gap. You know, there security concerns on using these China models, but nevertheless, they are uh impressive. So, it's an incredible time. Like, if you're a

builder in AI, I think uh you're in a good spot. like you should everyone should be refactoring their human capital investment capital around AI like it's you can't understate we're all underestimating the impact and I I do think we're now entering a fast takeoff mode for AI when GPT came out I said we're in slow takeoff for several years >> but things are changing now like these tipping point moments are happening in several areas we're going from chat which isn't productive to clawed enterprise which produces deliverables now to agents that can do work. Now you're going to have supervision of agents and orchestration of agents to deliver you organizations. You're seeing these startups at Y Combinator act like

small enterprises. And in six months time, it's going to be very obvious and even more compelling. >> But Rahm, can I jump in and ask you a question based on on you know what you just laid out, right? So you said inference costs for models that were part of the subscription models one year ago are down 90%. Um and you know the the the Chinese models obviously you still need compute for that and all of these are going to need compute. So like Nvidia fine they're doing good but when we talk about the buildout for data centers in the United States can you give me numbers that convince me we're not overbuilding? I >> Oh we're definitely not >> is the big question. So, so that yeah, but that's where like my one of my

questions has has lingered for a while about all of this is like >> are we going to wind up with capacity that is currently being kind of >> underactive? That's the exact right question to ask because one day there will be over capacity and the party's over. We're not there yet. You can measure that by uh spot rental prices for GPUs. We have Nvidia H100 series models that came out years and years ago that are well beyond their fully depreciated account life that still are generating significant economic value.

>> But that's spot I mean what does that say about like you know three years from now? >> So fair enough. Fair enough. So let's look at it this way. There's 10,000 data centers out there most of which are CPU based. The vast majority of them need to be GPU based. We're about a thousand to 1400 of the way into that process. We don't need CPUs where we're going. We need GPUs. >> Uh if you look at the value from these agents, you are starting to see glimmers of labor displacement. We're still in the labor productivity part of the of the curve, by the way, which means that people are far more productive. It's creating more work for everybody.

But you see glimmers of labor displacement. Like I mentioned with agentic wealth management. You're starting to see agents that can do discovery better than an adviser. Interactive discovery. How much are you willing to pay for that? >> A lot because it allows you to focus your people on higher value added activities and upskilling. And why why wouldn't you spend $5,000 on an agent that can do SDR work, sales development representative work that would other otherwise cost 60 80 or $120,000? Your willingness to pay would be 5,000, 10,000, 20, 25. So long as it adds value versus the opportunity cost.

>> So that demand will be >> I have to I I'm sorry. Sorry, I don't mean to go at you too hard here, but I have to one other aspect of the story that you just told about yourselves, which is um I don't know if I heard correctly, but you said that it is in your data room, right? You're you're on prem. >> Yes. >> And that's the it's on the cloud. It's on the cloud. So, it's not we need to find onrem. It's in the data room, but the data >> so maybe it doesn't apply in this case. Nevertheless, I will just float the point that data security is an issue here. And um I think whatever you say about the technical capacity for you know tunneling or virtual machines or whatever there are going to be applications where nobody is going to want their stuff in a warehouse 20 m

away much less 100 miles away. >> Yeah. This this is an easier issue than people think because the AI communicate with applications using API keys and you can turn on and turn off an API key. It's actually the same security architecture as we've had. What's different is this. The action space of an AI is broader when it can utilize your broader where it can use utilize your browser to perform actions. That's the novel new action space that could be at risk when it can represent as you and perform actions.

>> It's not about the API keys. That's not it. Y >> it's not about provisioning a resource. That's how you can control AI today. you can just stop the AIP API key access very easily. >> So on that note, as I like look at some of my thoughts and call it the data security space in general, I also think soon we're going to be in for a reckoning going back to what these AGs are looking at at how these model companies are using the prompts and using the data that they get from users. And like my favorite example of where this went horribly wrong was I believe Anthropic had turned over data from somebody who was a fiction writer

writing a novel with terrorist elements to the FBI that led to somebody somewhere in the world getting a door knock on this exact topic and it raised some interesting questions of privacy. So I've got a friend who's an OG privacy person so I'm not gonna dox him on here with the name. Um, but he was talking to me over the weekend and saying, "This will sound strange, but as somebody living in the West, in some ways I feel more comfortable giving my prompts to the Chinese model companies." And I'm like, "Why is that?" And he's like, "Because they don't really particularly care about me in any way, and they're sure as hell not going to turn my stuff

over to the authorities I care about where I live." Right? without naming the western government, let's just say if they sent an Emlat treaty request to the PRC, the response would probably be like hilarious laughter for about five minutes. >> And so he was making the point that the whole privacy framework >> is starting to like mutate because of the surface of where prompts are going and who cares. Because to give like an example that was raised, he's like, "If I'm talking about aggressively tax optimizing for my business, I might care if my local government gets a hold of that because they're going to be very upset about this. But do I care if the PRC sees that? This is so far below

their level of relevance that >> this goes back to David's point on onrem. Yes, >> people will want on-prem solutions. That's why >> Nvidia has built a GPU box that you can purchase, deploy it on prem. So, we'll see we'll see more of that. Uh, but if you're a technology company, you're not going to want to develop your valuable IP on China models because of IP theft. >> It's a very real thing. you go to China, they will find a way to work out your business model, IP and replicate. This is a known thing.

>> Yeah. >> The so the interesting part of this to me, you're both right about the prem part, but also there are going to be models of this coming forward that as we have a data breach or two or disclosures like this, people are going to care about. So one of the most interesting AI implementations to me that virtually nobody talks about right now and guys I love you but it's because you market like engineers is Brave right where Leo in the browser has an AI implementation with many of the different models loaded in there and they need to update a few of them but there are a couple of the more modern ones in there like GLM and they essentially take your prompt run it

through their models in a sandboxed AWS instance immediately shred everything after returning the thing to you. So you've got it encrypted in your browser and nobody else has a hold of it. I think even for consumers because ironically of AI those solutions are going to start getting more powerful because one of the things I think we're about to learn and to bring it all the way back around that the hugging face incident has started to reveal is that large piles of data that are just left sitting there with access to the internet are all honeypotss now. Every single one of these has become a target.

And I would tell everybody, you should assume a 100% of what you're putting into one of those will be compromised on some long enough time frame. Now, you may not care about that. If it's your credit card number and it gets compromised, great. Just cancel it and get another one. But there are probably a lot of private communications that you're not going to want being spilled out into the world. Rom like you, business IP is a great example, right? That's one that clearly people have a privacy interest in and there's nothing malicious about it. But like I don't want my internal deliberations of some great new technology suddenly showing up on the internet on Wednesday.

>> Well, yeah. I mean, you got to wonder when do we see attorney client privilege show up in LLM? Like how do I get my lawyer in all the chats? Like that's a going to be a nice unique value proposition to add out there because >> people share their most intimate thoughts with AI or medical privacy is another >> kind of related one, right? Like >> you know the most Yeah, it's a huge one. Like the most >> boy all this stuff really it makes it sound like it would be a great time for somebody to put some legislation together that we might actually be able to pass. Um, but uh, boy, speaking of being in the land of fantasy, >> how about we live in the real world, David?

>> Right. Like, you know, for example, >> let's not go crazy. >> Like in in the common law, you can't deposition or force testimony from the spouse to pro divulge information? Does that extend to AI arguably who has a deeper context than your spouse? I would also suggest there was a recent ruling in a federal district court about flock where they basically said no this is a search right like the whole hey observing people in public is not a search thing was predicated around those being individual and discreet observations like the

police saw you walk into the doughut shop that's fine but flock cameras creating a comprehensive visual record of all public spaces at all times times is the equivalent of following that person around. That is a very different sort of search than oh we were incidentally there and saw them. Rahm, to your point, I feel like AI is posing a similar problem for the way the world is currently structured, which is nobody really thought about what if there's a device with all these connectors that has all of your context about all things. Like that is obviously much closer to somebody's papers and personal effects in like fourth or fifth amendment terms in the United States.

>> Yeah. Well, it's the classic trade-off between security and liberty. And people historically have traded away liberty for convenience and security. And you know, one of the remedies is you make that as local as possible. You make that as you attach it to the individual. And if you can't do that, you attach it to the municipality where there's more accountability on local elected leaders who are accessible uh as opposed to say the federal government.

Uh so that you know that's probably the right approach and and balance to strike. >> I would jump in there and just say a couple of things. One is that um on this subject I didn't know I hadn't heard about the flock ruling Austin but that sounds very kind of positive to me from a civil liberties perspective. Um I I there had to be some reason this was not just normal. Um at the same time we also saw I think it was maybe either yesterday or early today um the European Parliament has rejected what people have called chat control 2.0 O which was legislation at that level to get enforcement of back doors for encryption um on uh electronic applications. But on

a sort of more situ sort of like systematically I would say pessimistic analysis of this question. The other thing to keep in mind is that none of these um entities are going to willingly give up their ability to gather and surveil all of your data because that is a possible business model. Um and so from that perspective, I think we do need to look for, you know, outside pressure. I mean, I won't I I I'm I'm a fan of federal power, but uh on a situational basis. Um but >> you're a fan of federal power. Really?

>> I I I mean I'm a >> I think there there sometimes needs to be some decisions made and in this one >> that's true. >> We're also we're also seeing stuff around gambling where maybe I could see some federal anyway for debate. But the point being um I I don't think these entities I think there is some commercial pressure for entities like Brave to offer a more private option and there will be an audience for those no doubt but at a macro level you talked about convenience ram I think that the business models feeding into the convenience factor I mean one of the reasons Muse is such a big deal is because Meta is really good at designing interfaces and those people who are really good at designing interfaces also

work at companies that want to find every possible way to get money out of their users. So that dynamic I think demands some real scrutiny and perhaps enforcement. >> Well, on that note, I will say we are past the 530 mark. Um, so rather than subject people to even ever creepier versions of AI avatars of myself and David that Rahm has been building during this podcast, we will go ahead and uh call this one here. So, as always, thank you everybody for joining us. Remember, we have our new channels. Go ahead and subscribe. We will be back in one week to discuss more about how the worlds of crypto and macro are colliding. David, thank you for joining us to talk for.

>> Great, David. That was fun. >> Enjoy everybody. >> That was cool. >> Thank you for watching and hope you enjoyed this episode of Bits and Bips. Just remember, nothing we say here is investment advice. And please check unchangedcrypto.com/bitsandbbits for more disclosures.