Bet on it: Zvi Mowshowitz on professional gambling, trading, and AI futures
![Bet on it: Zvi Mowshowitz on professional gambling, trading, and AI futures](/content/images/size/w1156/2024/08/Zvi-Mowshowitz-square-1.png)
This week I interviewed Zvi Mowshowitz, who I Internet-met 30+ years ago in middle school then became acquainted with for real through his recent public intellectualism on covid. We talked about his interesting career path through Magic, professional gambling, and equities trading, which are very different but rhyme in important ways.
[Patrick notes: I add commentary to the transcript using this convention. This is, necessarily, after the conversation and in my personal capacity. As always, nothing you read in my personal spaces is necessarily endorsed by other organizations, including Call The Shots, Inc., a 501c3 that I ran which is discussed (as VaccinateCA) below.]
Sponsor: This podcast is sponsored by Check, the leading payroll infrastructure provider and pioneer of embedded payroll. Check makes it easy for any SaaS platform to build a payroll business, and already powers 60+ popular platforms. Head to checkhq.com/complex and tell them patio11 sent you.
(00:00) Intro
(00:16) Meet Zvi Mowshowitz
(04:11) Trading and Magic: The Gathering
(07:24) Professional sports gambling
(11:58) Navigating the sportsbook market
(22:33) Sponsor: Check
(23:48) Financial markets vs. sports betting
(34:02) Covid-19 early predictions
(43:21) Covid-19 policy failures and blame
(49:52) Vaccine rollout chaos
(01:01:11) The importance of scaling effective strategies
(01:14:46) AI predictions
(01:23:58) Wrap
(Many of the above section headers are available via Ctrl-F if you'd like to skip to that part in the transcript. Sadly, for technical reasons, I can't reliably hyperlink them everywhere this shows up.)
Transcript
Patrick McKenzie: Hi everybody, my name is Patrick McKenzie, better known as Patio11 on the internets, and I'm here with my buddy Zvi Mowshowitz.
Zvi Mowshowitz: Yup, also known as TheZvi.
Patrick McKenzie: I don't know the quickest way to encapsulate you professionally for people who might not be familiar with your work – I would say public intellectual. Do you have anything that you would prefer over that?
Zvi Mowshowitz: I think ‘public intellectual,’ ‘rogue journalist about AI’ at this point, largely – something like that.
Patrick McKenzie: Currently you write a newsletter largely about roundups of what's been happening in the world of AI, but you've done a number of things over the years. I actually became aware of you when I was in middle school, 25+ years ago, because you were big on the Magic: the Gathering scene. So what happened in between Magic: the Gathering and today?
[Patrick notes: It’s difficult to overstate how formative M:tG was during my middle school years. Some of my first writing for the Internet was analysis of the metagame for the Dojo (the community’s premier site for cognoscenti in the early days). On the Internet, no one knows you are a middle schooler. (As I’ve recounted previously, some middle schoolers use this to substantially more advantage than I did.) ]
Zvi Mowshowitz: So many things. I played a lot of Magic: the Gathering in high school and college. Decided to become a professional; spent roughly 10 years in some combination of playing Magic and writing about Magic to pay the bills and keep going.
I tried to join a startup developing a game. That didn't work out very well. I got into sports betting in various capacities for a while; after that I tried to found a personalized medicine startup called MetaMed which lasted a few years and got reasonably far, but ended up not having good product-market fit, basically – and we didn't understand marketing – and it didn't work out.
Then I joined Jane Street Capital. Then I did a year or so of crypto trading.
Then COVID happened: I did a combination of making my own game, sort of in the style of Magic, which didn't ultimately work out, and I started spending a lot of time worrying about COVID and thinking about COVID and analyzing COVID.
I realized I might as well share this information with other people; I have a blog, I've written about rationality style stuff on that blog for a while, so I started writing COVID posts starting (I think) sometime in February 2020. That turned into weekly updates. Suddenly I realized there were hundreds of people who were like, “I really depend on your weekly updates to not have to otherwise think about COVID so much,” and so I kept going; one thing led to another.
When COVID was winding down and I had just written my last COVID post, entitled “Your Best Possible Situation” – which is that you get to stop writing these posts and go back to your life – I jokingly said weekly COVID posts should be replaced with weekly AI posts because there was a lot happening.
Then everyone said, “yeah, yeah, absolutely, yeah, you gotta do this!”
[Patrick notes: My Internet buddy Visa keeps a collection “Joke about the outcomes you want.” There is a corollary to this, which is: do not joke about outcomes you don’t want. Having seen how self-deprecation gets rewarded in the world for a few decades and then how rapidly response changes when one simply stops doing it, I would flag this as being far deeper wisdom than people generally appreciate. (Well, I mostly stopped doing it.)]
Zvi Mowshowitz: I was just thinking to myself, “oh no… I, I guess.”
I tried it, and it got a very positive reception. One thing led to another, and now this is my day – all day, all the time.
Patrick McKenzie: The best possible form of audience capture is when the audience says, “we think you're actually good at something, we'd like you to continue doing that for the next couple of years” – something that's pro social and actually useful for the world.
[Patrick notes: The less salubrious form of audience capture is when the audience uses your output for a form of wireheading which is against their interests to consume. You, as a producer, then get awarded for wireheading them harder and infecting more people with the memes that have adversely impacted the original audience. Eventually, this results in being the intellectual equivalent of PepsiCo, with you doingincreasingly elaborate features of chemical engineering, impossible in nature, while running a world-spanning highly optimize marketing operation. This stunning edifice of intellectual effort will be in the service of killing your customer base just slowly enough to be described as not intentionally killing your customer base.
…Well that got dark, but it is important.]
Zvi Mowshowitz: The thing is, I think it's super important. I've been a worrier in one form or another about existential risk from AI since about 2007 (or something like that) – well before any of these current things were happening. This has not come to me recently.
Suddenly AI is moving forward much quicker; a lot of things are happening and a lot of decisions are getting made. Helping people understand that, and helping people think better about artificial intelligence in all these different ways – and be able to do that faster (even though I write a metric ton of words that I expect people to mostly pick and choose from unless they want to go into the weeds) – a lot of people say they found it really helpful.
It has allowed me to speak directly on issues I feel matter to people at major labs, people who are in national security, people who potentially help shape the way things go.
I'm really excited to keep doing that, even as it completely overwhelms every hour of every day.
Trading and Magic: The Gathering
Patrick McKenzie: So I think we'll return to AI in a moment, but to go back into the mist of history, you've done three somewhat different, somewhat similar things that broadly look a little bit like trading. (I suppose Magic might look like trading in some ways too.)
Zvi Mowshowitz: Magic started out largely also as trading, because everybody had a limited pool of cards and there wasn't a good, strong marketplace – and we were all kids so we didn't have the money to just buy everything we wanted, we had to acquire them through opening packs and trading.
So trading was a substantial portion of that, and I kind of didn't like it. It was very much, you spend a lot of time trying to find people who will give you a reasonable deal – then when you know more than someone else and you can take advantage of them, you don't want to because you feel bad about ripping off a person in front of you, especially because it's often a kid.
You were basically in this world where you were trading to try and get the right basket of goods that you wanted, but you never really got value from it. Occasionally you'd be able to get some value, and feel good about it because you were giving them what they wanted – and that was a service, and you were organizing and providing – but mostly it was just trying to not be cheated all day and making sure that the trades were fair.
That just got very exhausting, and after a while I was just like “I don't have time for this.” I got enough money that I could afford just to start buying my cards, and that was a great relief.
But trading has always been fascinating to me – as a game, as a competition, as an intellectual exercise, as a thing that I believe makes the world a much better place by distributing resources properly and allowing people to figure out what matters and all that.
Patrick McKenzie: I think many people who work in finance in some capacity, or in the markets, or doing “formal trading” – or even high stakes negotiation, etc. – got their start on things that society broadly does not believe are real, but which certainly felt real to us in the moment.
People on RuneScape with the “I can't believe it's not stock market,” – a crippled stock market, but a stock market – the World of Warcraft Auction House systems, Magic: the Gathering cards, etc.
Some of them have had longevity into our adult lives, but the lessons people took from those various things have definitely echoed in real institutions over the years.
Zvi Mowshowitz: Absolutely, it makes you feel comfortable with it, and it makes you appreciate it.
The Magic dealers will take 30% of the transaction roundtrip, something like that. If you want to sell your cards to them, and then buy others, if you don't have the cards they specifically want, they will pay you almost nothing – and yet, those Magic dealers are providing an insanely valuable service, without which my life would have been a living hell.
I would have had to go around individual players, trying to engage in barter style transactions all the time. That was fun when I was 14, and it was not fun when I was 24.
Then I get into other markets, some of which are more clearly zero-sum and competitive, and some of which are clearly productive and useful.
I just found the whole thing fascinating. It's a contest of trying to figure stuff out and learn about the world, and it's a fire that teaches you probability and statistics, various forms of psychology, and strategy, and so many other things.
Professional sports gambling
Patrick McKenzie: We were talking earlier about – a hole in my financial knowledge is about sportsbooks – but you did that professionally for a while.
Can you tell the audience a little about what being a professional sports gambler looks like?
[Patrick notes: I have a complicated relationship with gambling. I’m not opposed to it, per se, and have been known to compensate independent providers of entertainment services on a probabilistic basis via e.g. playing at the same poker table as them. However, I think we’ve done a disservice to the broader community by making gambling hyperaccessible. There was societal value in making it something you had to get on a bus to do and in putting some friction between your bank account and the craps table, even if that friction was just walking from the table to the conveniently situated ATM. Now with everything on your phones, the marginal player is not just betting against the house, they’re betting their self-control will exceed the ability to provoke action of the best product and marketing teams given billion dollar budgets. Most will lose that metaphorical bet, and then go on to lose the real bets.
And on the gripping hand, at least we have deep-seated societal antibodies against this when we call it a sportsbook, in a way we don’t when the gambling app on your phone wears the clothes of a responsible financial institution.
Anyhow, qualms disclosed, the reason I wanted to talk with Zvi about his sports betting career is because it demonstrates some concepts relevant to more formal markets.]
Zvi Mowshowitz: Yeah, so being a professional sports gambler… first of all, it's one of those jobs where you choose where you want to draw your attention, where you want to pay attention to, where you want to focus on, what attacks you want to make.
The great advantage of being a sports bettor is the great advantage of any trader, which is that you always have the option to do nothing. The house has the edge, and the house will ultimately almost always win, or else they wouldn't be a house.
[Patrick notes: Even people whose job is capital allocation will necessarily not place a bet in 99.999%+ of opportunities available to them. This isn’t necessarily expressing a negative point of view on the opportunities not chose. Partly, it’s because search/evaluation/execution costs are non-zero. Partly, it is because you are likely capital constrained in some fashion and so don’t want to pick merely positive bets, you want to pick the best positive bets. This is also why galactically sized pools of capital tend to indexing over active management over time: if you no longer are capital constrained, then you just want to buy all of the positive bets at a minimum of search/evaluation/execution cost.]
Zvi Mowshowitz: But if you are a smart gambler trying to win, your great advantage is you can look at 100 markets a day; of those 100 markets, 97 of them, you will not find anything that you find interesting, potentially, if you're looking at one sportsbook in isolation. You can just not trade on those at all – that's totally fine.
Then you find the one big mistake they made, and you exploit it, and you also find two other places where you think they may have made a mistake, where you can supplement, and you tune in on that – then you can still make a lot of money.
The art of sports betting from the bettor’s perspective is a combination of many skills: you need to figure out how to watch sports, how to analyze sports, build a model of the sport on both a fundamental level – which teams are good, what's likely the outcome of the basic situation – and also…
One of the things I specialized in was derivatives, essentially: if the home team is a two-to-one favorite, and on average, it's going to be 42.5 points in this NFL game, how many points will the home team over-under be? And what are the odds on that? I specialized in questions like that and some more exotic versions of that, or, “what's the chance that goes over 48.5 points instead of 42.5 points?” “We know it's an underdog, but exactly how big an underdog?” They can sometimes severely misprice this.
[Patrick notes: I am glad I never learned until today that sports games had derivatives available, or I would have sunk much, much more intellectual effort into them than I have.]
Zvi Mowshowitz: Also, I focus on arbitrage and on comparing the odds at various different places.
That’s one of the things – in the American stock market, you go there and there's one price.
[Patrick notes: He used to work at Jane Street trading equities, folks; you do not need to tell him that this is an oversimplification to show degree of difference.]
Zvi Mowshowitz: You go to sports betting and yeah, the price is mostly aligned, but the sportsbooks largely do not have any mechanical reason why they have to have the same price except for people who do arbitrage.
When I got to crypto, it was just a feast of that. Everyone's price is different; moving money around takes time. People have different preferences. People are very irrational. You can take advantage of this.
[Patrick notes: A material portion of the market dislocation in crypto, particularly in the 2017ish timeframe, was that people simply did not have the physical capability to trade in disparate markets at the same time. This was part of Alameda’s stated core competencies; the unstated part was “also we’re surprisingly willing to commit crimes.”]
Zvi Mowshowitz: Back in sports betting, if you have a combination of:
- You understand derivatives- You understand price and line shopping, - You understand what all the numbers mean (i.e., you take a baseball game and it's like, okay, this is the Yankees -1.5 runs, and this is the Yankees to win, and you know which of those lines is better given the total of the game by exactly how much, and the other context, and you know whether or not it matters whether the closer was used yesterday, and blah blah blah – hint, it mostly doesn't.)
You put it all together, you line shop, you watch things develop over the course of the day, and you use all of your different attributes.
We used a strategy nicknamed “Metroid Prime,” the idea that we would combine all of our different sources of information and knowledge and heuristics, then we'd find opportunities where it aligned together – where this one edge says that we have something, then all of our other indicators also like it. It was all very human.
Also, back then, it wasn't a ruthless competition. At one point – before I got a really sophisticated setup, because we didn’t have a big enough bankroll – I was sitting in an apartment, I'd have one monitor and I’d alt-tab between the different sportsbooks.
I'd have eight sportsbooks up where I had money. And I would wait for the line at the sportsbook that I believed had the significant information. That “the line in motion is generally the trustworthy line” is one thing you learn – the market that's moving is the market that has action, the market that has actual volume, therefore you can trust it.
I'd wait for it – okay. I have a plan in my head. If this line moves here, I'm going to hit everybody who has this number still up before they can move it. It's called “chasing steam,” which is frowned upon, but at the time was largely tolerated.
I would often be able to get three or four distinct bets in, each of which takes two or three distinct website stages, with no bot aid whatsoever – just typing furiously and clicking the mouse – before the lines had moved at the end of the sportsbook, which means that there weren't that many people who were trying to do the same thing.
Navigating the sportsbook market
Patrick McKenzie: Correct me if I'm wrong, but on a sportsbook, the sportsbook itself is acting somewhat as a market maker, and that they're taking some amount of principal risk, but when they get unbalanced on one side or another of a particular market, they will move their prices to lay off some of that risk?
[Patrick notes: We are about to go into the deep end of discussing sports lines. I was not familiar with quoting conventions when speaking with Zvi, and assuming many in the audience are similar, I’ve put brief glosses in brackets italicized. In particular, a useful convention for “American odds” is that negative odds mean “The amount of money at risk to win $100 net of the price of your bet” (i.e. -$50 means you wager $50 to potentially win $150 gross. Positive odds mean “a bet of $100 wins $X net of the bet.” And thus -1 means 100:1 odds against the outcome in question and +1 means 100:1 odds in favor of the outcome in question. Make sense? Yeah, I’m also pining for the simplicity of Lighting Bolting an attacker in response to the Giant Growth, which obviously is entirely mechanically different than countering the Giant Growth.]
Zvi Mowshowitz: So this is one of those situations in which there are a lot of different strategies depending on what the sportsbook is, what it's trying to do, what level of confidence and bankroll they have.
The traditional version of this is doing the NFL game: the line is -7. [Team X to lose by 7 points.]
Some guy bets a lot of money at -7 at -110 [Team X to lose by 7 points or more with $110 risked to win $100 net if that comes to pass]; you move the line to 7.5 and try to get action there. The seven point is only worth about 14 cents so that – even though it's a relatively big point – you still make six cents on net because you got a -110 and a -110 and you're profitable, and someone's going to want 7.5.
[Patrick notes: Yep, that’s dense. If you had fully offsetting bets at -7, you’d be taking in $220 and paying out $200. You have less than fully offsetting bets if you move the line; the loss you take on exactly -7 is $14 in expectation. And so, $6 (six cents) in expected profit on net, if you can convince someone to take the proposition “my favorite team [the other guys] wins by 8 points or more [+7.5].”]
Zvi Mowshowitz: Occasionally the line will move 7, 7.5, 8.5, 9.5 and you'll be in trouble, but you try to often enough not let that happen – not let that happen for too much money – and you get the two-way and you earn, and you make your profit, you make your 5% ideally, and you go home.
[Patrick notes: The line moving in that fashion is similar to an equities marketmaker being “run over.” They did not get into this business to express an opinion on the underlying, but now they have a position and it is probabilistically losing money as we speak. Options include either a) waiting and hoping fortunes reverse or b) find someone to buy you out of the position at a guaranteed loss.]
Zvi Mowshowitz: But the modern sportsbook philosophy would treat making 5 percent at a place like a Bodog traditionally, back in the day, or today like a FanDuel or a DraftKings… if you were managing the lines at FanDuel and you're charging these prices they're charging – which are kind of outrageous, even on the main lines – if you reported back to your boss that you only earned the 5% (that is kind of the dream scenario for the getting two-way action), your boss would think you're bad at this. You are not doing a good job.
The reason for that is because the odds that balance the market are not the same as the best belief you can have about the outcome of the game by watching exactly which of your customers are betting on exactly which of these things, if you are on the inside, and watching the lines move and develop at all of these other sportsbooks.
[Patrick notes: This was a revelatory point for me, and to draw it out a little more: the odds that balance the market may involve less total wagered by punters. You should welcome accepting more total wagers if you’re pretty sure the marginal wagers are only low-information punters, accepting that you are now directly betting against them versus merely market-making, because betting against low-information punters is a good way to make money.]
Zvi Mowshowitz: You have a wealth of information because you can divide the sports betting world into – metaphorically, as a simplification – sharps and squares. The squares are the regular guy in the street; he goes to the bar, he has a drink, he bets on his favorite team. He uses intuition, he uses his basic eyes.
The sharp is the guy who's using a mathematical model. He's fighting for a living – he's trying to win very aggressively. And so you think of it as, “the consensus of the sharp models is pretty accurate.”
But then there's the other people in the market, and the sharps aren't gonna bet unless the line moves enough that they're interested.
Patrick McKenzie: So in terms of zero sum markets, this is pretty close to a pure zero-sum market with a large fee attached to it from the sports betting platform; essentially, the source of excess returns for sharps is successfully identifying the squares, the regular punters who are betting on the team that they've been following since childhood, or other less signalful things than attempting to trade with each other as little as possible.
Zvi Mowshowitz: Certainly that is one way to look at it. There are a lot of different strategies you can use with the sharp.
One thing you can do with the sharp is you can be very, very good at identifying when the public is all on one side, and then be good at line shopping so you find a way to bet on that without paying the full 5 percent or even worse to get on the other side.
That's profitable even if you have no idea what these teams are, who is good, what the number is supposed to be in some sense, although all of that knowledge on a basic level is very useful to avoid making stupid mistakes.
Patrick McKenzie: Can you explain briefly what line shopping is?
Zvi Mowshowitz: Line shopping is where I have open FanDuel, and I have open DraftKings, and I have open Bodog, and I have open Pinnacle, and I have open all these other things, or I have an automated feed – Don Best sells you a heads-up display that'll show you the prices everywhere.
You say, okay, “here it's -110, here it's -110, here it's -110, here it's -105 – hmm; here it’s -110, here it’s +100 where I get one-to-one instead of paying 11 to 10.” Then you take the +100.
Then ideally you look and you find a place that's got a number that is past the median of the other people's lines; if the average line is -110 / -110 everywhere, and then there's some place that you don't think has any intelligence, that has gotten unbalanced action and now moved to -125 / +105, so now you can get odds instead of giving up odds, you get a little bit – and if you don't have any reason to think they know anything, you just knock that +105 off.
[Patrick notes: Again, -$110 is $110 wagered to gain $210 if you win, or about 2:1 odds. Zvi, seeing this prevailing in many markets, thinks that 2:1 is approximately right; it’s 50/50 whether team X wins or not. One market, due to heavily asymmetric play by its playerbase, is attempting to lay off (hedge) action on one side by giving favorable pricing to one side: for a 50/50 bet, they’ll pay you $205 if you wager $100 and win. Zvi’s point is “Assuming you think this market is not correct, you should be willing to take that bet automatically even if not hedgeable. It is positive expected value; your “hedge” for it is intertemporal. You will take a lot of small-for-your-bankroll +EV bets over the years and variance will eventually result in you making money contingent on you being right with other parts of this analysis.]
Zvi Mowshowitz: You can't hedge it – you can't find an even-money bet on the other side, necessarily – but now if you're confident the true odds are 50/50, just by looking at the rest of the markets, you've made money (or even if they're like 49.9 because the sportsbook occasionally is doing something for reasons but mostly it's got a bunch of idiots). You can use that to make money.
Other people are just pure handicappers – they say, “I have a detailed mathematical model of the English Premier League; I'm going to figure out which team is how likely to win this game, I'm going to know who's injured, who’s likely to play, whose motivations are what, where we're at home, and this many days rest. I'm going to put it all together, and compare it to the odds. When the odds are significantly different from what I have, I'm going to bet.
“And I will naturally end up on the correct side of these games – presumably – because the sharps will agree with me because they've also done the same analysis and they'll figure it out.”
But that's not why I'm here. I'm just here because I like the stock. I like the odds, so I’ll just buy it because it's cheap.
Over time, the lines tend to move in the direction of that not being as good a strategy. The odds will be bought towards fair value. Because there's 10 of you, right, or 100 of you, that all kind of know that Chelsea is a good bet today. Mm-Hmm. You don't wanna wait until they bet and tell the sportsbook that Chelsea is a good bet today and they move the odds towards Chelsea.
So you want to get it there first. Sometimes it moves the wrong way because the squares overwhelm you in a surprising fashion, but not by default.
You're trying to maximize all these different things at once and the strategy is building up the superstructure and figuring out how to profit.
Patrick McKenzie: I know relatively little about the market, but a fascinating thing I learned was that the sportsbooks aren't blind to the identity of the customer in the way that market makers would be typically blind to the identity of the customer in traditional markets. You’re allowed to give worse prices to people who you think have an edge versus people who don't have an edge; you're allowed to do various shenanigans to attempt to incite interest in the market, and then pull those back before you get action, etc.
[Patrick notes: Importantly, as covered in the conversation with Ricki, this is a) not formally allowed in many real markets and b) because it is not formally allowed, we create market infrastructure which allows us to generate the outcome without formally doing it. For example, internalization of orders at retail brokerages / payment for order flow sells specialized firms the opportunity to interact with low-information traders at better prices than they’d quote in venues where e.g. Goldman Sachs can possibly have an account.]
Zvi Mowshowitz: It's much less worse prices than it is lower volume, lower size. The idea is, ok, you are famous gambler Billy Walters, and you wanna bet at our casino on an NFL game. Normally we would let you bet $10,000, $20,000, $100,000, because this is a pretty standard thing that people bet on, but if we know who you are… “you know what, Billy, you want $100? You can have it, but we're not going to give you a real bet because we expect you to win. But if you want to give us your information for almost no money, we're not going to stop you – and we're not legally allowed to strictly kick you out of a legal sportsbook these days.”
So instead DraftKings lets you bet, like, a dollar – or three cents in extreme cases, if it just goes too far and the guy just doesn't quit. But this is what they do: they analyze the player. If the player is winning, most sportsbooks’ responses will be “take a hike, we don't want your action now.”
[Patrick notes: After this episode was recorded, the WSJ ran an expose on this topic. Some years before that, the NYT ran a piece titled “tips to avoid getting limited by a sportsbook.” (And yeah, I am aware that the NYT will say “Oh that piece on the nytimes.com domain name is not really the NYT”, but you pick your brand strategy and you take your chances.)]
Zvi Mowshowitz: They're not mad at you. They don't think you cheated – it's not like counting cards in a blackjack game. They just say, “okay, we don't want your action anymore, because it's not worth the information that we gain and the liquidity that you provide to have to be on our toes all the time and to take this bath” – and again, if you're trying to make 8% or 10% instead of 5%, you have to have a price you know is wrong – so even if they break even, you're reducing this guy's percent return, and you're hurting his metrics, he wants you gone.
Obviously, you have to balance the idea to have a balanced book. You don't want to take on too much risk. You don't want to have too much variance. You don't want to risk – like, the public isn't an idiot. If you just opened 50/50 odds on every game, the public would clean your clock completely and utterly immediately, even if they had no access to another market, because they know who the good teams are.
It's just about this precision. It's the whole question of the efficient market hypothesis – people say, “oh, there's no way to make a profit on the stock market. You might as well buy an index fund, because anything you know is already priced in.” And the answer is mostly – right?
You're not going to make huge amounts by using public information and using common sense, but it doesn't mean that everybody's acting correctly, because the people who are paying attention, that attention is valuable; they only have so much capital; there are a lot of things pushing against it – you know, you have idiosyncratic knowledge, doesn't mean it doesn't matter.
In sports that mostly means you're going to get relatively small edges. The sportsbooks are pretty good at figuring out when they have mispriced and the public knows they're crazy, and the line moves a lot very quickly – including people just anticipating that other people will realize the line is wrong, even if they don't actually disagree with the line.
There are a lot of these dynamics. The key skill – when you have access to all these markets and want to make a profit purely on trading – is, you want to figure out what the good prices are, where you can make money, but you also simultaneously have to avoid getting kicked out of all these casinos.
There's a handful in the world that will say, “we welcome winners.” You can go to Pinnacle and you can say, “I'm going to make millions of dollars off you every year because I'm just going to be better at the largest markets than everybody else, and I'm going to bet winners over and over and over again. “
Pinnacle will say, “you know what? That's our business. We will find a way to get action against you, we'll use that information. Come on in.”
But FanDuel and DraftKings will tell you to take a hike because that is not their business. William Hill does not want your business.
You might or might not try to use what's called ‘beards,’ or try to convert whales to try and sneak in, when they don't know it's you after you get kicked out – it's against the terms of service, don't do that – you might just try to use it once and try to disguise what you're doing.
One of the things you never want to do is join for the first time and then suddenly make a bunch of weird prop bets and obscure bets on Romanian third league that give away the game, that tell them right away who you are, because the Bayesian evidence is so strong. That's the first thing you're betting in your accounts in Chicago – the chance you are anyone I want anything to do with is now 1 percent or less, and if I lose that customer when I shouldn't have every now and then, you know what? That's a risk I'm willing to take. And you're probably just gone already.
Financial markets vs. sports betting
Patrick McKenzie: Segueing a little bit from the kind of informal, non-financial markets to the actual financial markets – in what ways is that similar? In what ways is it quite different?
Zvi Mowshowitz: The biggest thing is that it's unified and it is much, much more competitive and much, much more efficient. Also, the skill level is through the roof. I was, in my opinion, the best trader – both as a gambler, in terms of my execution, my attempt to do skill; with no computer aids, I was ahead of everybody. As a professional trader, I felt like I was the best guy in the room.
Then I go to Jane Street, and despite all this experience, I am the worst trader on my desk when I come in. I have a lot of great instincts, they're really excited to work with me, they're really excited to see what I can do, these people are whip-smart – and I'm not saying I'm the dumbest guy there, I'm not, but these people know what they are doing.
I'm not going to tell you exactly what they know or how they do it because that's trade secrets – but like, “oh my god, they are very very good traders.” Any one of them, if they tasked them with beating the sports markets, they'd all crush it. They'd all just completely destroy this thing.
[Patrick notes: One reason why some markets stay inefficient and somewhat grubby is that capitalism is constantly performing a sort across all human capital. If you’re really, really good at trading, capitalism will happily stake you with much more of society’s resources to grind out a much smaller efficiency improvement at a much more important problem (the real economy versus play markets). The people needed to make the market fully efficient would be better used elsewhere, and so the market stays inefficient.
This argument generalizes far away from particular gambling arenas. It is also why there are excess returns available for small software developers but the process of going out and getting them will cause many to end up in AppAmaGooFaceSoft instead. Or, I suppose, running podcasts.]
Zvi Mowshowitz: The edges you're looking for are now vastly, vastly smaller. In sports, you're looking at edges of a few percent. Often people say the edges are small, but they're not as small as they think they are – but there's still a few percent at most.
And the best thing is that it turns around very quickly. The game is tomorrow, the game is today, the game is maybe on Sunday, and you get your money back – and you can do it again, whereas when you're in finance, you're using all of this leverage, and you're looking for edges that are not measured in basis points – hundredths of a percent.
If you could have a trade that returns five basis points today, that's a killer trade in a place like Jane Street. You are amazed to find that trade.
I mean, they do trades that make more than that, obviously, but if that trade safely has that EV, even if it's wildly risky in every individual case, as long as it doesn't take too long so the cost of capital isn’t an issue, you're super excited – and you expect that not to last. You expect Citadel to find that trade next week and Virte to find that trade three weeks from now and so on. It’s going to be gone, so use it while you can.
There's always a question of, “why am I allowed to make this trade?” The financial market's like, “Who is selling to me at this price? This price is wrong, so why am I getting to do this? Who is my counterparty? What do they know that I don't know?” – these questions, sometimes you don't have good answers to this. But you need to think about these questions, you need to be very sophisticated.
Everyone's constantly getting smarter, everyone's constantly getting faster; everyone's automating more things, they're bringing in various forms of AI. Financial markets are much, much way, way, way ahead. So that's the biggest difference.
The other difference, of course, is they're positive-sum, in the sense that these are real investments; the numbers can go up. There are people who have economic reasons to be here, who are fine with losing money versus the efficient allocation of capital, in the expected value sense, in order to hedge their risk, in order to put it to work in the way that they need.
Therefore there's a pool of money that gets to be distributed amongst the people who are making the market run well. That is very prosocial, and you get to be part of that.
Patrick McKenzie: One of the interesting things for me about trading is that you have one layer which is views on the fundamentals, and then you have another layer where you're trying to connect a high-fidelity model of the minds of people who you might or might not know the identity of, and how those play off each other.
[Patrick notes: I credit Luke Constable for most improving my model here: there are the fundamentals, and then there are capital flows, and all models of the world which do not include structural reasons for capital flows will fail catastrophically to model the world.]
Patrick McKenzie: Then one level above that it's like, some variety of trades are where there is someone in the ecosystem who will be doing something for other than strictly economic reasons. One of the classic ones is index funds rebalancing; there's lots of reasons for index funds to be more efficient than paying investment managers to make decisions where they historically do not have alpha and edge over the cost of trading fees and taxes.
Everyone has been piling into index funds; it’s a computer program, the computer programs are described in some fashion and you will know on which day an index change. There's an entire shadow economy which is built around anticipating index changes before they happen.
There is a case to make that that is legitimately prosocial, in that a large portion of the public will decide to make consequential decisions in their life, e.g. to invest their IRA in index funds, and that fact should be factored into the market. If there is kind of an arbitrariness in the computer code that the decision will be made six months from now, that thriving market in anticipating the decision helps to kind of backpropagate that information from the future to today, so that those folks get more efficient prices rather than being wedded to the prices that naive execution of the computer would make.
Zvi Mowshowitz: Yeah, it feels like a distortion when you're part of it.
I definitely participated in index-rebalancing trades, where you're trying to ask questions like, “How many other people are anticipating this index rebalance? When are they committing to it, given how much capital costs? Are they going to overdo this trade? If the index has to buy” – it was in some other country, because I was on the international desk – “they have to buy 5 million, they have 5 million shares of this to rebalance all the people who are going to have to buy. Well, collectively, are we going to buy 1 million shares, are we going to buy 5 million shares, or are we going to buy 10 million shares? Because if we all buy 1 million shares, whoever buys those is going to make a bunch of money. They're going to have a percent. We're going to sell only 1 of the 5 million to them. They're going to have to buy the rest from retail, or someone else who naturally has it; that's gonna, that's gonna be a missed opportunity.
“If it's 5 million, it's basically not going to move very much. They're going to come in and buy; we're gonna come in and sell.” It's fine. You want to stop at 3 or 4, collectively speaking – if you think that the combined number is getting close to 5, you're like, “maybe I should back off and not do this trade very much.”
If it's 10, then everyone loses money. And it definitely does happen – sometimes you overestimate and it doesn't happen. But in general people wouldn't come back and occasionally overload it if it wasn't generally good. It's generally good on a small scale – you're making bits, you're not making percents, especially in America.
But yeah, I think that it's a helpful thing to do.
I don't own S&P index funds; I own individual American stocks because I have access to the American exchange. Even though I don't necessarily think I have a huge edge on the market, I can do the Metroid Prime thing of like, I can combine all of my edges and if I have any advantage, then I can do better than picking at random – plus the tax advantage of, if I buy 20 individual stocks, and 5 of them are down and 15 of them are up (as opposed to having an S&P 500 index where it's just up) then if I want to get money out, I can balance a win, a number going up, with a number going down and suddenly I don't have a capital gain, and there’s nothing I can do about that otherwise.
[Patrick notes: Approximately 100% of the professional investors I’ve asked about what retail traders should do that isn’t the usual advice (“just use low-cost index funds”) have mentioned that retail traders have structural tax advantages and should, as an economic matter, sell those advantages to the rest of the economy. My conversation with Byrne contained an example. I think that net of attention and transaction costs most people should just go with the standard advice, but one wonders about either a) outliers among retail investors and b) what a product team could do here.]
Zvi Mowshowitz: So as an individual investor, I have reasons to diversify like that – but also because there are some bad companies out there, and one easy way to identify them is that they have high borrow costs on their stocks. If people are so eager to bet against a stock, that it will go down, that they are paying large amounts of money to borrow the stock so that they can short the stock, and you are long this stock, you better at least be getting the interest – and even with the interest, you should be highly suspicious that this is not a good idea.
[Patrick notes: Interactive Brokers and Schwab have formal opt-in programs to share the interest with you, FWIW. IB splits it 50/50; Schwab is historically more opaque. Risks and conditions apply, etc etc.]
Zvi Mowshowitz: You can use that, and you can do common sense and say, “this is a good company, making good things, that's making reasonable decisions” – these things are just not going to be factored in by the computers fully in my mind.
So you basically make reasonable decisions and then you diversify enough, and that's fine for me – and then for international I have an ETF because like it would be expensive and annoying – but for an average person, of course they should just buy some combination of ETFs and forget about it.
It's a wonderful service. It costs three bips a year or something and yeah, it also costs a few bips a year maybe in these rebalances. They lose a little money every time; it's not gonna be quite as good
But compared to what they used to charge, compared to, what, the costs – the cognitive costs, the error costs of doing things the other way – this is a tenth or a hundredth of what they used to lose. It’s very good.
Patrick McKenzie: Present-day, my children have college funds in a tax-advantaged vehicle in Japan. There exists a fund that I won't name which is wrapping an American index fund.
The index fund charges three basis points and the fund for the service of making it legal to buy it in Japan charges 1.9% – so a good business if you can get it, but obviously injurious to the returns of people who must do investment to the United States economy through Japanese financial institutions.
[Patrick notes: I’m still not going to identify them, but FWIW I looked up a current prospectus to fact check this. Even the Japanese financial industry has gotten better in living memory: the expense ratio is down to 0.67% where the underlying U.S.-domiciled fund is 0.32%. It’s not clear to me from the prospectus whether the Japanese investor also eats that but, well… suffice it to say I’ve never read a Japanese prospectus where a financial institution voluntarily gives away half their margin. Oh also there is an undisclosed fee in both directions on the yen/USD cross.
Do I recommend using Japanese asset managers? Ahem, only to the degree you absolutely have to… which you do, if you’re my kids, attempting to get a tax-advantaged college savings account in Japan.]
Patrick McKenzie: Anyhow, one of the ongoing themes of working in the finance industry and looking at it, there are many mistakes made, to be clear – but in many ways this is where society in the United States broadly decides to allocate the time, attention, and effort of a lot of very, very smart people.
And you get the kind of results that you expect to find when you spend many people's time, attention, and effort on it.
The artifact that sticks out most of my mind for this is, for folks who've seen The Big Short, there's the Jenga scene, which is outlining a strategy for essentially shorting the housing market prior to the 2008 collapse.
There is an actual deck which is available on the Congressional Record, a PowerPoint deck, that was played at the actual meeting that this scene dramatizes. In reading that deck, – the level of analysis about the US housing market, about the trends, the loan underwriting deficiencies, the market in “the sand states” versus the rest of the country, etc. – I was extremely struck. The people who wrote this have observationally a better, more detailed, more rich understanding of what is going on in the housing market than any other sorts I've ever seen with respect to the housing market – be it regulatory reports, or talking with people who are actually in real estate, etc. etc.
[Patrick notes: Due to the U.S. government’s constitutional inability to operate a website the best way to find that deck is Googling the exact title, “Shorting Home Equity Mezzanine Tranches.” [Scribd link, sorry.] It seems to get whackamoled every few years every time I have linked to it in the past. The best link to direct testimony I have handy is here [docx link].]
Covid-19 early predictions
Patrick McKenzie: Sometimes we wander away from our professional interests and look in places where society chooses to invest a little bit less of its time and talent. Do you want to talk about the COVID years, and being in the awkward position that you were one of relatively few people looking at a pandemic as it was happening in real time?
Zvi Mowshowitz: One of the relatively few people who was looking at a pandemic and willing to share the information. I can speak to my time – I can't share details, but there were things in the world where we built models of real things, and our models were better than other people's models of those things – by a lot – than the publicly available ones.
This is known to be a common thing in finance; the housing thing is not a coincidence. In sports, it's the same thing. If you want to know who should start tomorrow's game as the pitcher in a playoff game, if you're not sure who should go on short rest – put up odds under both conditionals (which has been done in the past), see what the price is, and that's how the team should choose.
The team has no idea. The team has some inside information, but that inside information is dwarfed by the amount of insider knowledge that the gamers will have. The gamers watch teams make mistakes and just adjust the odds because they made a bad tactical decision – all the time.
So we go back and we think about COVID: I think there were very good models of COVID at various hedge funds – but that doesn't mean they were telling anyone. (I don't have any specific knowledge. I'm just saying that obviously there were.)
With COVID, I noticed it somewhat in January, understood in February a model of some of the things that were going to happen.
I tell the person I'm working on the game with – mid-February, something like that – I'm like, “this is probably going to be the last time we go into the office, very soon, and we're probably not gonna see each other for a while.” He's looking at me like, “What is this guy talking about? But he's not bluffing. He obviously means it.”
[Patrick notes: In early March, I remember walking past packed crowds in Nakameguro drawn by flower watching on a Sunday, and remarked to Ruriko “Honey, this might be the last normal weekend in a very long time.” It was, at the time, conventional wisdom that Japan had successfully opted out of the covid pandemic. This was not accurate.]
Zvi Mowshowitz: Then I try to explain a bit, but… I'm trying to figure it out, and then I'm trying to make decisions, because everyone has to make important life decisions that are going to impact how these next few years are going to go, and potentially your life and death – they're a really big deal.
I start writing in advance, and I notice I'm really concerned, “what am I going to do?”
I stocked the pantry in advance with a month's worth of stuff in the middle of February because I expect that to happen. Then we have this revelation of, “oh, if me, my wife, and our two kids are stuck in this Stuyvesant Town two bedroom apartment, and can't go outside for weeks or months on end, effectively, my wife predicts crazy. People will go crazy. And the fact that you have prepared for the physical needs of everybody involved to survive what's going to happen is irrelevant. You need to get out of there, get out of there now.”
The thing is, you don't have to only understand what's going to happen, you have to understand what your reaction is going to be and what you can and can't handle.
We move out to Warwick, New York, so we can rent half of the house and have a yard and have space, and we can hopefully stay sane. We just pile all our stuff in and we just move. We get out of there inside of a week, [to] where her parents are right there, so we have some people we can form a pod with – we can do something reasonable.
Then I just spend a lot of my time being terrified of what's happening, watching the news, analyzing the conditions, trying to figure out where it's going to go.
I figure I'll just write my stuff up, what I'm thinking about, and the first few are just warnings and explaining basic information.
There's an Exodus metaphor that I'm still pretty proud of, a Purim thing, but then pretty soon I get into the groove of every week I'm like, “Here are the numbers, here's how it's doing in each region. here's where things are moving, here's the news that I've compiled.”
I start pulling a list on Twitter of people I'm following specifically for this purpose, and I get into a groove, and I'm trying to figure this out – but it feels like nobody else is doing this job, like I'm the only person in the world who is taking this kind of level-headed approach of trying to figure things out.
I make predictions every week – “Here's what I think the next week's numbers are going to look like.” Then I see, “here's what I had, here's what I said last week, here's where I was wrong; here's how I'm updating based on the fact that it missed high, or it missed low, or it missed in this region, or this particular thing happened.” You're looking at depths and cases separately, because there's a lag, so it's different graphs that you're drawing, and different things you're understanding.
Over time I built up a model of this, and I got increasing confidence. I started a debate with Robin Hanson, and then writing up an early thing gave me more confidence people would want to hear it, and be willing to receive it and use it.
One thing leads to another. I'm still making my game, but effectively that's stalled because we can't raise money in the middle of everybody freaking out about a pandemic – everybody's like, “I don't have time for this phone call.”
So instead I'm just writing about COVID all day, and that just keeps going and I get better at it; it keeps being useful for a while and this lasts longer than I or a lot of other people expected – I definitely made some mistakes. I feel like I was doing very well cognitively compared to almost everyone else who was trying to make strong predictions, but there are still some very big misunderstandings.
The biggest one I think a lot of people made, and [which] I definitely made was, I didn't understand the homeostasis aspect of the situation.
I didn't understand how we would be so foolish to get the problem just enough under control that the situation was only terrible, but not try to actually solve it – and also not just take our lumps and get back to get back to business and do something reasonable. We ended up in just this nightmarish middle ground forever.
When we moved out, we signed a lease that was month to month because we figured, “we're going back to the city in a few months,” because obviously this thing will explode, it'll be very, very bad for a few weeks, maybe a few months, and then enough people get infected and that'll be that, because I can do math. I can do various charts and graphs, and this equilibrium scenario just seems so implausible. And then it happened.
In hindsight, I know exactly why it happened, but at the time I just wasn't thinking that way. I came around to this in the summer [of 2020] and figured out what we were going to do. It became clear how we were reacting to various things; we were settled in for the long haul.
Luckily the landlord was happy to have tenants who were nice, so we were there for as long as we needed to be. We were flexible. I just settled in, and I did the thing; I was doing it completely unpaid for most of it, and then after a while, I was just like, “well, I'd like to be able to do this with support, does anybody want to support?”
I was very, very fortunate that some people really appreciated what I was doing, and they reached out and said, “actually, we would!”
They've asked to remain anonymous, which shows you how these things work sometimes – but yeah, an anonymous small group of people are supporting me, and I get to do what I do full time.
Patrick McKenzie: To zoom in on a couple of parts of that, there were contemporaneously with this – and it seems to slip down the memory hole in a lot of ways – but there were “legitimate experts” that made many, many very confident pronouncements about how things were going to go, and the correct policy responses, and similar – and those confident pronouncements were wrong about the future. They were obviously wrong when they were made.
[Patrick notes: As I noted in my conversation with Kelsey, Vox’s now-deleted tweet that there would be no pandemic was on January 31st. And, hey, twenty-something liberal arts majors working for a private equity fund, one shouldn’t necessarily expect them to be experts in epidemiology. But then the CDC and WHO spent months assiduously refusing to call the disease airborne where that was already manifestly obvious, etc etc.]
Patrick McKenzie: The fact that they were made has not quite been sandblasted out of the record, but many people profess to not remember things that were obviously, truly said.
But at the same time as they're being public experts who were advising the governments on this – and many people and firms were taking their cues from the public experts – there were also intense private research efforts done on this, in places like hedge funds and… without giving away any specifically NDA information, it's obviously the case that every large employer had to make a decision on whether to close the office or not, and those decisions were not made randomly.
You could, if you wanted to keep score on decisions on, without loss of generality, AppAmaGooFaceSoft – Apple, Amazon, Facebook, Google, etc. – they have tens of thousands of employees. They would prefer they not die. And they all made various decisions around closing and opening offices.
You can score the timing of those decisions against the pronouncements of society’s legitimately constituted authority with respect to these, and figure out who did a better job of predicting the near future.
I'm very confident about how that actual scoring size would go. I think many people are very confident. The PR departments of those companies would prefer no one actually does that exercise, because embarrassing the people in positions of authority is not plus-EV for them.
But I'd prefer that we as a society do remember that that happened.
[Patrick notes: Compare Google on March 10th versus, I don’t know, the Surgeon General of the United States that week.]
Covid-19 policy failures and blame
Zvi Mowshowitz: Oh yeah, it was horrible epistemics throughout.
I developed the “theory of blame,” that essentially the goal of these people was to avoid being blamed in a two week time horizon for the things that they decided.
But you can have back propagation, so if people concede in advance that six months from now, that's going to be a horrible decision, they would notice that two weeks before that but they can notice they're going to notice it and so on, and so you can get blamed for long term things, but there was no plan.
There was no need for a plan. Everyone was just trying to survive politically and logistically in various ways – they just didn't see that continuing to try and muddle through in this way would be so much more supremely expensive than the two options that actually made any sense, which were to either try and solve the problem and go to COVID zero or so damn near that you could contain it (which was never an option for us really), or acknowledge after a few weeks, or at least a few months that this can't go on, it's just too expensive; we can't make it to the vaccine, not really in this form – it's not worth making it. We're gonna do this, or we're gonna race to the vaccine as fast as possible, and then we're gonna radically change everything.
None of these options were chosen. I think the moment that struck home for me was [when] there was a model of COVID that was getting a lot of play, and was treated as if it was a legitimate thing to think about and model and plan on – and it made absolutely no sense.
I am not an epidemiologist, but I can do math – I can do statistics, I can run charts, and I can look at numbers. I wrote a post titled “Seemingly Popular Covid Model Is Obvious Nonsense,” because it wasn't just that I thought it was wrong, I thought it was obvious nonsense. If you looked at this thing and you thought the future was going to look like that, I have no idea how you came to that conclusion. I just tore it a new one several times over and I got very strong reinforcement of, “yes, yes,” everybody I saw commenting on it, everyone who thought about it, “yes, you are just correct. This is just obvious nonsense.”
[Patrick notes: The essay thoroughly earns that title. Indeed, pointing out the ways it earns it is shooting fish in a barrel, which is why it is extremely useful to point out that a Magic: the Gathering geek correctly and specifically pointed out the failure in real time.]
Zvi Mowshowitz: So much more of what we were doing was just also obvious nonsensical decision making – and what did it cost us? It cost us an extra year of our lives and trillions in basically wasted capital.
Patrick McKenzie: Many things from that period are burned into my memory; one that is particularly burned into my memory is the sudden campaign in the US media and public health circles against the efficacy of masks for containing airborne illnesses.
[Patrick notes: As has been observed before, one of the most striking bits about the early pandemic was credentialed experts very confidently recommending against masking then very confidently recommending mandatory masking. “We have always been at war with Eastasia.”, they said, with this obediently reported by a press corps that couldn’t remember their own articles or tweets from the ancient mist of four weeks prior.
Patrick McKenzie: I'm pretty confident that I've seen masks worn in medical settings before, and I assume that that is not a ritual that is done for no reason. What is the physical basis for this conclusion? Then a couple of months later, the same folks were saying, “no, masks everywhere, we're going to have a public policy around this, strong enforcement, etc.” – what has changed in your world model?
Can you own having made some of those changes for a legitimate reason?
Zvi Mowshowitz: As someone who was masking in March – obviously – I was one of those who fought that good fight.
That was not one of the ones where I felt I was being a genius. That was one of those where I was like, “This is obviously crazy. Everybody I know with half a brain has realized that masks are a good idea here. Everybody who's saying the masks don't work is obviously lying” – and I mean lying, in March.
I think that the regular people who got the impression masks don't work and kept it up, or it became political – I think a lot of them just got the wrong idea and got mistaken. But the people who were saying it in March were obviously lying their asses off, and then when they flipped around and started digging the opposite case – in many cases [in] ways that the evidence clearly didn't justify their confidence levels and their magnitude estimates – it was just, “Oh my god. Now we know who you are. Now we know how you act.”
The question in my mind became, “to what end? What is their ultimate reason for doing this?” I actually have the generous interpretation there, I think the charitable interpretation is correct: that they did this because they thought it would save more lives, that it would have better outcomes for us, and they didn't understand the price they were paying.
Patrick McKenzie: Yeah, I think the most charitable interpretation is, around March 2020, we thought there was a supply chain issue for masks, we wanted them allocated preferentially into the medical system versus being diverted to the population at large, and that required lying. We lied in full knowledge of that fact, and then walked back to lie after the supply chain issues were solved.
That’s a bit gobsmacking me to articulate in as many words because it’s a fundamental contradiction of how I assume democracy should work, but I can at least credit that line of reasoning.
But in a model where we are both a functioning democracy and also adopted that, there would be some owning six months later on, like, “Sometimes propaganda and war requires telling people things that aren't true. We've done that in wars before. We will do it in wars in the future – We fibbed massively. We're going to do it again sometime. Update your models.”
That conversation has not really happened.
Zvi Mowshowitz: I don't think they had any idea the order of magnitude of damage to public trust they were doing when they did this. I don't think they thought about it. I don't think this was top-of-mind.
I don't think they thought about the damage they were doing when they walked it back without admitting they were walking something back, and tried to just memory hole it. You say “when we have a functioning democracy” – I think this is the way we end up with not a functioning democracy, in a lot of ways. It’s because you destroy trusted institutions and then people don't trust a word that comes out of your mouth. Then they make very different choices in various ways in the future, when they need to trust you – because they can't.
It was a horrible set of decisions. I like to think that people doing the kind of role that I'm doing are largely there to point out when this kind of thing happens again.
But yeah, it just made everything so much harder – the vaccine situation if nothing else. That was a huge, huge problem.
Vaccine rollout chaos
Patrick McKenzie: Yeah. On the topic of ‘things that there were no plans for,’ it wasn't strictly speaking that there was no plan in 2020 for rolling out the vaccine eventually, but it turned out that that plan was vastly underbaked relative to the societal importance of it.
Zvi Mowshowitz: The plan seems to be less baked than the plan that was formed by the person I'm talking to right now spontaneously at the time.
[Patrick notes: A few geeks, knowing little of medical delivery, piled into Discord one evening and whiteboard out an operational plan for successfully finding vaccine doses statewide without requiring explicit cooperation of an agency. That plan basically worked. Zvi has written about VaccinateCA before, more focused on the meta takeaways than my writeup. I’ve also discussed it with Kelsey and with Dwarkesh Patel in recorded appearances.]
Patrick McKenzie: So yeah… (both laugh) For those of you who don't know: I have no medical or epidemiological background. The vaccine arrived in December of 2020; in early January of 2021, the state of California was reporting to the federal government that it was successfully injecting 25% of the strictly limited number of doses that it was receiving, and that sounded a little bit crazy.
There was widespread reporting in the media that people could not find the physical locations they would need to go to to find those three out of four doses that were not getting injected, and they were making dozens of calls in a row to find a hospital or doctor that had it available.
I thought, “hmm, plausibly there's at least one person in the state of California who's capable of making a website which will list names and addresses of the places where the vaccine is available. Perhaps someone should do that.” I tweeted this out on Twitter and Karl Yang, who follows me on Twitter, invited 10 people into a Discord to start hacking on that website.
I'll elide the long story here, but we effectively became something that was very close to “the public-private clearinghouse for vaccine location information in the United States” because in ‘the plan,’ knowing where the vaccine was was not a priority and so not ordered and not done – so we bluntly decided to do it.
Zvi Mowshowitz: None of the things that they did with the vaccine rollout were the things you would do if your goal was to roll out a vaccine as quickly as possible to as many people as possible, and ideally to the right people – but mostly to some people – nor were they the one that would allocate it in any efficient fashion. They just were very, very worried about other things, and really bad at their jobs.
Patrick McKenzie: We made gestures in the direction of allocating things efficiently, and made a couple of decisions which are better than the worst of all possible decisions. The decision to prioritize elderly people first is a genuinely medically good decision and uncontroversial.
Then there was a massive politically inflected spoil system around the vaccine, which – the administration of that spoil system and strict adherence to it outcompeted administration of the vaccine as a goal at the margin, consistently for many months, in a way that killed many, many thousands and tens of thousands of people.
Zvi Mowshowitz: It also just crippled the economy in very real ways. I remember trying to figure out the logistics of, “What is the first possible day I can, without violating ethical principles, get my hands on a shot, and then get my hands on a second shot, and then have my life back? The moment me and my wife have our shots, we can have our lives back.”
We spent a lot of time and effort on that, and then a lot of time and effort also managing this process manually for a bunch of other people we knew, including her parents. It was a nightmare.
Patrick McKenzie: Many people were in the position that they were managing care for others. I was likewise trying to assist my parents, who were in Illinois at the time, in getting the vaccine.
I flew over to California to run VaccinateCA, which was originally California-centered and then covered the entire United States. A thing we learned early was that pharmacies got a huge allocation of the vaccine relative to all the medical care providers, and if you were approaching the system from a position of no knowledge, you should attempt to get it from a pharmacy.
The few people who figured that out would go to the first pharmacies that would come to mind when they thought of the word pharmacy, which tends to be one of the two largest ones in the United States – so the number of seekers versus number of doses for those two pharmacies was grossly out of whack.
Relatively few people continued calling to the third largest pharmacy in their town. And because of the foreseen state reasons, the third largest pharmacy got the third largest allocation. There was a grossly disproportionate – in the other direction – imbalance with respect to seekers at the largest pharmacy.
Once we figured this out for California, I immediately phoned my dad and said, “Dad, you've been trying to get the COVID vaccine for weeks and have gotten turned away many times at the pharmacy that you usually go to; who's the third largest pharmacy chain in Chicago?”
He knew that off the top of his head and called them. He calls me back 30 minutes later, he's like, “mom and I have an appointment now.” I'm like, “how is this not on the nightly news every night until the situation is resolved?”
[Patrick notes: The vaccines needed a halfway competent marketing department, and did not have that, despite being the most anticipated product launch in the history of the world.]
Patrick McKenzie: We tried to get out that message at the margins too, get it into all the places… it was not a priority for the federal government that people would be able to Google where the vaccine was – for a combination of understandable and absolutely maddening reasons, one maddening reason being that various political actors wanted credit for the work that “their people” were doing, versus merely solving for ‘getting as many Americans vaccinated as quickly as possible’ – so “ship the information onto Google” was not a north star of project plans despite this
Zvi Mowshowitz: Right – Google being down for helping. I'm sure if they had been asked nicely and told they would be looked upon well for, for helping –
Patrick McKenzie: I will not speculate as to the contents of any calls made with Google, because one has to credibly promise people to keep confidences, but…
Zvi Mowshowitz: (I mean, I can speculate but you can’t.)
The biggest thing about this to me as an update – because I'd already figured out the government was dysfunctional in roughly these ways, although this did exceed my expectations for dysfunction: the inability of almost everybody to be able to navigate this information environment, and the inability of any sources to figure out this was the situation and then share that information with other people, that kind of gobsmacked me.
I think this is one of the places where I didn't do as good a job as I should have. I should have been harping much louder on the question of “how do you personally get your hands on the vaccine as quickly as possible?” Because if you can get your hands on the vaccine and that vaccine is just sitting there and not being used, you should go out there and get it and then tell your friends and tell your family and tell everybody in your neighborhood until there are no more doses left.
I have a process that's basically, I imbibe information from various sources and I curate my feeds carefully to make sure that this gives me a good mix of information and I'm not going to miss something, and then I respond to it. That process is less good at understanding when you need to be very proactive about a specific thing, and you need to go out and do the thing yourself.
Somehow even though I understood that – vaccine distribution was all I was talking about – this is really important, I just didn't focus anywhere near enough on this question.
I think that's probably the biggest failure I have after early 2020 in the COVID coverage. I think the first few months, I made some cognitive errors that were definitely expensive errors to make (even though I did reasonably well) but yeah – I had the right idea, the right overall picture, but I just did not hammer the thing that mattered anywhere near enough.
That's a classic mistake: you understand what's important, and you invest in it, but you should have invested so much more in it. It's kind of the story of AI, where people are now understanding, “no, just add extra zeros to everything” – extra zeros are better.
Patrick McKenzie: That brings up a classic blog post of yours – but just to complete the thought on regrets: I spent the later half of 2020 trying to… I dealt with some depression in the wake of the pandemic – trying to get my own house in order, to get back to being productive at work, etc. I took my ball off the global situation, national situation.
One of the great regrets in my life is, if I had thought seriously in October of 2020 about what the vaccine distribution plan was in the United States – and what the built information architecture was around that – I would have been able to confidently predict the manifest failures that we saw in in December and January and could have done something about it.
That sounds a little bit gobsmacking, that a person who was working in a marketing department in central Japan at the moment feels bad that they did not single handedly overturn the United States policy environment with respect to distribution of the COVID vaccine – but based on realized results, I think that is actually something that was possibly in the solution set, if I had figured it out three months earlier.
I kicked myself a little bit for that one.
[Patrick notes: Meaning, explicitly, I could have gotten Google to have a responsive answer to queries about the covid vaccine shipped in time for there being supply ready to show in response to that query. Your prior on that might be low, but you should adjust it far upwards when you realize VaccinateCA successfully got Google to do that a few months later. The difference in those two outcomes is rather stark because of how many more lives 10,000 doses saves in January versus in March due to relative vulnerability of eligible population, progression of the pandemic, etc.]
Zvi Mowshowitz: I'm not beating myself up over my failure to realize the same thing. I do think that if I had thought carefully about what the government is likely to do, I would have put a very high probability on this being botched, and would have realized that covering in case that did happen in vaguely this way just had much, much higher EV than anything else I was possibly doing with my time, including writing the bulk of the COVID posts.
You were just a random person (in some sense) in a marketing department and you made it work. I don't think there's any particular reason any number of other people couldn't have also made efforts – and also if five people had separately tried it and two of them had worked, that's one of the best possible situations that you’ll see.
[Patrick notes: To be clear, there were many sibling efforts, some directly inspired by us and some not. One of the takeaways that has received the least widespread recognition: there was a repeated user behavior of people organizing on Facebook into local vaccine hunter groups, which swapped up-to-date location information and also had members of the professional-managerial class call into pharmacies/county health departments/etc on behalf of people who can’t successfully operate a bureaucracy but can successfully operate Facebook.]
Zvi Mowshowitz: I just didn't think ahead to the problem. It just didn't come up to me on my radar. I guess that's actually, now that we talk about it that way, a good lesson to be like, “I don't necessarily think ahead to what the problems [are that] I'm going to inevitably be in the future as much as I probably should – and maybe you can actually intervene on this now?” I may do some [thinking ahead about those problems], but just not concretely enough. That’s a really good note.
[But] given you are the person who did the most compared to your positioning to fix this problem, it's kind of the trader’s lament. The trader’s lament is that you are never ever happy with your decision on the trade, because either you made money, in which case you didn't do enough, or you lost money, in which case you're sad you did the trade.
I mean, it's not that simple, but fundamentally speaking, it's very rare that you come out and you say, “I had interesting decisions to make and I got exactly the right amount of this.” Sometimes you do – sometimes you feel like “I played that really well,” especially if you've been doing the same trade for weeks or months, and you've gotten the idea of how much can be supported and how to respond in terms of sizing – but that sizing or timing of the response kills you.
Same thing with any exponential; COVID's an exponential situation, AI's an exponential situation. You only have two ways to respond to an exponential: you either respond too early or you respond too late. So if you respond to the COVID vaccine failure too late, at least you're responding. That's still something.
You can respond too early and that's the ideal here, but you can't beat yourself up over not doing that in this situation.
The importance of scaling effective strategies
Patrick McKenzie: Ah, morale management – a challenging problem.
Speaking of response curves, you’ve returned many times over the years to the concept of More Dakka. The title encapsulates both what I love about our weird corner of the internet and a little bit of what frustrates me about our weird corner of the internet: it’s a concise reference – by way of a second-order reference to Warhammer 40k – to a concept which is really, really useful for people.
Can you unpack More Dakka?
Zvi Mowshowitz: Dakka essentially is ammunition – it’s like “more firepower.”
The idea is, so much in our lives, we kind of know what the answers are. We know the things that are improving matters, but then when we notice a good response, we don't then scale up what we're doing. We [might] do a little bit of it, and then we're like, “oh, that was helpful, and then we don't realize we can just keep going.
One classic example of that that I used in the post was gratitude journals. Everybody knows who looks at this – it's just a universal result – if you just physically buy a notebook, and you just write down every day, “here are some things I am grateful for that happened today, and here are some people I would like to thank,” and [then] just file the journal away, you will be happier.
The more often you do this, the happier you will be – and this is vastly more efficient, and also almost free, compared to all the other things people do to try and be happy.
And yet almost no one does it.
I don’t do it either. I will sort of do the mental step version of it, and that is helpful, it's incredibly helpful – and the step where you email people and you thank them is incredibly helpful.
It’s one of the few things that actively has the potential to make you happier in general.
Think about the people in your life who you legitimately should be grateful were in your life – even if they're not in your life anymore – just send them an email like, “thank you,” and specifically, “here's why you made my life better.” That's great. Things get better. No one does enough of this.
Then there are so many other things where it's like, “okay, I’ve tried doing this 10 times, but I haven't tried doing this a hundred times.” So often it is just keep doing it until it stops being better to do more – it is a very, very good principle.
Patrick McKenzie: This has wide applicability in so many domains. I've seen it up close and personal.
I ran small software businesses for many years, consulted many other software businesses, and worked in marketing departments.
The classic thing is, if you have ten channels – email, SEO ads on Google, etc. – and you explore those ten channels, one of them will work out really, really well, two of them will work out “eh, we're glad we did that,” and then you'll have seven failures.
What the marketing department or the person in charge will often do is like, “Okay, we've run ten experiments, we've gotten seven failures, so let's run ten new experiments.” The thing that very rarely gets said is, “The thing that works out really, really well… should we do a lot more of that?”
“Should we do that – not necessarily to the exclusion of all other things – but should there be someone specifically in charge? If SEO has been the magic bullet for us, should we have The SEO Person that spends all of their time on SEO and makes sure that we have an update about the new SEO things every Friday?”
Zvi Mowshowitz: And our next 10 experiments – maybe eight or nine of them should be variants on SEO and just tweaking something a little bit, or pouring more money into it, just gathering more information because that's where the money is. Exactly.
Patrick McKenzie: Many people are familiar with the efficient market hypothesis; I think a lot of us have kind of a shadow version of it in our head, like, large, well resourced organizations must be making rational decisions all the time. And –
Zvi Mowshowitz: No! (both laugh)
Patrick McKenzie: False. This is false. If you understand one thing about the world, understand that a collection of very smart, extremely well resourced people can be making absolutely terrible decisions, and know that while they are making them, and that state can continue for years on end.
If you look at some of the largest, most sophisticated marketing shops in the world, you will see this pattern over and over and over again: “Yeah, we have the one thing that makes all the money. How much do we invest into that one thing versus all the other things?”
“Eh, It's 5 percent of the budget.”
“Have we considered changing that?”
“Well, we've debated it a few times, but for various institutional reasons, we have not pulled that trigger” – and then if you say, “this is insane, let's fix it,” you have at best a 20% chance of actually motivating behavior in a way that matters, if you devote the next six months of your life to it.
Zvi Mowshowitz: The most maddening version of this is when people figure out the right thing to do – which often just involves throwing money at the problem in a reasonably intelligent way – and they do a little of it, and it has a small effect, and if you do the math, this implies that you could in fact solve the problem by just scaling this up.
But right now that didn't do very much – and then they say, “that didn't work,” and they dismiss it as a failure.
To me the canonical version of this, whether or not you approve of the goal – I think it's very important – is the fertility problem.
Right now, in many countries, there are almost no children. You have South Korea – now they have 0.6 children per woman [Patrick notes: perhaps you care about second significant digit but this is one of those times where needing to care about second significant digit is very bad news], and that means there will very rapidly be very very small numbers of Koreans versus what there are now. This is very very bad for their country in various ways (even if you don't just value the idea of there being people around).
You run these experiments and people say, “okay, we gave new mothers $2,000 and then, yeah, the number jumped a little bit, but it didn't seem to last that much compared to the other things that are going on.”
It's like, okay, you're telling me that they're having babies that cost, in the US, $270,000 to raise, and have all these other costs and lifestyle transformations. You tried paying them $2,000, and you saw a small effect, and you gave up. This is just obvious absurdity. Have you tried thinking about $20,000? Have you tried thinking about $200,000? Even if you're not going to do it now, you don't think it's worth it, just figure out whether or not this would work, and then you can have an actual discussion about whether or not it makes sense.
Patrick McKenzie: You can do the experiment on the micro scale for an amount of money that the state routinely spends on toilet paper in government offices – offer an absurd amount of money to the lucky cohort of 50 people, and just plot out the response curve and see.
Zvi Mowshowitz: The cultural context is different, so you are going to have a very corrupted experiment, but you just get so much more information if you just try different numbers and see what happens.
As a gamer you learn to do this, as a gambler you learn to do this, as a trader you learn to do this, and as a rationalist you learn to do this – and in AI a lot of people are indeed learning to do this.
But at the same time, so many people see this new AI tool and they say, “huh! that's kind of cute. This does some cool new things.” And then they just kind of forget about it. And they never think about what would happen if they invested somewhat more resources into figuring out how to make this improve their workflow, figure out how to learn things better.
This alien just showed up on your doorstep that knows all the things and will explain them to you, and respond to whatever you want.
You are so bizarrely incurious. You asked it to write three poems and then you had some laughs and put it away. What's wrong with you?
Patrick McKenzie: I think this is a story of our lives for writers as well. I've written about 5 million words or so and people ask me for writing advice. I often tell them, “however much you think you want to write right now, that's kind of homeopathic relative to the amount you'll write if you become a good writer.
“Can you just decide to write a lot more right now, about anything? Any topic, any format, any cadence, whatever works for you, just write 10 times more than you think you need to, and then 10 times more than that.” Visakan on Twitter has a good articulation, “do a hundred things – the same thing a hundred times. Write a hundred blog posts, record a hundred videos, etc.”
Zvi Mowshowitz: Teller, of Penn and Teller: the definition of magic is often just someone who put way, way, way more time and attention into this thing that anyone thought anyone would.
It’s just that simple.
Yeah, you get good at it. I was a professional magic player, and the periods where I was really, really good were when I just had the gumption: I sat down, and I'm just like, “I am going to practice this. I am going to train this over and over and over again. I am going to get really, really good.”
The first time I made a pro tour top eight, I was literally in class, in a lecture, and my deck is in my hand; I'm shuffling the whole lecture, and I'm taking test draws, and I'm playing them out, and I'm thinking about the sequencing, and I'm thinking about the details.
It just doesn't stop. It doesn't. It would seem like maximum overkill to most other people, but it just gets better. You just keep going, get that little extra edge.
Patrick McKenzie: I think the way that this becomes an edge, and [that] it's replicable in so many domains, is one reason why the current shape of the world is in some ways more hospitable for people who are wired a little bit differently, than it has been for much of history.
But neither here nor there. I think both of us are probably wired differently across some spectrums if you can measure it, whether you want to put a label on it or not – sort of, people who will happily do enough of something that the rest of society says is clearly too much time spent reading about financial infrastructure.
Zvi Mowshowitz: Yeah, on another podcast a few days ago I heard the expression, “the future belongs to the obsessed,” or something like that. That just seems totally true. You don't have to be obsessed with the same thing all your life, but you have to be able to be obsessed with it today, and you need to be able to understand when it's time to be obsessed with something until you got what you wanted out of it. That is one of the most underrated skills.
Patrick McKenzie: Yeah. Cultivating the ability to be obsessed about the right things and to continue that until it's no longer useful.
One of my worries about this, as someone who has spent more than a little bit of their life playing video games, is that the world and actors within it are very good at giving you synthetic things to obsess over where that is maybe not the optimal thing to do for large periods of your life.
If I could just get those years running the World of Warcraft raid guild back, that'd be awesome.
Zvi Mowshowitz: Don’t be so sure! That might have been a much better education than a lot of other things, because it taught you so many valuable skills on so many levels, including how to be obsessed with something, and care about the details, and work through it, and make it happen.
I think that a lot of the stuff we do, the most valuable part of it is the skills you develop along the way and the things you learn – obviously the trap is, you learn video game skills that only apply inside video games and then you learn to play more video games and then that's a disaster if you take it too far.
But yeah, I have three kids, and when I watch my kids play video games, as long as they are progressing, as long as they are trying to get better, as long as they are trying to figure things out, as long as they are doing some form of deliberate practice in any way, I am pretty much thrilled.
I think this is one of the best things they could be doing with their time if I don't have a great social activity or something lined up for them right now – whereas when they're just in Minecraft doing the same thing over and over again because it amuses them, I need to take that device away.
Patrick McKenzie: I think one of the positive changes of the world is…
A reason why I found myself shuffled into running a World of Warcraft raid guild – versus having 60 direct reports in some organization that did things more important than killing virtual dragons – was because I was in a weird place in the economy and was not on a management track anywhere, et cetera, et cetera.
One of the iterations about the world in startupdom: you can take people who are on paper not ready for the management track – they haven't paid their dues, they don't have an MBA, et cetera, et cetera – and just saying, “okay, start with managing a team of four people, and maybe you're good at that, maybe you're not. If you are good at it, we will increase the n until we find a level at which you are no longer productive for the economy.”
That sort of sorting function has identified very, very stellar people early in their career relative to like the traditional track for, “Okay, we're going to apprentice you for a few years, you'll get an MBA and then you will be successfully apprenticed through a variety of increasingly larger divisions until you eventually manage 250 people in your 50s.”
So [these are] interesting ways that the world is reconfiguring itself to take advantage of what we are learning (in fits and spurts) about how people cooperate together, how that is mediated with software, how to deploy brains more effectively.
AI predictions
Patrick McKenzie: I wouldn't want to end without talking about AI a little bit. I think we come from slightly different perspectives on it.
Patrick McKenzie: Assuming an audience that is aware that something has been happening a little bit for the last two years, but might not be super up on the details, what’s the one paragraph version of, “if you understand nothing else about where we are, understand this”?
Zvi Mowshowitz: A handful of AI labs are spending exponentially more resources to build smarter and more capable language and other models that are getting rapidly more powerful, more capable, more useful – that are already highly useful now and that you should be using on a regular basis to make your life better – and we don't know when this process is going to hit a wall and stop being economical and effective, and able to pump more intelligence into the system.
If it doesn't stop soon, it’s going to, at minimum, transform the economy and everyone's lives in ways that are very difficult to anticipate and predict – that might be very good, very bad, or something in the middle – and potentially poses an existential risk to the existence of humanity or anything that most of humanity actually cares about. [It] threatens to spiral out of control essentially, because if you have a bunch of computerized virtual researchers, essentially, that are as good as the current researchers in AI, and then they make smarter AIs and smarter assistants than them and work so on, this can get out of control very quickly.
All of this is on a very, very rapid exponential. Algorithmic improvements alone are meaning that this year it costs several times less resources to do the same thing, the same way, as it did last year.
Again, we don't know whether the end is in sight, and we have to solve a lot of very difficult problems if we are going to get through this and have a world that has value, that we get to live in and get benefit from – and that seems really hard.
And in the meantime, there's this amazing tool around and you shouldn't miss out.
Patrick McKenzie: Yeah. I come at it from a very slightly different angle with respect to the existential risk part of the equation.
[Patrick notes: Here my salaryman communication style where I autominimize disagreements and use ironic understatement may bite me in the hindquarters. My p(doom) is somewhere around p(nuclear winter), but I have spent a few year marinating with the arguments by people much more concerned than I am, which is why my p(doom) is not p(orcas hunt humans to extinction). Zvi’s is higher.]
Patrick McKenzie: Way back in the day I had an undergrad concentration in AI back in 2004 and it's essentially a different technology compared to what we're using these days.
I think it's important to understand both the broad shape of this and also understand individual technical artifacts, because the technical artifacts – the sum over all of them is the broad shape of things.
GPT 2 came out in, let's see, 2020, I think? [Patrick notes: 2019.] I saw GPT-2, and thought, “wow, I've been involved on the internet around various scaled approaches to creating relatively low-value content, and this is going to replace that form of economic activity within a couple of months. If you want to have a plausible recipe for blueberry pie, this will kick out an infinite number of plausible recipes for blueberry pie.
An interesting thing about blueberry pie recipes on the internet is that it is not necessarily the case that a blueberry pie recipe on the internet has to actually physically produce the result of a blueberry pie in real life for that to be an economically rational blueberry pie recipe for the person writing it.
Be that as it may, it was obvious that it could kind of ramble in a semi-coherent fashion on a variety of things.
Then a year, year and a half later, GPT-3 came out. Since then we've had 3.5, and 4.
I think people who are not steeped in this topic should be amazed the first time that they experience 3.5 and 4.
To the extent that they're not amazed, that's a, “fix this in your world model” – particularly because, even if you think we're on the down-slope of, “10x-ing the amount of resources to get to the next model might not cause the the same qualitative gap in performance that we saw between [the previous iterations of] these things,” it implies that whatever the next model is, is going to be very, very impressive. And the status quo is already very, very impressive.
We use a bunch of words that kind of obscure the truth, use jargon; people don't understand what “one-shotting an exam” means. The thing I tell less-technical people is, “you have something that is functionally about as smart as an undergraduate and capable of doing most things that an undergraduate can do relevantly in intellectual domains – and which makes lots of mistakes which, if you've met undergraduates before, you might be familiar with.
“And unlike human undergraduates, it's very, very responsive to correcting when you point out mistakes – and also has infinite capacity for working through the night and turning around term papers in 300 milliseconds. It's very clearly the most important technological advance that we've seen in our lives.”
(...Hmm, okay, we’ve seen the Internet too.)
[Patrick notes: As I remarked to Tyler Cowen, I think the Internet is the Great Work of humanity.]
Zvi Mowshowitz: I think of it this way: the downside of AI, if AI basically doesn't get much better than it is – and I mean downside in terms of how big a deal it is, not in terms of how bad it could be. – the pessimistic case is what I would call “only internet-big.” That's a good way to think about it. This is going to be as big a deal as the internet, if it disappoints us as much as it could realistically disappoint us. That is the downside case for impact.
So be prepared for something like that, and you should try and get in on this at the ground floor in terms of understanding it, and preparing (at least psychologically), and have an outlook for it. This should absolutely impact your plans – and again, none of this involves any of the existential risk questions or even the possibility these things would get vastly vastly better; just the current generation, one more generation that maybe isn't that much better, and years for this to cycle through the economy, people to figure out what to do with it, people to build programs around it to use it better, people hooking it up to tools, people hooking it up to your personal contacts and resources, people putting it in an assistant on your phone – so this is gonna power your version of Siri or ‘Okay Google’ or anything like that – very soon.
It’s going to take a bunch of jobs. There will be other jobs that people will do instead. It's going to make a lot of people a lot more productive. It's going to transform the way we do all sorts of things, certainly the way we handle information, the way we handle anything digital is going to change radically. This stuff is baked in. This is just going to happen.
The question is, “does it then go the next level beyond that?”
Patrick McKenzie: Yeah. For relatively intelligent, relatively sophisticated people who are looking at this and looking at the centers of the probability distribution of that impact, I would agree that the most reasonable, least impressive outcome is, “oh, it's only as big as the Internet is” – and as someone who says that the internet is the most important thing that human race has ever created, there's a reasonable case to be made that AI is only as important as the Internet is.
[Patrick notes: To refine what I am saying here, assume I’m saying “Only as big as the typical NYT reader thinks the Internet is.”]
Patrick McKenzie: There are also reasonable cases to be made that we are in the first years of something that is approximately as societally transformative as the industrial revolution.
I'm very happy we went through the industrial revolution. I'm also very happy I'm living hundreds of years after the industrial revolution, both because we get the benefits of it, and because we avoided a very tumultuous time in history.
Zvi Mowshowitz: That one seems like it worked out.
Is this only Internet-big? Is it only industrial revolution-big? Is it only humanity-big? Is it only solar system-big? Is it going to transform the lightcone, in ways that could be anything?
That's the range that we're looking at. This is going to be a huge deal, unless our civilization just collapses for other reasons before it has a chance – but that's the only way this isn't going to be a gigantic deal, as far as I can tell.
Patrick McKenzie: I think people will feel like, “okay, I've heard this before from VCs et cetera, ‘Bitcoin is gonna be the next big thing.’”
That was me, hating on Bitcoin because I'm a bit of a crypto skeptic – I never really bought the pitch that it was going to be the next internet. I did not buy the pitch that mobile computing was going to be about as big as the computer was; I think I was directionally wrong on that, while being directionally right on cryptocurrency.
Zvi Mowshowitz: I mean, Bitcoin is still trading at $68,000.
Patrick McKenzie: Yeah, it has a price, but if we look around in our lives every day and look at the impact of the iPhone versus the impact of Bitcoin: Bitcoin is still trading at $68,000; the iPhone mattered a lot to humanity and Bitcoin descriptively has not.
Zvi Mowshowitz: Agreed.
Patrick McKenzie: I would bet everything I had and and borrow more to bet against the outcome that AI is only as big as cryptocurrency in terms of realized impact on human life.
I think that seems like a reasonable place to wrap things up.
You continue writing as an independent researcher – do you want to tell people where they can find you on the internet?
Zvi Mowshowitz: I write at TheZvi: thezvi.substack.com.
I put out newsletter updates on Thursdays, and then I try to post between one and three other things during the week, depending on what's happening and what naturally forms its own thing; I try to do a monthly of non-AI, non-other themed topics and otherwise keep up with things. That's my job, that's essentially what I do all day.
I also have a 501c3, Balsa Research, that is currently working on potentially repealing the Jones Act.
(There’s a very long story as to how that came to be something I was focusing some of my attention on, but it is like the platonic ideal of a completely terrible law – that both in and of itself it would be incredibly valuable, and would be this gigantic civilizational morale boost and enabler, if we could get rid of it. I'm excited by the opportunity to try it, because I think the resistance could be remarkably low.)
(A reasonably small amount of effort and money could potentially give us a single-digit chance of making this work, maybe even double-digit. Not promising anyone victory there, but you never know.)
Mainly I write about AI. I try to figure this stuff out; I try to have people think better about it, and try to move us in directions that will enable us to navigate this critical period and minimize the chance that all value in the universe is lost or all the humans die or lose control.
Patrick McKenzie: Thanks very much for your time today Zvi, and thanks very much everyone for your time and attention. I'm Patrick McKenzie, patio11 on most places on the internet, including Twitter (which will always be Twitter to me). Most of my writing is at bitsaboutmoney.com these days. I also have a blog with a terrible domain name that I should eventually rip off the band aid and replace – but one thing at a time.
Thanks very much. See you around the Internets.
Zvi Mowshowitz: Absolutely. See you around.