What We’re Reading (Week Ending 03 November 2024) - 03 Nov 2024
Reading helps us learn about the world and it is a really important aspect of investing. The late Charlie Munger even went so far as to say that “I don’t think you can get to be a really good investor over a broad range without doing a massive amount of reading.” We (the co-founders of Compounder Fund) read widely across a range of topics, including investing, business, technology, and the world in general. We want to regularly share the best articles we’ve come across recently. Here they are (for the week ending 03 November 2024):
1. We Need to Have a Talk About “Bond Vigilantes” – Cullen Roche
In May the 10 year interest rate was as high as 4.75%. By September it was as low as 3.6%. Today it’s bounced back to 4.2%. And as rates tick higher in recent months there’s been a growing chorus about how “bond vigilantes” are going to teach the Fed a lesson. This has been especially loud coming from Stanley Druckenmiller, Paul Tudor Jones and Elon Musk. The basic narrative is that bond markets will teach the Fed and US government a lesson about reckless spending which will drive up interest rates and bankrupt the USA. Except there’s a huge problem in this narrative – the bond market has a lot less control over interest rates than this story would have you believe…
…Individuals go bankrupt. Large aggregated sectors do not. For example, the aggregated private sector cannot go bankrupt. The US government is a huge aggregated sector in the US economy. It does not go bankrupt. In the aggregate it can print money to fund all spending and it cannot run out of this money unless it borrows in a foreign currency, which it doesn’t do. Of course, it can cause wildly huge inflation. We know that after Covid, but comparing the Federal Government to an individual is just wrong…
…The Fed is the reserve monopolist. So, as I’ve explained in the past, they have absolute control over something like short-term interest rates. The market quite literally cannot force them to change this rate because the market cannot compete with the Fed. If a bank tries to set the short-term interest rate the Fed just comes in and smashes them with their bottomless pit of money…
…The best way to understand the yield curve is to think of the Fed as a dog walker who has absolute control of the leash at the handle and allows the dog to wander further out the leash. The Fed has absolute control over the handle (the Fed Funds Rate) and lets the 30 year wander from side to side, but still within a certain control. It might look like the dog is walking the Fed, but the Fed always has the ability to pull that leash in and grab that dog by the neck. This is what “yield curve control” would look like and if the Fed entered the market for 30 year bonds and started explicitly setting a target they could drive that rate to whatever they wanted. But they let it float a bit…
…I wrote nearly this exact same article 11 years ago in response to a WSJ op-ed in which Druckenmiller said the exact same thing. He said the USA was bankrupt and that the Fed was too loose…
… In my humble opinion, the error in this analysis is two-fold:
- Assuming that interest payments are problematic – they are not because the Fed can control them with a dial and also because high rates put DOWNWARD pressure on inflation.
- Assuming high inflation must result from government deficits. I think it’s absolutely true that large deficits put upward pressure on inflation. I’ve said this a billion times during Covid. But government spending is 23% of GDP. That’s the same level it was at in 1982! And while it’s a large portion of aggregate spending we should remember that 77% of spending is coming from OTHER sources. In most cases, it’s much more efficient sources such as the most efficient corporate machine the human race has ever seen (corporate America).
2. You’re Not Paranoid. The Market Is Out to Get You – Jason Zweig
Graham wasn’t only one of the best investors of all time; he may have been the wisest. His intellectual brilliance, six decades of investing and study of history gave him a profound understanding of human nature.
As he wrote: “The investor’s chief problem—and even his worst enemy—is likely to be himself.”…
…To be an intelligent investor doesn’t require a stratospheric IQ. It does require discipline and the ability to think for yourself.
As Graham pointed out, individual investors are “scarcely ever” forced to sell stocks or funds and—unlike professional portfolio managers who are continually measured against the market—are never compelled to care what other investors are doing.
That independence is your single most valuable asset, a luxury most professional investors can only dream of possessing. It’s what Graham called the “basic advantage” of the intelligent investor. But, he warned, “the investor who permits himself to be stampeded [by other people’s behavior]…is perversely transforming his basic advantage into a basic disadvantage.”
As I argue in the new edition of the book, it has never been harder to be a disciplined and independent investor. In today’s incessantly twitchy, infinitely networked markets, the siren song of smartphones, social media and streaming video can tempt you to trade more and copy the crowd.
After all, it often makes sense—and just feels right—to join the herd…
…Yet crowds aren’t always right, and their errors are contagious. What separates the wisdom from the madness of the crowd?
In 1907, the statistician Francis Galton described a contest at an agricultural fair in which nearly 800 visitors tried to guess the weight of an ox. Although many knew little or nothing about oxen and their guesses varied widely, their average estimate turned out to match the weight of the ox exactly.
Galton’s guessers had a variety of viewpoints, sought to win a prize for accuracy, didn’t know other people’s estimates and had to pay an entry fee. The sponsors of the contest collected and tallied all the guesses.
The judgments of that crowd were independent, confidential, diverse, incentivized and aggregated—and, therefore, remarkably accurate at estimating simple values.
But the judgments of today’s crowds are often the opposite of Galton’s…
…The weight of an ox doesn’t change with people’s estimates of it. However, if thousands of speculators decide a stock or cryptocurrency is worth $100,000, it will skyrocket—at least temporarily—even if it’s worthless.
Joining the crowd can change how you think, no matter how much you pride yourself on your independence. That’s especially insidious because it occurs subconsciously.
One recent study found that investors on social media are five times more likely to follow users who agree with them and will see nearly three times more messages they agree with than disagree with. Falling into such an echo chamber, the study showed, leads people to trade more—and earn lower returns.
Meanwhile, bucking the consensus engages circuits in the brain that generate pain and disgust. Experiments have shown that when you find out your peers disagree with you, your choices become up to three times more likely to match theirs, although you have no conscious awareness of being influenced…
…If you have views about which asset or investing strategy is right for you, write down your reasoning before you explore what some online group is saying. Take no action without reviewing your original rationale and determining that there’s a reasonable basis for changing it—grounded in independently verifiable evidence, not just the opinions of random people online.
Use a checklist to focus on the stability of the underlying business rather than share-price movements. Have I read the company’s financial reports? Do its executives admit mistakes, use conservative accounting and avoid hype? Have I written down at least three reasons why this is a good business that will be even better five years from now? What, exactly, do I understand about this company that most other investors are missing, and why?
3. Why the Fed Cut Rates and Mortgage Rates Jumped – Joe Weisenthal, Tracy Alloway, and Tom Graff
Joe (05:45):
But I actually have to refinance a mortgage in a couple of years. I could do it today, I guess, but I have to do it at some point. Alright. Government 30-year yields are 4.3%, 4.32% as we’re talking right now. I’ll probably want to get a 30-year fixed. Why can’t I just borrow at 4.32% if the government is already backstopping it?
Tom (06:05):
Well, so the key difference between a mortgage bond and a Treasury bond is that, in the United States, virtually all mortgages and all the ones that Fannie Mae and Freddie Mac back can be refinanced at any time without any penalty.
Joe (06:18):
Can’t I just promise not to? Well, I guess because I can always sell the house or something like that.
Tom (06:20):
Yeah. You can’t do that, Joe. And so, from an investor perspective, what that means is, if interest rates rise, no one refinances, everyone just stays where they are. Witness all the people kind of stuck in 2.5%, 3% mortgages right now, right? And so, those mortgages just stay outstanding and they might stay outstanding for 30 years for all we know, right?
Whereas if interest rates fall, you kind of don’t get any of the benefits. So if I buy a 30-year Treasury and interest rates drop, I can make 10%, 15%, 20% price appreciation as that happens. But in a mortgage bond, if interest rates fall, everyone just refinances, I just get all my money back at par, I’m no better off. And so you got to get paid for that, what — we’ll get into it —but what’s called negative convexity, you’ve got to get paid for that risk and that’s why there’s a spread between mortgage bonds and Treasury bonds…
…Tracy (09:53):
So talk to us about what goes into producing a mortgage rate. So if I want to buy a house and I go to a bank and I ask for a mortgage, what are the individual factors that go into the number that eventually gets quoted back to me?
Tom (10:07):
Okay. Yeah. So let’s assume for sake of argument, this is a loan that conforms to Fannie and Freddie’s standards. because That’s the ones we’re talking about here. So assuming that right? Your bank has to pay Fannie or Freddie a guarantee fee. The G-Fee. And that is based on your credit situation. So how much you’re putting down, what your credit score is, that sort of thing. And it’s all algorithmic. So they’re just typing it into a computer, Fannie and Freddie kicking back, here’s the rate, right?
Then they’re also going to think to themselves, okay, well where can I sell this mortgage? Right? What price am I going to get when I sell it in the open market? And that depends mostly on just what the general price is for the going rate for mortgages, but it might depend a little on your situation so we can get into it how certain kinds of mortgages command a bit more of a premium in the market than others and that will go into the rate you’re going to get quoted. And so every night the bank’s mortgage desk is sort of plugging in, hey, for mortgage like this, we’ll we’ll offer this rate for mortgage like that we’ll offer this rate and all these factors are going into that. So when your loan officer’s typing this into his computer, that’s what’s spitting out, right?
Joe (11:17):
Actually, let’s back up. What makes a mortgage conforming versus non-conforming?
Tom (11:21):
The biggest thing is the price. So the price relative to used to be a hard number, but now Fannie and Freddie do it relative to your sort of MSA or what, what your area? Yeah.
Joe (11:31):
So wait, above a certain price? Can you go into that a little further? Above a certain price, Fannie and Freddie just won’t back?
Tom (11:37):
Yeah. They’re just not backing it. And that has to do with their mandate from Congress to be about affordable housing…
…Tracy (25:56):
So I’m going to ask the question that I’m sure is on everyone’s minds per that Google trends chart, but when do mortgages come down?
Joe (26:04):
Yeah, or what will it take at least?
Tom (26:06):
Yeah, so, so we should, let’s, let’s talk about why they’ve risen since that Fed meeting, and then I think that’ll inform where they’re headed, right. So look, the 10-year Treasury is not a function of where the Fed is today. It’s a function of where people anticipate the Fed being in the next year, two, three. Beyond three, it’s sort of fuzzy, but like, you know, year two we sort of have a sense we can make a guess.
And so going into that September meeting, people started thinking themselves, boy Fed may cut 50 basis points in September, 50 basis points in November, maybe even 50 more basis points in December, right? If you pull up your WIRP chart on the Terminal, you can see this, right? If you go back to then, but since then what happened, we got a big jobs report the beginning of October. That was the September report, but came out in October.
And that was kind of a game changer because not only did we get a solid number for September, but it was huge upward revisions kind of erased what looked like a downward trend in hiring, right? Well now all of a sudden we’re like, boy, the Fed might be a lot closer to that neutral rate than we think, right? Eh, probably going to still cut in November, but maybe they’ll cut in December, maybe they won’t. But if they do, it’s certainly not going to be 50 basis points unless something changes.
And so that change in expectations has caused the tenure to rise. So commensurately the mortgage rate has risen, right? And so from that story you can say, all right, well it becomes pretty easy to see what’s going to cause mortgage rates to drop, the tenure needs to drop, right? And what’s going to cause the tenure to drop? Well, we’re going to need more Fed cuts priced in. Well what’s going to cause more Fed cuts to get priced in? We need the economy to get weaker.
4. An Interview with Hugo Barra About Orion and Meta’s AR Strategy – Ben Thompson and Hugo Barra
HB: Yeah, and this is worth taking a step back and talking about in a bit of detail, because there’s a few things that we don’t think about too much. The thing that annoys me having worked on smartphones for the last 15 plus years is that our smartphones make us work too hard. These workflows, these mobile app workflows are too repetitive, they’re not smart, they treat us generically.
It doesn’t make any sense that this world will continue for a lot longer and we know that AI is going to fundamentally change this. All apps are going to become agentic. Think of developers writing apps in their respective agents. Agents will make it possible to have much, much simpler workflows, which are highly, highly personalized. They’re still being rendered by the app, but the flows themselves are highly personalized, they have a much lower burden on users. Agents can do a lot of the prediction and anticipation and pre-thinking on a user’s behalf so that everything is boiled down to hopefully a small number of simple choices or no choices at all.
Oh, here you go, I have an analogy, you have to tell me if this fits what you’re going for here. So arguably the ultimate agentic experience that people experience right now, even though they don’t realize it, are their social media feeds, in that the feed is perfectly customized to serve up to you the entertainment that it thinks you want at every moment, and it actually turns out based on engagement numbers, it works pretty well, and while people claim they want a chronological timeline or whatever, that’s like saying you want a grid of apps and the reality is revealed preferences says that no, they don’t want that. Is that a good analogy for what you’re going for?
HB: I think that’s halfway there. I would say an agentic version of Instagram is going to be a little bit different. Instagram thinks it’s pretty smart, but it doesn’t have a lot of context from your life. As much as people say that Instagram listens to their conversations, that’s not true.
If only it did.
HB: Exactly. If only it did, it’d be great, but it of course doesn’t. So Instagram, to use an example, knows very little about you relatively speaking, about the broader context of your life. So it’s like a poor man’s agent that tries to represent your interests and serve what you want. A true agentic version of Instagram has an agent that represents what you want and can do a much better job ranking, filtering the content that you see at any point in time based on a bunch of other things, and it’s very tricky because Instagram can’t know about these things, because if they do, they will create this massive profile about you. So there’s a whole new architecture of the Internet that will have to be invented for these agentic capabilities to become unleashed because you have to keep your data…
…HB: Yeah, and this is where we get into I think the meat of the topic, which is what does a world of AR apps look like? What does it feel like to live in it? I’ve used this YouTube video called Hyper-Reality multiple times when I’ve given talks on AR. It’s completely absurd, it’s a world that we don’t want to live in, but it’s a joke, but it’s also not. So I always encourage people to go watch Hyper-Reality, it’s a beautiful artistic piece.
Before we get into what living in this AR world looks like, there’s a couple of things that I always like to talk about. One is that direct manipulation, which is what Apple brought to the world with multi-touch — we’ve had other forms in the past, but that’s really when it arrived — is genius and it has and will continue to exist, and direct manipulation when you’re literally touching something with your fingers has to be tactile, meaning it requires a physical surface. Pinch-to-zoom in midair isn’t nearly as useful as something on a tactile hard surface, so that’s the first thing.
The second thing is our arms get tired. This idea of midair computing is only really useful for quick actions. There’s this hilarious scene in Minority Report where Tom Cruise is probably sweating by making lots and lots of gestures in midair, and perhaps ahead of their time in their vision, but that’s not a thing, people don’t want to be computing in midair.
And look, if Tom Cruise can’t do it, none of us can do it.
HB: (laughing) Exactly. So anyway, we have to keep those things in mind. Direct manipulation is genius and your arms get tired, so there are three modalities of UI and UX in the Spatial Computing paradigm. The first are your tools, they’re like your utility belt, they’re things that walk with you wherever you go. They might be body-locked or in some cases head-locked, your notifications tray, your settings, your menu, etc., these things walk with you where you go, and you will access them through both 2D gestures and 3D gestures. But it’s all quick, it’s just how you get into the thing that you want to do.
Right, this is almost like the mechanical wristwatch of UI layers.
HB: Exactly right. So that’s your utility belt, we’re going to bring that with you everywhere.
The second thing are world-anchored apps. So it’s basically walking to your house and instantiating an app on your table sitting down and then playing with it. That app might be a 2D iPad style app, it might be a 2D app on a massive surface, it might be a little 3D app, like a tabletop app. Imagine calling an Uber using a tabletop 3D map that allows you to say exactly where you want to get picked up.
Right. You can pick up the car and put it on the map where you want it.
HB: You can put it on the map. So this is really useful because you can instantiate any app on any surface at any time.
Then the third thing, which is where it gets really exciting, are world-anchored virtual objects and maybe screens as well. So these are things that are just in the world. You walk into your house, you’ve got art on the walls, you’ve got maybe a control panel where ordinarily a light switch would be, and it allows you to do all sorts of things with your house because it’s not actually there, it’s just the wall. But you see something, you see a control panel on the wall that’s rendered for you and it will be agentic, etc., all that stuff.
That’s like real augmented reality because you are actually augmenting reality.
HB: Yes, this is real augmented reality. Think about annotating the world as well. You saw in the Orion demo the recipe thing where it annotates the ingredients and visually tracks them so if you walk away and look back, it’s still tracking them, they’re still there. This is crazy interesting stuff, and that’s where a lot of the new types of use cases are going to come from, and that’s it. Those are the three categories of UI in an augmented reality world…
…Yeah, this is the challenge here. You have a couple Apple points here, the one thing about linking it to the smartphone is, if you can offload all that compute and offload all that battery and offload all that connectivity into one device, it makes it a lot easier. I mean, you said for Apple, “Number eight, Apple will continue to slow-follow Meta on camera glasses and mixed reality headsets, but will be several years behind on AR glasses”.
HB: Yeah. I think that it’s a really easy win for Apple to launch a competitor to the Ray-Ban glasses. I mean, it’s a proven form factor, just do it and I think they’ll do a fantastic job at it. It makes total sense because of Apple Visual Intelligence. It’s just, just do it. So that was a rumor from Mark Gurman from I think last week, which I really believe in. The earbuds with cameras, I’m not so sure, but I do believe that camera glasses are a thing that makes sense for Apple to be building.
Now, the AR glasses though is not, in my opinion, a product that we’re going to see from Apple in much less than 10 years.
Wow.
HB: I think that one is going to take a very, very, very long time.
And why is that?
HB: I just think their product bar is going to be insanely high, and I think they’re going to have some hard architectural decisions to make. Is it attached to your iPhone as an accessory or is it a standalone thing with its own puck like Meta did? There’s a lot of trade-offs there that I think people don’t necessarily think about carefully. It is not easy for Apple, that’s my next point.
Number nine.
HB: To ship AR glasses as a smartphone accessory, because in practice they have significant cost margin, thermal envelope constraints on the iPhone because the iPhone is a single, super high volume product that needs to be a great product and a highly profitable smartphone, first and foremost, so as soon as you have to start to add more components to this thing to power AR glasses, you’re tasking your primarily profit center for the whole company and creating all these architectural constraints.
Couldn’t they just make an extra model like the AR model? But I guess then that ruins your TAM.
HB: Yeah, I think that’s like the worst of all worlds, in my opinion.
This is really interesting, 10 years does blow my mind because yeah, your thought immediately, let me restate your argument, make sure I get it, your thought immediately goes to Apple already has a smartphone, they can just do an accessory, but actually the issue is the smartphone is so successful and so profitable and so essential that, 100 million, is that in a quarter, whatever, all those smartphones can’t be compromised to support this because they’re so important.
HB: And the attach rate just doesn’t justify it.
That’s right.
HB: Look at the attach rate of Apple Watch, the attach rate of Apple Watch is still fairly small.
And yet if you did a separate model, you’re giving away your entire advantage so you’re stuck.
HB: Exactly. So I think it’s a harder trade-off space than people realize for Apple, and my guess is that this is a discussion that is highly unresolved.
5. Researchers say an AI-powered transcription tool used in hospitals invents things no one ever said – Garance Burke and Hilke Schellmann
Tech behemoth OpenAI has touted its artificial intelligence-powered transcription tool Whisper as having near “human level robustness and accuracy.”
But Whisper has a major flaw: It is prone to making up chunks of text or even entire sentences, according to interviews with more than a dozen software engineers, developers and academic researchers…
…More concerning, they said, is a rush by medical centers to utilize Whisper-based tools to transcribe patients’ consultations with doctors, despite OpenAI’ s warnings that the tool should not be used in “high-risk domains.”
The full extent of the problem is difficult to discern, but researchers and engineers said they frequently have come across Whisper’s hallucinations in their work. A University of Michigan researcher conducting a study of public meetings, for example, said he found hallucinations in eight out of every 10 audio transcriptions he inspected, before he started trying to improve the model.
A machine learning engineer said he initially discovered hallucinations in about half of the over 100 hours of Whisper transcriptions he analyzed. A third developer said he found hallucinations in nearly every one of the 26,000 transcripts he created with Whisper.
The problems persist even in well-recorded, short audio samples. A recent study by computer scientists uncovered 187 hallucinations in more than 13,000 clear audio snippets they examined.
That trend would lead to tens of thousands of faulty transcriptions over millions of recordings, researchers said…
…The tool is integrated into some versions of OpenAI’s flagship chatbot ChatGPT, and is a built-in offering in Oracle and Microsoft’s cloud computing platforms, which service thousands of companies worldwide. It is also used to transcribe and translate text into multiple languages…
…Professors Allison Koenecke of Cornell University and Mona Sloane of the University of Virginia examined thousands of short snippets they obtained from TalkBank, a research repository hosted at Carnegie Mellon University. They determined that nearly 40% of the hallucinations were harmful or concerning because the speaker could be misinterpreted or misrepresented.
In an example they uncovered, a speaker said, “He, the boy, was going to, I’m not sure exactly, take the umbrella.”
But the transcription software added: “He took a big piece of a cross, a teeny, small piece … I’m sure he didn’t have a terror knife so he killed a number of people.”
A speaker in another recording described “two other girls and one lady.” Whisper invented extra commentary on race, adding “two other girls and one lady, um, which were Black.”
In a third transcription, Whisper invented a non-existent medication called “hyperactivated antibiotics.”
Researchers aren’t certain why Whisper and similar tools hallucinate, but software developers said the fabrications tend to occur amid pauses, background sounds or music playing…
…Over 30,000 clinicians and 40 health systems, including the Mankato Clinic in Minnesota and Children’s Hospital Los Angeles, have started using a Whisper-based tool built by Nabla, which has offices in France and the U.S.
That tool was fine-tuned on medical language to transcribe and summarize patients’ interactions, said Nabla’s chief technology officer Martin Raison.
Company officials said they are aware that Whisper can hallucinate and are addressing the problem.
It’s impossible to compare Nabla’s AI-generated transcript to the original recording because Nabla’s tool erases the original audio for “data safety reasons,” Raison said.
Nabla said the tool has been used to transcribe an estimated 7 million medical visits.
Disclaimer: None of the information or analysis presented is intended to form the basis for any offer or recommendation. We currently have a vested interest in Apple, Meta Platforms, and Microsoft. Holdings are subject to change at any time.
