slashing away parts of their humanity

One of the best reasons to write online, which I hadn’t anticipated when I started, was to “find the others”. The people who make you feel less alone in your thoughts. The ones taking the same crazy pills as you, whose minds wander the same alleys.

David Fu is one of those people for me. We bump into each other at the intersections of education, games, and the type of idealism we should have long outgrown. He pings me last week with an email subject:

watched this pod and thought of mathlete vs mathematician

The subject is a callback to Benedict’s reflection math team and other horrible things you do to get into stanford and his message was a paraphrase of an interview he watched:

If we live in a world that hyper awards those with power and those willing to cut out their humanity and hyperfocus on the explicit measure (win math competitions, win status and prestige games like getting into Stanford) then we should expect to get a world in which the people who have the most power to communicate and control things are the people who have been most willing to slash out their humanity and hyperfocus

Sociopaths are running the asylum and I’m still stuck on how we got here, so ok, you got my attention. I open the YouTube link.

Dork f’n Christmas.

It’s an interview with C. Thi Nguyen!

Nguyen is a philosophy professor at the University of Utah with a focus on games. Not game theory but theorizing on games. His first book Games: Agency as Art (my notes) is amazing, but I don’t recommend it unless you are in the market for an academic treatise on the philosophy of games.

Nguyen’s interview with Sean Carroll 5 years ago is still one of my favorites (my notes), so I was stoked to settle in to this one.

It delivers.

Turns out he has a new book and if that wasn’t self-recommending enough, I was delighted to see Dan Davies write:

If you liked “The Unaccountability Machine” and you got a book token for Christmas, spend it on “The Score” by C Thi Nguyen. I haven’t seen an advance copy or anything, but I met the author at a conference and he’s extremely funny and clever. I will be trying to promote the book to as many business podcasts as possible, because I very much think that this is a case in which the dead hand of academic analytical philosophy may have robbed the world of one of its greatest management consultants.

Dan’s another author whose powers of observation are galactic, so I’m getting a lot of convergence on healthy brain food.

I’ll leave you with some of my favorite excerpts:

Hobbes, Power, and Defining Reality

What Hobbes says is that the ultimate power is not military or economic power. It is the power to define terms and control language. If you control language and define terms, you control people from the inside.

Games are like artificial governments. They’re things where we play around with incentives and rules and constraints to shape people’s action.

When someone says, ‘Here’s a watch. It measures your health,’ and if you accept that, you’re letting somebody else define what health means for you—typically in what’s easy for one of their devices to measure cheaply.

Value Capture (Core Concept)

Value capture is any case where your values are rich or subtle or in the process of developing that way and you get put in an institutional setting and that institutional setting offers you a simplified—typically quantified—version of that value, and then that simplified version starts to take over your conception of good…It’s not just an incentive. It’s when you start to care about the metric as the thing itself.

Law School Rankings and the Death of Deliberation

Before rankings, colleges described their missions in totally different languages—money, research, activism, community. Students had to deliberate about what they cared about.

The moment U.S. News and World Report started issuing rankings, students stopped talking about what they wanted and started assuming that ‘best’ was whatever the ranking said.

You’re outsourcing not just your values, but the process of deliberation about your values.

Bernard Suits and what a game actually is

Bernard Suits defines a game as the voluntary attempt to overcome unnecessary obstacles. In ordinary practical life, the outcome is valuable on its own, so you try to get it as efficiently as possible. But in games, the value is inseparable from the obstacles themselves.

If the goal of basketball were just to pass the ball through the net, you’d use a ladder at night with no opponents. But that’s stupid — because the value of making a basket is intrinsically tied to dribbling, jumping, shooting, and resistance. Whatever the value is, it’s in the process, not the outcome… And when you hyper-optimize for winning, you destroy the spirit of the game.

Why metrics are fun in games — and dangerous elsewhere

The articulation of a metric has clarity, and that clarity is incredibly powerful over us. The best way to talk about this is to explain what metrics do for us in games — and then why that’s okay in games and not okay elsewhere… Games are this incredibly interesting art form where someone designs a new self for you. A game designer tells you what to want, how to pursue it, and what constraints you have, and suddenly you become a different kind of being — a being of feet, or balance, or precision.

What makes games so beautiful is their simplicity and clarity. We all have the same goal. It’s blissfully clear what a good move is… But that same clarity lets us outsource a complicated judgment about ourselves. The system says: I will let you know when you’re doing great. And the danger is that the clarity of a metric can keep us playing even when it makes us miserable.

Goal vs purpose

One of the most important distinctions in thinking about games is the difference between the goal of a game and your purpose in playing the game. The goal is the target you’re trying to hit. The purpose is why you play at all… For some people, the goal and the purpose collapse into one thing — winning. I call these achievement players. But for a lot of us, there’s a deep difference.

The easiest place to see this is party games. In party games, the goal is to win, but the purpose is to have fun. You have to try to win for the game to work, but if everyone had a great time and you lost, you’d be ridiculous to feel bad about it… The larger purpose has clearly been fulfilled.

Rock climbing is another good example for me. I love rock climbing. I am a terrible rock climber. I am mediocre beyond belief… The goal of rock climbing is to get up the rock. But the purpose, for me, is the beauty of movement and the clarity of mind it gives me. It’s one of the only things that actually gets my brain to shut up.

What’s interesting is that I cannot get that feeling without trying to win — without genuinely trying to complete the climb. But it also doesn’t matter if I fail all day. I leave feeling good. My body feels good. My mind feels cleansed.

When you have the right attitude toward games, you keep goal and purpose separate. The game tells you the goal. You choose the game for your own purposes… And that separation is a huge dimension of freedom that we often don’t have with metrics.

The gap between what matters and what’s measurable

The thing to get really interested in is the gap between what’s really important and what’s easy to measure institutionally. ‘Likes’ claim to represent communicative value. Steps and VO₂ max claim to represent health. And when those representations are too thin, they don’t just miss what matters — they actively change it.

Qualitative understanding is rich, subtle, and context-sensitive, but it travels badly. Quantitative knowledge works by isolating a context-invariant kernel — something everyone can understand — and stripping away nuance so it can move easily between institutions. The problem isn’t that data is bad. It’s that we reach for it compulsively, even when it’s inappropriate.

Recipes, accessibility, and the loss of judgment, and the “facade of objectivity”

The facade of objectivity as well that this folds into. It’s this notion that we are also being sold this actively sold this by tech companies as well as our governments as if this metric data-driven system is also democratizing, it is populist it is access expanding… but what you’re describing is the thing that I think I have felt and have been frustrated by which is the decline in the value of expertise of editorial judgement of human decision

Old recipes don’t say “two cups of flour.” They say things like “add water until it feels just under sticky.” That’s actually a very good recipe — if you know what you’re doing… Modern recipes give you accessibility. Anyone can follow them. But what they take away is the cue to adapt and use judgment.

Metrics are that for values. They tell you something anyone can use and understand. And accessibility isn’t bad — but there’s a price. And the price is expertise.

When data is genuinely good — and when it turns on us

Large-scale data is really good at optimizing for things that are easy to count. It’s why I am alive and why my child is alive. Not dying of an asthma attack is a very clear target, and data is incredibly good at that.

It’s also incredibly good at debiasing. If an institution is convinced it isn’t biased, numbers are often the only thing that can knock the door down. You can point and say: look, women are getting the same scores, but you’re hiring men nine times as often… That clarity is powerful.

But then you move decades forward. And what you start to see is those same data-driven approaches getting thinned down into numerical quotas and proxies that miss the heart of what they were meant to fix… The system that was good at breaking down the door becomes the system everyone optimizes against.

Data-based approaches are very good at the blunt stuff. And then, over time, they tend to miss the subtle stuff — while capturing everyone’s attention. People start gaming the metric. Institutions start optimizing the proxy. And the thing that actually mattered quietly slips out of view.

Metrics are best at targeting what everyone can count easily together. And the uncomfortable question is how much of what makes life meaningful is actually easy to count together… Because if power accrues to those willing to hyperfocus on the explicit measure and cut out everything else, then we should expect a world run by people who have been willing to slash away parts of their humanity in order to win.

Metrics, Shame, and Modern Power

Historically, the guardrail was shame—‘you’re not fun to play with.’…Shame has never felt less effective than it does right now. We live in a system where the green arrow going up and to the right is the only thing that matters.

…and then there’s this little exchange with a very 90s thought. Where “sell-out” was an insult instead of the goal of a person’s life.

Pablo: When I was growing up, obviously, I knew there were like popular things, but it was not as if I read an article or read a book and had my view of it informed by how many other people simultaneously were doing that…We all feel the way in which stuff is worse. It’s hard for me to say that it’s disconnected from the entire conversation we’ve been having.

One of my takes that I have for the new year, popularity will become uncool. And I say that just because we are all watching the mechanisms of what it is to get all of the views and all of the likes and all of the retweets, right? And I just think we’re due for a movement that we’ve seen before, by the way, in which pop culture becomes uncool. And I just think that that is it’s just one of these things that feels like we’re ripe for.

Nguyen (shooting it down): Yeah. Unless that gets captured and large-scale forces successfully gain this sense of uncoolenness and manage to [garbled]. Like punk points. What happened with punk? It became pop punk, right? Some people resisted and were like screw popularity and then large-scale forces figured out how to game that, market that, and we got, you know, “top” alternative radio.

[Random bit I just noticed: Nguyen teaches at University of Utah. Another one of my favorite thinkers and conversation partners, Robert Wuebker, teaches there. I may need to make a field trip to this Utah place.]

when we realized our deltas were wrong

In Thursday’s paid subs post, embedding spot-vol correlation in option deltas, I buried this story but I thought worth sharing since it’s broadly suggestive of what happens when you list options on an investments touted as worth being in your asset allocation:

I started in commodity options just before the listing of electronic options markets. When I first stepped into the trading ring, many market-makers were still using paper sheets. We had spreadsheets on a tablet computer, but heard of a fledgling software called Whentech. Its founder, Dave Wender, was an options trader who saw the opportunity. I demo’d the product, and despite it being a glorified spreadsheet, it centralized a lot of busy work. It had an extensive library of option models and it was integrated with the exchange’s security master so its “sheets” were customized to the asset you wanted to trade.

I started using it right away. Since it was a small company, I was able to have lots of access to Dave with whom I’ve remained friends. I even helped with some of their calculations (weighted gamma was my most important contribution). I was a customer up until I left full-time trading. [Dave sold the company to the ICE in the early 2010s. It’s been called ICE Option Analytics or IOA for over a decade.]

The product evolved closely with the markets themselves. Its nomenclature even became the lingua franca of the floor. Everyone would refer to the daily implied move as a “breakeven” or the amount you needed the futures to move to breakeven on your gamma (most market-makers were long gamma). Breakeven was a field in the option model. Ari Pine’s twitter name is a callback to those days. Commodity traders didn’t even speak in terms of vols. They spoke of breakevens expanding and contracting.

What does this history have to do with a spot-vol correlation parameter?

This period of time, mid-aughts, was special in the oil markets. It was the decade of China’s hypergrowth. The commodity super-cycle. Exxon becoming the largest company in the world. (Today, energy’s share of the SPY is a tiny fraction of what it was 20 years ago.)

Oil options were booming along with open interest in “paper barrels” as Goldman carried on about commodities as an asset class. But what comes with financialization and passive investing?

Option selling. Especially calls.

Absent any political turmoil, resting call offers piled on the order books, vol coming in on every uptick as the futures climbed higher throughout the decade.

A little option theory goes a long way. Holding time and vol constant, what determines the price of an ATM straddle?

The underlying price itself: S

straddle = .8 * S *σ√T

If the market rallies 1%, you expect the straddle price at the new ATM strike to be 1% higher than the ATM straddle when the futures were lower. Since the “breakeven” is just the straddle / 16, you expect the breakeven to also expand by 1%.

But that’s not what was happening.

The breakevens would stay roughly the same as the market moved up and down.

If the breakevens stay the same, that means if the futures go up 1%, then the vol must be falling by 1% (ie 30 vol falling to 29.7 vol)

It dawned us. Our deltas are wrong.

If we are long vol, we need to be net long delta to actually be flat.

When your risk manager says why are you long delta and you explain “I need to lean long” to actually be flat, you can imagine the next question:

“Ok then, how many futures do you need to be extra long for this fudge factor?”

We need to bake this directly into the model because it’s getting hard to keep track of. Every asset and even every expiry within each asset seems to have different sensitivities between vol and spot. The risk report can’t be covered in asterisks detailing thumb-in-the-air trader leans.

Whentech listened. Whentech introduced a new skew model that allowed traders to specify a slope parameter that dictated the path of ATM IV. Their approach was simple and numerical…

Earnings IV Glide Paths

I want to expand briefly on Wednesday’s HOOD: A Case Study in “Renting the Straddle” because HOOD’s implied volatility that contains earnings actually declined for the rest of the week and disentangling that is a good chance to reinforce your understanding.

On Wednesday, Feb 13th HOOD vol (which encompasses earnings on Feb 10) lifted a bit from when I wrote the post. We’ll call it 68% IV.

To make 68% IV fit smoothly with the non-earnings vols from the preceding expirations, we need to assume an earnings move that allow the ex-earnings vol to be ~56%

That corresponds to about a 9.5% earnings move (a bit higher than the average move of 8.55% for the past 8 quarters).

This table shows implied trading day IVs net of various-sized expected earnings moves.

Let’s tie this idea back to theta or option time decay.

A one-day move of 9.5% corresponds to a single-day implied vol of ~119%

9.5% / .80 = 119%

This comes from remembering that an ATM straddle is 80% of the implied vol

As you approach the earnings day, the implied vol of the option will be dominated by the fact that the stock is expected to move 9.5%. Therefore, we know the implied vol is going to increase.

We think of theta as “how much value the option loses as time passes” but because we know that vol is going to steadily rise, we can conclude that the actual experience of theta is going to be much less than the model says. The model doesn’t “know” the implied vol is going to increase, but you do.

As vol increases, the option will gain value that offsets some of the theta. It won’t offset all the theta. If it did, then you would just buy all the options today, have free gamma for a month, and sell them right before earnings.

So much of the theta will be offset?

We can answer this if we hold our assumptions constant:

  • trading day IV is 56%
  • earnings move is 9.5%

(I added the assumption that the earnings date is also the expiration date. It’s stark that all the theta we defer happens on the last day.

You can see how the vega offsets part of the theta.

Just like with any option, the theta still accelerates as you approach expiry but at a slow rate (theta is left axis).

All the theta happens at the end.

Oh, as a matter of pragmatism, I should add that HOOD option markets are wide. And yet there’s millions of contracts of open interest! Amazing for market makers. To quote Alanis…isn’t that ironic?

haters

The following statements are simultaneously true:

1) You can do anything if you put your mind to it” is a lie.

If your last name is McCaffery, you have a chance of engineering elite athletes. If you are an Abdelmessih, you’ll be waiting for the metaverse for the sensation of what a 4.3 40 feels like.

2) You are currently very far from your ceiling.

You can drive a truck through the gap between these 2 ideas so they are not really in conflict.

To let the first disappoint you is to let perfection thwart the good. The cost of this pedantically true statement is self-defeat. It’s the kind of victory only an intellectual would recognize because it’s familiar territory — an unnatural use of technicalities to excuse failure because they define success as adherence to fine print. It’s a strange inversion of “It’s better to be roughly right than precisely wrong”. They are precisely right but roughly wrong, but the wrongness touches their life ceaselessly and in the most material ways.

To let the second statement disappoint you is known as a “start”. Congratulations. Recognition is the first step. This should be obvious, but the path to improvement starts by realizing there’s room for it. In you. Not in the world changing such that your conditions are improved, but for you to improve your station, with a smiling indifference to a world you can’t control anyway.

But I stay “start” because beginnings are sensitive to expectations. If you start anything expecting it to be easy, you will likely not finish. It’s such a simple observation, but it bears a life-changing load. It means that anything you are serious about doing should start with the expectation that it will test your resolve, so when the moment comes, you are not hit with the double indignity of difficulty but also surprise.

And one of those negative surprises always comes from others. Haters. But haters also come from people who don’t actually hate you. They may even love you. But this is how they deal with being disappointed in themselves.

This is not an easy subject. It’s at the root of how everyone relates to everyone else. It’s wrapped up in status, luck, a sense of narrow justice when it has to do with the promotion at work, and global justice in the sense of being born on third (or America…although whether the runner is heading home or to second is today’s “dress” debate).

It’s not as easy as saying “ignore everyone else”. There’s a scammer in jail or even just a common internet grifter who dismissed sober advice from someone they respect who they dismissed as speaking behind a veil of risk-aversion. Or less scandalous scenarios like “I’m dropping out of school to pursue acting”.

It’s a good idea to consider the judgement of those you are certain love you. But even then you need to grade them on a curve based on their own risk bias which takes some judgement of your own. Parents want to see their adult children on solid ground. If you win an Oscar and they get to walk the red carpet, that’s just gravy. That will never be in their calculus. But it might be in yours. They’re running a max-min strategy, you want to win the tournament.

(Rob Carver’s analogy to the Wordle starting word choice is a tangible expression of this for most of the English-speaking world who got swept up in that game.)

So if you should consider the judgment of loved ones and even then with skepticism, you know who you should definitely ignore? Randos and water-cooler friends. There’s just too much at stake.

I posted this on X in a thread where Ryan was parrying haters.

When it comes to haters its useful to remember that the correct retaliation is nothing but apathy which if the detractor was smart in the first place, they would realize that themselves. What’s that line about hate being like drinking poison and expecting the other person to die?

Hate seems like ultimate confession of weakness.

It’s very rare that anyone changes anyone else’s mind. Unlearning hurts.

The internet fools us because when life’s most important moments happen, your world shrinks. The volume on everything turns down, and you are left with a few people. The hater? Might as well be an atom in another galaxy. Why would they occur to you?

Attention is everything. Lots of people on here give you the gift of permitting yourself to ignore them. Accept it gratefully

Scott Adams, the Dilbert cartoonist, died this week after a battle with prostate cancer. He’s a politicized figure (Scott Alexander’s memorial post is a bizarre mix of tribute and psychoanalysis). But like many others, I’ve read his work on career advice and even the thought experiment book “God’s Debris” which I remember precisely nothing about. But I did see a quote from it this week, which I strongly agree with:

“People think they follow advice but they don’t. Humans are only capable of receiving information. They create their own advice. If you seek to influence someone, don’t waste time giving advice. You can change only what people know, not what they do.”

Moontower #299

Friends,

The following statements are simultaneously true:

1) You can do anything if you put your mind to it” is a lie.

If your last name is McCaffery, you have a chance of engineering elite athletes. If you are an Abdelmessih, you’ll be waiting for the metaverse for the sensation of what a 4.3 40 feels like.

2) You are currently very far from your ceiling.

You can drive a truck through the gap between these 2 ideas so they are not really in conflict.

To let the first disappoint you is to let perfection thwart the good. The cost of this pedantically true statement is self-defeat. It’s the kind of victory only an intellectual would recognize because it’s familiar territory — an unnatural use of technicalities to excuse failure because they define success as adherence to fine print. It’s a strange inversion of “It’s better to be roughly right than precisely wrong”. They are precisely right but roughly wrong, but the wrongness touches their life ceaselessly and in the most material ways.

To let the second statement disappoint you is known as a “start”. Congratulations. Recognition is the first step. This should be obvious, but the path to improvement starts by realizing there’s room for it. In you. Not in the world changing such that your conditions are improved, but for you to improve your station, with a smiling indifference to a world you can’t control anyway.

But I stay “start” because beginnings are sensitive to expectations. If you start anything expecting it to be easy, you will likely not finish. It’s such a simple observation, but it bears a life-changing load. It means that anything you are serious about doing should start with the expectation that it will test your resolve, so when the moment comes, you are not hit with the double indignity of difficulty but also surprise.

And one of those negative surprises always comes from others. Haters. But haters also come from people who don’t actually hate you. They may even love you. But this is how they deal with being disappointed in themselves.

This is not an easy subject. It’s at the root of how everyone relates to everyone else. It’s wrapped up in status, luck, a sense of narrow justice when it has to do with the promotion at work, and global justice in the sense of being born on third (or America…although whether the runner is heading home or to second is today’s “dress” debate).

It’s not as easy as saying “ignore everyone else”. There’s a scammer in jail or even just a common internet grifter who dismissed sober advice from someone they respect who they dismissed as speaking behind a veil of risk-aversion. Or less scandalous scenarios like “I’m dropping out of school to pursue acting”.

It’s a good idea to consider the judgement of those you are certain love you. But even then you need to grade them on a curve based on their own risk bias which takes some judgement of your own. Parents want to see their adult children on solid ground. If you win an Oscar and they get to walk the red carpet, that’s just gravy. That will never be in their calculus. But it might be in yours. They’re running a max-min strategy, you want to win the tournament.

(Rob Carver’s analogy to the Wordle starting word choice is a tangible expression of this for most of the English-speaking world who got swept up in that game.)

So if you should consider the judgment of loved ones and even then with skepticism, you know who you should definitely ignore? Randos and water-cooler friends. There’s just too much at stake.

I posted this on X in a thread where Ryan was parrying haters.

When it comes to haters its useful to remember that the correct retaliation is nothing but apathy which if the detractor was smart in the first place, they would realize that themselves. What’s that line about hate being like drinking poison and expecting the other person to die?

Hate seems like ultimate confession of weakness.

It’s very rare that anyone changes anyone else’s mind. Unlearning hurts.

The internet fools us because when life’s most important moments happen, your world shrinks. The volume on everything turns down, and you are left with a few people. The hater? Might as well be an atom in another galaxy. Why would they occur to you?

Attention is everything. Lots of people on here give you the gift of permitting yourself to ignore them. Accept it gratefully

Scott Adams, the Dilbert cartoonist, died this week after a battle with prostate cancer. He’s a politicized figure (Scott Alexander’s memorial post is a bizarre mix of tribute and psychoanalysis). But like many others, I’ve read his work on career advice and even the thought experiment book “God’s Debris” which I remember precisely nothing about. But I did see a quote from it this week, which I strongly agree with:

“People think they follow advice but they don’t. Humans are only capable of receiving information. They create their own advice. If you seek to influence someone, don’t waste time giving advice. You can change only what people know, not what they do.”

Money Angle

I want to expand briefly on Wednesday’s HOOD: A Case Study in “Renting the Straddle” because HOOD’s implied volatility that contains earnings actually declined for the rest of the week and disentangling that is a good chance to reinforce your understanding.

On Wednesday, Feb 13th HOOD vol (which encompasses earnings on Feb 10) lifted a bit from when I wrote the post. We’ll call it 68% IV.

To make 68% IV fit smoothly with the non-earnings vols from the preceding expirations, we need to assume an earnings move that allow the ex-earnings vol to be ~56%

That corresponds to about a 9.5% earnings move (a bit higher than the average move of 8.55% for the past 8 quarters).

This table shows implied trading day IVs net of various-sized expected earnings moves.

Let’s tie this idea back to theta or option time decay.

A one-day move of 9.5% corresponds to a single-day implied vol of ~119%

9.5% / .80 = 119%

This comes from remembering that an ATM straddle is 80% of the implied vol

As you approach the earnings day, the implied vol of the option will be dominated by the fact that the stock is expected to move 9.5%. Therefore, we know the implied vol is going to increase.

We think of theta as “how much value the option loses as time passes” but because we know that vol is going to steadily rise, we can conclude that the actual experience of theta is going to be much less than the model says. The model doesn’t “know” the implied vol is going to increase, but you do.

As vol increases, the option will gain value that offsets some of the theta. It won’t offset all the theta. If it did, then you would just buy all the options today, have free gamma for a month, and sell them right before earnings.

So much of the theta will be offset?

We can answer this if we hold our assumptions constant:

  • trading day IV is 56%
  • earnings move is 9.5%

(I added the assumption that the earnings date is also the expiration date. It’s stark that all the theta we defer happens on the last day.

You can see how the vega offsets part of the theta.

Just like with any option, the theta still accelerates as you approach expiry but at a slow rate (theta is left axis).

All the theta happens at the end.

Oh, as a matter of pragmatism, I should add that HOOD option markets are wide. And yet there’s millions of contracts of open interest! Amazing for market makers. To quote Alanis…isn’t that ironic?

Money Angle For Masochists

In Thursday’s paid subs post, embedding spot-vol correlation in option deltas, I buried this story but I thought worth sharing since it’s broadly suggestive of what happens when you list options on an investments touted as worth being in your asset allocation:

I started in commodity options just before the listing of electronic options markets. When I first stepped into the trading ring, many market-makers were still using paper sheets. We had spreadsheets on a tablet computer, but heard of a fledgling software called Whentech. Its founder, Dave Wender, was an options trader who saw the opportunity. I demo’d the product, and despite it being a glorified spreadsheet, it centralized a lot of busy work. It had an extensive library of option models and it was integrated with the exchange’s security master so its “sheets” were customized to the asset you wanted to trade.

I started using it right away. Since it was a small company, I was able to have lots of access to Dave with whom I’ve remained friends. I even helped with some of their calculations (weighted gamma was my most important contribution). I was a customer up until I left full-time trading. [Dave sold the company to the ICE in the early 2010s. It’s been called ICE Option Analytics or IOA for over a decade.]

The product evolved closely with the markets themselves. Its nomenclature even became the lingua franca of the floor. Everyone would refer to the daily implied move as a “breakeven” or the amount you needed the futures to move to breakeven on your gamma (most market-makers were long gamma). Breakeven was a field in the option model. Ari Pine’s twitter name is a callback to those days. Commodity traders didn’t even speak in terms of vols. They spoke of breakevens expanding and contracting.

What does this history have to do with a spot-vol correlation parameter?

This period of time, mid-aughts, was special in the oil markets. It was the decade of China’s hypergrowth. The commodity super-cycle. Exxon becoming the largest company in the world. (Today, energy’s share of the SPY is a tiny fraction of what it was 20 years ago.)

Oil options were booming along with open interest in “paper barrels” as Goldman carried on about commodities as an asset class. But what comes with financialization and passive investing?

Option selling. Especially calls.

Absent any political turmoil, resting call offers piled on the order books, vol coming in on every uptick as the futures climbed higher throughout the decade.

A little option theory goes a long way. Holding time and vol constant, what determines the price of an ATM straddle?

The underlying price itself: S

straddle = .8 * S *σ√T

If the market rallies 1%, you expect the straddle price at the new ATM strike to be 1% higher than the ATM straddle when the futures were lower. Since the “breakeven” is just the straddle / 16, you expect the breakeven to also expand by 1%.

But that’s not what was happening.

The breakevens would stay roughly the same as the market moved up and down.

If the breakevens stay the same, that means if the futures go up 1%, then the vol must be falling by 1% (ie 30 vol falling to 29.7 vol)

It dawned us. Our deltas are wrong.

If we are long vol, we need to be net long delta to actually be flat.

When your risk manager says why are you long delta and you explain “I need to lean long” to actually be flat, you can imagine the next question:

“Ok then, how many futures do you need to be extra long for this fudge factor?”

We need to bake this directly into the model because it’s getting hard to keep track of. Every asset and even every expiry within each asset seems to have different sensitivities between vol and spot. The risk report can’t be covered in asterisks detailing thumb-in-the-air trader leans.

Whentech listened. Whentech introduced a new skew model that allowed traders to specify a slope parameter that dictated the path of ATM IV. Their approach was simple and numerical…

 

From My Actual Life

I definitely have more couch potato tendencies in the winter. I’m currently watching Mad Men (for the first time!) I’m almost finished with Season 2 which means I like it.

I recommend the movie Eden on Netflix. Go into it knowing nothing. That’s how I went in (Yinh said let’s watch some movie called Eden and I said ok knowing nothing else). I’m so out of touch sometimes, we were a quarter of the way through the movie before I said “Isn’t that Jude Law?”

It’s definitiely one of those movies where right after you finish it, you’re googling “how true were the events in [movie title]?”

Wednesday night was the first time I ever went to a Cal game which is kinda pathetic since I’ve lived less than 20 minutes from Berkeley for over a decade now. But St. Mary’s College usually has a better hoops team, is even closer, and has a much smaller arena. It’s more of a gym than a venue.

Cal was able to hang with the Blue Devils for the first half before Duke started being Duke.

So many nepo babies in the game. Marbury’s son was is a sophomore walk-on for Cal (he’s only played 5 minutes all season though), Justin Pippen is Cal’s starting PG as a sophomore, and the freshman Boozer twins play for Duke (although only the 6’9” one sees the court. 6’4” bro MIA). The taller Boozer is a force. Much savvier than you might expect from a freshman big.

Duke brought out some local celebs. We didn’t see them, but Steph and Del Curry were there with family. We did see these guys one of whom’s life is basically a victory lap. Getting dapped up every 3 seconds, everyone taking selfies with him. You can decide who I’m talking about:

 

 

Stay groovy

☮️

Moontower Weekly Recap

Posts:

HOOD: A Case Study in “Renting the Straddle”

On Monday, I noticed that Robinhood ($HOOD) vol screened cheap in the Trade Ideas tool. But that tool uses 30-day constant maturity IV. Since HOOD earnings was just about 30 days out on Monday, the interpolation gave the earnings vol no weight. The pre-earnings vol is in the low 50s, which is, indeed at the bottom of the range for HOOD implied vol.

I looked at the vol that includes earnings.

HOOD reports earnings on February 10th. The February 13th expiry is currently priced at 64% ATM implied volatility.

At first glance, 64% might seem elevated but let’s decompose what the market is actually pricing. When a known event, like earnings, falls within an option’s expiry, the market assigns extra volatility to that expiration. But how much of that IV comes from the event itself versus normal trading day volatility?

I’m gonna lay out the numbers and then get to the process.

• Expected earnings move or straddle: 8.55%

• Event volatility (earnings day): 10.72% single-day vol (169.8% annualized)

Why?

The ATF straddle approximation tells us that a straddle ~ .8 x vol

Well, if we assume the earnings straddle is 8.55% then we just divide that by .80 (or multiply by 1.25 which is the arithmetic burned into trader brain) to get 10.69%

• Trading day volatility (pre-earnings): 54% annualized = 3.41% per day

Where do these numbers come from?

Let’s start with the earnings straddle…why 8.55%?

Here’s a handy secret. A good first guess what the market’s estimate for an earnings moves is the mean move size of the last 4 or even 8 earnings.

I just asked Gemini.

Title: Historical Earnings Moves - Description: HOOD historical moves

It’s a good first guess but then you run that number through our Event Volatility Extractor:

Once you’ve extracted the lump of variance that comes from an 8.55% move on a single day, the remaining variance until expiry is then divided over the remaining days. That’s what that calculator does. It tells you that the ex-earnings implied vol is 54% IF you accept that the earnings move is 8.55%

Since the IVs that precede the Feb 13th expiry are in the low-50s then the term structure ex-earnings is smooth and sensible. If it wasn’t, then we know the market is pricing a very different move size for earnings.

We are just slicing a pizza pie. The whole pizza is the total variance until Feb 13th, currently encompassed by 64% IV. The bigger you make the earnings slice, the smaller the remaining slices (regular trading days) have to be. If the extracted trading day vol turned out to be much lower than 54%, then the market must be expecting a bigger earnings move to account for the difference. Conversely, if it extracted to 62%, the market is pricing a smaller earnings move than 8.55%. The smooth term structure tells us 8.55% slices the pie correctly—each regular day gets roughly the same-sized piece. The Feb 13 expiry sits naturally in line with surrounding expirations.

But…

  • If you think that’s too high for earnings, you could sell the Feb 13th expiry and buy the expiry preceding it. If you think it’s too low, you could do the opposite.
  • If you think it’s a fair price, then you can simply judge the implied trading day vol on its own merit — 54%.

[Our tools will programmatically do this so that we can then use the ex-earnings vols in our standard Trade Ideas cross-section algo. Until then, we are adding a filter that allows you to exclude names with earnings upcoming from the cross-section sorter.]

So is HOOD vol cheap?

The Trade Ideas algo thinks it’s relatively cheap. Relative depends on your universe. Based on the universe I calibrated on (over 100 liquid ETFs and stocks) it screens cheap.

But an obvious follow-up question is…does it look absolutely cheap compared to its own history?

The answer is ‘“yea”. It’s not screaming cheap, but it’s on the cheaper side.

[This is where it helps to have context. Like if you follow the stock closely and have any feels on it then knowing the options are a bit cheap can inspire some trade structures that get you more juice for your knowledge.]

A few views into its history:

The current IV curve is lower than median realized vols,and a bit higher than current realized vols. BUT…current realized vols are also less than 25th percentile. They only need to sneeze up to median levels for these options to price much higher (especially if they maintain the same VRP ratio which is totally reasonable).

Title: Event Volatility Extractor - Description: HOOD event vol decomposition

If you prefer time series, the current 30-day IV is sitting near the 1-year low for 1-month implied vol (red line).

Recapping some of the more challenging points:

  • The entire “cheap vol” thesis depends on whether the 8.55% expected move is reasonable.
  • While 8.55% matches HOOD’s historical average, that’s not how we finalized the number we should use. It’s a starting point that we then test to see if that move size would produce a smooth or humped term structure. If it causes the term structure to jump higher than we are using too small of an estimate, if it causes it to invert sharply, then we are using too high an earnings estimate. If your head hurts, you’re doing this right. Maybe 8.75% or 8.35% makes the term structure a touch smoother but you can use the calculator to see how much little adjustments like that flow through to an implied trading day vol. It has a bigger impact than you might think…changing the earnings day straddle by .25% can move the trading vol by a .5 to 1 point. This is below the threshold anyone except high volume vol traders and market-makers should care about.
  • The embedded risk you take when “renting the straddle”: the implied earnings move compresses as you approach February 10th – perhaps because the market decides HOOD’s earnings will be less volatile than historical patterns suggest – then your “cheap” pre-earnings vol becomes less cheap. You’d be holding a position where the event vol component is shrinking, pulling down the value of your straddle beyond normal theta decay. You’re not just betting on realized vol exceeding 54%. You’re also betting that the market continues to price in an ~8.55% earnings move.

Key Takeaway

Decomposing event volatility matters for cross-asset comparison and relative value analysis. A 64% implied volatility might look high in isolation, but after extracting a 170% event vol component (calibrated to produce a smooth term structure), you’re left with 54% trading day vol – which can then be evaluated against your regular toolkit.

 

_______________________________________________________________________________

Appendix: Recipe for Cross-Sectional Analysis With Earnings Names

I used Claude to encapsulate and synthesize a recipe. You can decide how it did:

One of the most powerful applications of event extraction is enabling apples-to-apples comparison across tickers – even when some have earnings and others don’t.

The Problem

Standard cross-sectional vol analysis breaks down when comparing:

• AAPL at 35% IV (no events)

• NVDA at 48% IV (earnings in 30 days)

Which is really “cheaper”? You can’t tell without extracting the event component.

The Recipe

Step 1: Identify Events in Your Universe

For each ticker in your analysis:

• Check earnings calendar (next 30 days typically)

• Note FOMC weeks for macro-sensitive names

• Flag other known catalysts (FDA decisions, etc.)

Step 2: Extract Base Vols Using Term Structure Smoothness

For each ticker with events:

a) Pull the full term structure of ATM IVs

b) Use the Event Volatility Extractor with different move size assumptions

c) The “right” move is the one that produces a smooth, non-humpy base vol term structure

This is the key insight from the NVDA example: too high an earnings move creates an unnatural dip after earnings; too low creates a spike. The correct assumption produces a smooth power law curve.

Step 3: Record Your Assumptions

For each extraction, document: Ticker, Earnings date, Assumed move size (%), Term structure fit quality (R²), Your confidence level (tight/loose)

Step 4: Run Cross-Sectional Analysis on Clean Vols

Now compare:

• AAPL: 35% IV (no adjustment needed)

• NVDA: 44% base vol (extracted from 48% dirty vol with 6.5% earnings move)

Calculate percentile rankings using the clean vols for all four dimensions: IV percentile (using base vols), RV percentile, VRP (base IV – RV), and Term structure steepness (using base vol term structure).

Step 5: Understand What You’re Betting On

When you identify NVDA as “cheap” after extraction, you’re making TWO assumptions: (1) Base vol of 44% is cheap relative to history, and (2) Market will continue pricing ~6.5% earnings move (your assumption holds).

The Cross-Sectional Edge

By extracting events, you accomplish two things:

1. Expand your opportunity set: Instead of excluding 30-40% of your universe during earnings season, you can analyze everyone on equal footing

2. Identify hidden opportunities: Sometimes the “expensive looking” ticker with earnings is actually cheap on a base vol basis, or vice versa

The market often prices earnings mechanically (historical average moves), but base vol can be at extremes. Finding names where base vol is at the 5th percentile but dirty vol looks “normal” because of earnings—that’s where edge lives.

Games we played during Christmas 2025

We played a lot of games over the holiday break. Some recs.

Timeline games

Hitster: Draw a card. Scan the QR code and a song plays on Spotify. Place it on your timeline based on the year the song was released. First player to line up 10 cards wins.

Hitster is really simple but a fun music themed game. Listen to songs from a QR code and try to place its release year in a time line in relation to other
image via FB Group

Chronology: Same idea as Hitster but cards with historical events written on them. Did you know the first looping roller coaster preceded the Gettysburg Address? Neither did I.

Weirdly, these games are not by the same company despite the same mechanic of completing a timeline of 10. I’ve never played a game with that mechanic and then played 2 with 4 days. Baader-Meinhof game moment, I guess.

 

Trickery games

Imposter: This is a free social deduction game that got a lot of play since we had several large gatherings.

 

Skull: I’ve boosted this game before. It’s reminiscent of poker or Liar’s Dice but we played a bunch over break with several groups and it universally loved. Even the 9 year-olds were super into it. Take 2 minutes to learn but then it’s very rich.

My favorite game review channel is Shut Up & Sit Down:

You really don’t need to buy the game to play it. Here’s the same game played with whatever cards you have around the house:

Finally, we played a giant round of a game I wrote about last year:

Left Center Right (1 min video)

This game is pure degeneracy and takes less than a minute to learn. Asian grandmas and 5-year-olds alike will lose their minds over it. Huge party hit this holidays. It’s actually an old game, but new to me. It has zero skill so when I heard how it works I immediately poo poo’d it but playing it in a group of 15 for a little cash is amazing.

If you want to make it skillful just create an open outcry side-market on who the winner is. Let’s say “Ann” is playing…Ann futures settle to 0 or 100 depending on if Ann wins so you can bid, offer, or trade any integer price between 0 and 100 based on your assessed probability of Ann winning. It’s a faithful simulation of mock trading (and really similar to the StockSlam game I was playing a couple years ago).

This year, we played a single round of LCR on Christmas Day that took close to an hour. 17 people with a $20 buy-in. Winner got $340. Asian aunties were rabid. Call me whatever you want, but “these people” love gambling.

[Because of the buy-in, we used poker chips instead of singles for tokens. I was told this took away from the experience since “grandma wants to see the cash”. Noted for next time.]

work is going to feel very different by next Christmas

I started to feel it over the break, but the feeling is inescapable after the past week.

This has nothing to do with current events. It’s me having the same reaction to Claude Code that early adopters using the terminal have already felt:

Work is going to feel very different by next Christmas. Yinh and I were talking about how long 2024 felt. There were a lot of life events, but just in terms of workflow, it felt like a year ago I was mostly using LLMs for transcription, editing, and giving it photos of broken stuff for help.

Today, I can write a description of a bug in Linear or Jira (who am I kidding — I upload a screenshot with a blurb and have AI write a detailed bug spec complete with testing protocol) and assign it to…”Claude bot”. A dev approves the change. Push to prod. Hundreds of hours saved over the course of a year. It’s accelerating by the week.

I’ll share more about what I’m doing personally in the letter this week, but here are a few must-reads if you are curious about getting more out of AI.

1) Claude, Code, and What Comes Next (6 min read) by Ethan Mollick

This is a strong overview of why Claude Code feels so stepped-up in capability. I’m using Opus 4.5 regularly and for one project in particular, its ability to compress the chat is lengthening the context window. This post is a nice primer to read while the idea of agents working 24/7 for you floats in the back of your mind.

2) How I code with agents, without being ‘technical’ by Ben Tossell

Khe texted me this post and it’s the next thing to read after Mollick’s. This gives you a glimpse into the near future (which is already here for Ben despite his humility in this post) with very concrete ideas. PSA: Khe’s letter is mandatory if you are a regular person trying to get the most out of the tools around us. I feel like I’m literally stepping in his tracks, just 3 months behind as I’m finally using Claude Code (I just needed it to be in a desktop app rather than command line because I just think DOS and my brain powers off.)

In this article, Ben says:

Not to be like everyone else on Twitter when they see Andrej Karpathy tweeting something, but this really rang true to me: **there’s a new programmable layer of abstraction to master.**

First of all, I have the Claude extension in my chrome browser. This lets you talk to Claude about anything you are seeing. Like the design of the website you’re on? Ask it to extract a style sheet. Don’t want to read the whole email or article? No need to copy/paste, just ask the sidekick to summarize it. There’s even a Google Sheets add-on if you want it to spreadsheet for you.

In this case, I just asked Claude in my browser for the Karpathy thread that Ben is referencing.

Boom, it just goes out to the web and finds it.

Here’s the Karpathy quote:

I’ve never felt this much behind as a programmer. The profession is being dramatically refactored as the bits contributed by the programmer are increasingly sparse and between. I have a sense that I could be 10X more powerful if I just properly string together what has become available over the last ~year and a failure to claim the boost feels decidedly like skill issue. There’s a new programmable layer of abstraction to master (in addition to the usual layers below) involving agents, subagents, their prompts, contexts, memory, modes, permissions, tools, plugins, skills, hooks, MCP, LSP, slash commands, workflows, IDE integrations, and a need to build an all-encompassing mental model for strengths and pitfalls of fundamentally stochastic, fallible, unintelligible and changing entities suddenly intermingled with what used to be good old fashioned engineering. Clearly some powerful alien tool was handed around except it comes with no manual and everyone has to figure out how to hold it and operate it, while the resulting magnitude 9 earthquake is rocking the profession. Roll up your sleeves to not fall behind.

If Karpathy feels behind, I guess the rest of us shouldn’t feel so bad. But the part I bolded feels big. Like you need to stop your first reflex about how you’d approach a problem and embody someone to whom this is all native to (while recognizing that nobody is perfectly native to it. Instead, there’s a continuum of how far along people are in how easily they consider problems in light of the new capabilities.)

3) Claude Codes by Zvi Mowshowitz

Things are moving fast. This came out 48 hours ago. Highly practical and honest assessment of the current state. Also happens to echo my opinion — this is going to be a vertical year in terms of workflow.

4) Greg Isenberg on “what young builders do” (X thread)

This thread is just a mind-eff because it shows the frontier of the kids building in the context of entrepreneurship.

5) Everyone Is Wrong About the Skilled Labor Shortage (5 min read) by Jon Matzner

. I tend not to think about things along the lines of “what are the jobs of the future”. It feels like when you do that, you are choosing a self-alienating frame that favores the predicate over the subject.

Anyway, a local friend is a lecturer at Cal in AI. Kids similar age as mine. He gets asked about future jobs all the time and I can’t pretend I never think of that even if I resist the impulse.

His answer is “fix people, fix animals, or fix robots”. He’s also partial to the “trades”. Basically, work that AI will eat last.

It makes sense. I can’t say I’m sold. My own view is that the acceleration is so fast that any prediction on those lines is swamped by the error bars, but insofar as you must choose, it’s as good a guess any. But that’s not a great foundation for deciding, so I just treat that topic as entertainment.

My view here even disappoints me because it sounds helpless with respect to planning. But then I read an article like Matzner’s and it’s an example of how a lot of consensus thinking (like going into the trades) is perfectly risky. The frictions to knowing how to do something will melt. The asymmetry in info that a tradesperson has compared to the client has been narrowing over time (YouTube) but the “last mile” of actually doing is going to get shorter. You’re going to know how to fix anything at home, it will be a question of whether the time is worth it or not. If there are no jobs, we’ll have plenty of time to fix things. I think I’m kidding. But what if I’m accidentally right?

6) Dos Capital by Zvi Mowshowitz

And now we get to the macro. Provocation instead of practical. For the lolz.

Zvi’s post is a reaction to Trammell and Dwarkesh’s post about the unprecedented wealth inequality we are about to see. What Zvi calls absurd is effectively Trammel & Dwarkesh not taking their premise seriously enough.

Zvi (emphasis mine):

They affirm, as do I, that Piketty was centrally wrong about capital accumulation in the past, for many well understood reasons, many of which they lay out.

They then posit that Piketty could have been unintentionally describing our AI future.

As in, IF, as they say they expect is likely:

[redacted list of assumptions in order]

Does the above conclusion follow from the above premises if you include the implicit assumptions?

Then yes. Very, very obviously yes. This is basic math.

Sounds Like This Is Not Our Main Problem In This Scenario?

In this scenario, sufficiently capable AIs and robots are multiplying without limit and are perfect substitutes for human labor.

Perhaps ‘what about the distribution of wealth among humans’ is the wrong question?

I notice I have much more important questions about such worlds where the share of profits that goes to some combination AI, robots and capital rises to all of it.

Why should the implicit assumptions hold? Why should we presume humans retain primary or all ownership of capital over time? Why should we assume humans are able to retain control over this future and make meaningful decisions? Why should we assume the humans remain able to even physically survive let alone thrive?

Note especially the assumption that AIs don’t end up with substantial private property. The best returns on capital in such worlds would obviously go to ‘the AIs that are, directly or indirectly, instructed to do that.’…

Even if we assumed all of that, why should we assume that private property rights would be indefinitely respected at limitless scale, on the level of owning galaxies? Why should we even expect property rights to be long term respected under normal conditions, here on Earth? Especially in a post calling for aggressive taxation on wealth, which is kind of the central ‘nice’ case of not respecting private property.

The world described here has AIs that are no longer normal technology (while it tries to treat them as normal in other places anyway), it is not remotely at equilibrium, there is no reason to expect its property rights to endorse or to stay meaningful, it would be dominated by its AIs, and it would not long endure.

If humans really are no longer useful, that breaks most of the assumptions and models of traditional econ along with everyone else’s models, and people typically keep assuming actually humans will still be useful for something sufficiently for comparative advantage to rescue us, and can’t actually wrap their heads around it not being true and humans being true zero marginal product workers given costs.

That’s the thing. If we’re talking about a Dyson sphere world, why are we pretending any of these questions are remotely important or ultimately matter? At some point you have to stop playing with toys.

I don’t even know that ‘wealth’ and ‘consumption’ would be meaningful concepts that look similar to how they look now, among other even bigger questions. I don’t expect ‘the basics’ to hold and I think we have good reasons to expect many of them not to.

Ultimately all of this, as Tomas Bjartur puts it, imagines an absurd world, assuming away all of the dynamics that matter most. Which still leaves something fun and potentially insightful to argue about, I’m happy to do that, but don’t lose sight of it not being a plausible future world, and taking as a given that all our ‘real’ problems mysteriously turn out fine despite us having no way to even plausibly describe what that would look like, let alone any idea how to chart a path towards making it happen.

The AI discourse is a ready reminder that there are no rules. There’s only power. We are bears on a unicycle. To some of our tech overlords humanity is but an experiment. A branch of a codebase we can’t see the extent of. And most likely NOT ‘main’.

To be overwhelmingly confident that this path is humanist is either hubris or motivated by the next round of funding. It just doesn’t seem clear to me that human flourishing is a layup end state of this trajectory.

Google’s mission is “to organize the world’s information and make it universally accessible and useful.” Google’s AI division’s mantra?

“Solve intelligence, and then use that to solve everything else.”

There’s mission creep and then there’s MISSION CREEP. If a business’s goal is to solve a problem and this is a quest to solve all the problems, I think it’s only fair to ask, while we still can…

“What is the last problem?”

[Nate Bargatze voice: Nobody knows]

Not a bad place to insert Asimov’s famous short story, The Last Question.

Moontower #298

Friends,

I started to feel it over the break, but the feeling is inescapable after the past week.

This has nothing to do with current events. It’s me having the same reaction to Claude Code that early adopters using the terminal have already felt:

Work is going to feel very different by next Christmas. Yinh and I were talking about how long 2024 felt. There were a lot of life events, but just in terms of workflow, it felt like a year ago I was mostly using LLMs for transcription, editing, and giving it photos of broken stuff for help.

Today, I can write a description of a bug in Linear or Jira (who am I kidding — I upload a screenshot with a blurb and have AI write a detailed bug spec complete with testing protocol) and assign it to…”Claude bot”. A dev approves the change. Push to prod. Hundreds of hours saved over the course of a year. It’s accelerating by the week.

I’ll share more about what I’m doing personally in the letter this week, but here are a few must-reads if you are curious about getting more out of AI.

1) Claude, Code, and What Comes Next (6 min read) by Ethan Mollick

This is a strong overview of why Claude Code feels so stepped-up in capability. I’m using Opus 4.5 regularly and for one project in particular, its ability to compress the chat is lengthening the context window. This post is a nice primer to read while the idea of agents working 24/7 for you floats in the back of your mind.

2) How I code with agents, without being ‘technical’ by Ben Tossell

Khe texted me this post and it’s the next thing to read after Mollick’s. This gives you a glimpse into the near future (which is already here for Ben despite his humility in this post) with very concrete ideas. PSA: Khe’s letter is mandatory if you are a regular person trying to get the most out of the tools around us. I feel like I’m literally stepping in his tracks, just 3 months behind as I’m finally using Claude Code (I just needed it to be in a desktop app rather than command line because I just think DOS and my brain powers off.)

In this article, Ben says:

Not to be like everyone else on Twitter when they see Andrej Karpathy tweeting something, but this really rang true to me: **there’s a new programmable layer of abstraction to master.**

First of all, I have the Claude extension in my chrome browser. This lets you talk to Claude about anything you are seeing. Like the design of the website you’re on? Ask it to extract a style sheet. Don’t want to read the whole email or article? No need to copy/paste, just ask the sidekick to summarize it. There’s even a Google Sheets add-on if you want it to spreadsheet for you.

In this case, I just asked Claude in my browser for the Karpathy thread that Ben is referencing.

Boom, it just goes out to the web and finds it.

Here’s the Karpathy quote:

I’ve never felt this much behind as a programmer. The profession is being dramatically refactored as the bits contributed by the programmer are increasingly sparse and between. I have a sense that I could be 10X more powerful if I just properly string together what has become available over the last ~year and a failure to claim the boost feels decidedly like skill issue. There’s a new programmable layer of abstraction to master (in addition to the usual layers below) involving agents, subagents, their prompts, contexts, memory, modes, permissions, tools, plugins, skills, hooks, MCP, LSP, slash commands, workflows, IDE integrations, and a need to build an all-encompassing mental model for strengths and pitfalls of fundamentally stochastic, fallible, unintelligible and changing entities suddenly intermingled with what used to be good old fashioned engineering. Clearly some powerful alien tool was handed around except it comes with no manual and everyone has to figure out how to hold it and operate it, while the resulting magnitude 9 earthquake is rocking the profession. Roll up your sleeves to not fall behind.

If Karpathy feels behind, I guess the rest of us shouldn’t feel so bad. But the part I bolded feels big. Like you need to stop your first reflex about how you’d approach a problem and embody someone to whom this is all native to (while recognizing that nobody is perfectly native to it. Instead, there’s a continuum of how far along people are in how easily they consider problems in light of the new capabilities.)

3) Claude Codes by Zvi Mowshowitz

Things are moving fast. This came out 48 hours ago. Highly practical and honest assessment of the current state. Also happens to echo my opinion — this is going to be a vertical year in terms of workflow.

4) Greg Isenberg on “what young builders do” (X thread)

This thread is just a mind-eff because it shows the frontier of the kids building in the context of entrepreneurship.

5) Everyone Is Wrong About the Skilled Labor Shortage (5 min read) by Jon Matzner

. I tend not to think about things along the lines of “what are the jobs of the future”. It feels like when you do that, you are choosing a self-alienating frame that favores the predicate over the subject.

Anyway, a local friend is a lecturer at Cal in AI. Kids similar age as mine. He gets asked about future jobs all the time and I can’t pretend I never think of that even if I resist the impulse.

His answer is “fix people, fix animals, or fix robots”. He’s also partial to the “trades”. Basically, work that AI will eat last.

It makes sense. I can’t say I’m sold. My own view is that the acceleration is so fast that any prediction on those lines is swamped by the error bars, but insofar as you must choose, it’s as good a guess any. But that’s not a great foundation for deciding, so I just treat that topic as entertainment.

My view here even disappoints me because it sounds helpless with respect to planning. But then I read an article like Matzner’s and it’s an example of how a lot of consensus thinking (like going into the trades) is perfectly risky. The frictions to knowing how to do something will melt. The asymmetry in info that a tradesperson has compared to the client has been narrowing over time (YouTube) but the “last mile” of actually doing is going to get shorter. You’re going to know how to fix anything at home, it will be a question of whether the time is worth it or not. If there are no jobs, we’ll have plenty of time to fix things. I think I’m kidding. But what if I’m accidentally right?

6) Dos Capital by Zvi Mowshowitz

And now we get to the macro. Provocation instead of practical. For the lolz.

Zvi’s post is a reaction to Trammell and Dwarkesh’s post about the unprecedented wealth inequality we are about to see. What Zvi calls absurd is effectively Trammel & Dwarkesh not taking their premise seriously enough.

Zvi (emphasis mine):

They affirm, as do I, that Piketty was centrally wrong about capital accumulation in the past, for many well understood reasons, many of which they lay out.

They then posit that Piketty could have been unintentionally describing our AI future.

As in, IF, as they say they expect is likely:

[redacted list of assumptions in order]

Does the above conclusion follow from the above premises if you include the implicit assumptions?

Then yes. Very, very obviously yes. This is basic math.

Sounds Like This Is Not Our Main Problem In This Scenario?

In this scenario, sufficiently capable AIs and robots are multiplying without limit and are perfect substitutes for human labor.

Perhaps ‘what about the distribution of wealth among humans’ is the wrong question?

I notice I have much more important questions about such worlds where the share of profits that goes to some combination AI, robots and capital rises to all of it.

Why should the implicit assumptions hold? Why should we presume humans retain primary or all ownership of capital over time? Why should we assume humans are able to retain control over this future and make meaningful decisions? Why should we assume the humans remain able to even physically survive let alone thrive?

Note especially the assumption that AIs don’t end up with substantial private property. The best returns on capital in such worlds would obviously go to ‘the AIs that are, directly or indirectly, instructed to do that.’…

Even if we assumed all of that, why should we assume that private property rights would be indefinitely respected at limitless scale, on the level of owning galaxies? Why should we even expect property rights to be long term respected under normal conditions, here on Earth? Especially in a post calling for aggressive taxation on wealth, which is kind of the central ‘nice’ case of not respecting private property.

The world described here has AIs that are no longer normal technology (while it tries to treat them as normal in other places anyway), it is not remotely at equilibrium, there is no reason to expect its property rights to endorse or to stay meaningful, it would be dominated by its AIs, and it would not long endure.

If humans really are no longer useful, that breaks most of the assumptions and models of traditional econ along with everyone else’s models, and people typically keep assuming actually humans will still be useful for something sufficiently for comparative advantage to rescue us, and can’t actually wrap their heads around it not being true and humans being true zero marginal product workers given costs.

That’s the thing. If we’re talking about a Dyson sphere world, why are we pretending any of these questions are remotely important or ultimately matter? At some point you have to stop playing with toys.

I don’t even know that ‘wealth’ and ‘consumption’ would be meaningful concepts that look similar to how they look now, among other even bigger questions. I don’t expect ‘the basics’ to hold and I think we have good reasons to expect many of them not to.

Ultimately all of this, as Tomas Bjartur puts it, imagines an absurd world, assuming away all of the dynamics that matter most. Which still leaves something fun and potentially insightful to argue about, I’m happy to do that, but don’t lose sight of it not being a plausible future world, and taking as a given that all our ‘real’ problems mysteriously turn out fine despite us having no way to even plausibly describe what that would look like, let alone any idea how to chart a path towards making it happen.

The AI discourse is a ready reminder that there are no rules. There’s only power. We are bears on a unicycle. To some of our tech overlords humanity is but an experiment. A branch of a codebase we can’t see the extent of. And most likely NOT ‘main’.

To be overwhelmingly confident that this path is humanist is either hubris or motivated by the next round of funding. It just doesn’t seem clear to me that human flourishing is a layup end state of this trajectory.

Google’s mission is “to organize the world’s information and make it universally accessible and useful.” Google’s AI division’s mantra?

“Solve intelligence, and then use that to solve everything else.”

There’s mission creep and then there’s MISSION CREEP. If a business’s goal is to solve a problem and this is a quest to solve all the problems, I think it’s only fair to ask, while we still can…

“What is the last problem?”

[Nate Bargatze voice: Nobody knows]

Not a bad place to insert Asimov’s famous short story, The Last Question.

Money Angle

These 3 videos had me dying. The dude behind them, Benjamin, clearly has a strong grasp of trading and investing, but this is totally underselling the talent.

Erik pointed him out to me. Benjamin publishes videos rarely (they take a long time to create), but despite being sparse in his output the quality is so good that you can see why they get millions of views and why the channel has over 600k subs. You don’t generally see these numbers on an account that has published 33 videos total.

Enjoy!

Money Angle For Masochists

From Kevin Mak’s outstanding Price Discovery and Trading:

What’s elegant is that this example very quickly teaches people to go from thinking with an individualistic perspective (what do I think it’s worth?) to thinking with a market perspective (what is the market saying it’s worth?). This shift in the way you think is extremely counterintuitive to non-market people (who are the majority of my students, and a majority of the population in general).

I urge you to read the post because it contains beautiful game that Kevin created to give students a visceral sense of how prices emerge from the interplay of public and private information.

Just after Christmas, I spent a day at the Arbor Quant Bootcamp run by Ricki & Ross. The bootcamp is 4 days. I attended “options” day (and even had the honor of speaking for an hour).

Options day is the last day of the camp. The capstone game brings together everything you learned over the course of 4 days. It was, by far, the coolest simulation I have ever seen of a trading situation. I won’t say too much but it involves teams huddled around computers live trading a situation that brings together options, index arb, game theory, probability, and an incredibly layered scenario where the right approach is not a settled matter. An incredible canvas for socratic teaching on top of a basic corpus of financial plumbing.

Given the audience of this letter, I hope you can all experience this one day. I sent my neighbor’s son to this bootcamp a few months before entering college because he wanted to know how he’d know if he liked trading. You will walk away from this bootcamp either unable to think of anything else other than what you just experienced or you will want to run far away. Either way you win because you’ll understand what trading really is (and why it’s a general skill set — which explains how many traders at prop firms have traded many different asset classes and markets).

It is a stark experience. You go on the internet and get some impression of what trading is and then you attend this lobotomy and can’t see markets the same way. For the uninitiated, it will spark the “Omg, I can see how looking at the world through this lens prints money for Jane Street et al”.

I have no financial interest in Arbor’s business, I’m just a huge fan. I’ll obviously boost the next session once it’s announced.

(Also, I’ve been working on a card game and one of the mechanics in their simulations might be the unlock I’m looking for. We’ll see. It’s an ongoing project in the moontower skunkworks.)

Advice

I had lots of conversations with people in attendance who were just starting their trading careers ,either at banks, private funds, or prop firms. I was asked for advice quite a bit. I’m generally uncomfortable with that because a lot of the best advice is banal (“don’t be late to work”) and more specific advice is so overfit that you can find its opposite in other people you admire.

But I did get asked one question that I had the rare fast, confident answer to. An eager fella asked me what asset class he should try to get into “for his career”.

Hold your horses, kid. You’re pivoting your data on the wrong column.

You should be more concerned about who you learn from. Asset classes go hot and cold, often for long stretches. They also teach you different things. Single stock option trading is very different from trading options on macro assets or index.

[I should probably write about this, but one is much closer to heads-up no-limit poker and the p/l glide paths from t-zero to t+x is totally inverted between the 2 businesses. If you want to have some laughs, put an index trader in a stock trader’s seat or vice versa. One can learn the other, but there’s an unlearning and learning curve.]

You may not have a choice in who you learn from. You will also not have a mature enough taste when you are starting out to distinguish mentors. So the beginning of your career is an especially sensitive starting condition to an already-wiggly path. But you’re better off at least being aware that “who” is more important than “what”. The right surroundings can turbo-charge your career or saddle you with habits you might never unlearn.

On advice, Gappy’s updated memo is terrific:

2025 Buy-Side Quant Job Advice

 

From My Actual Life

We played a lot of games over the holiday break. Some recs.

Timeline games

Hitster: Draw a card. Scan the QR code and a song plays on Spotify. Place it on your timeline based on the year the song was released. First player to line up 10 cards wins.

Hitster is really simple but a fun music themed game. Listen to songs from a QR code and try to place its release year in a time line in relation to other
image via FB Group

Chronology: Same idea as Hitster but cards with historical events written on them. Did you know the first looping roller coaster preceded the Gettysburg Address? Neither did I.

Weirdly, these games are not by the same company despite the same mechanic of completing a timeline of 10. I’ve never played a game with that mechanic and then played 2 with 4 days. Baader-Meinhof game moment, I guess.

 

Trickery games

Imposter: This is a free social deduction game that got a lot of play since we had several large gatherings.

 

Skull: I’ve boosted this game before. It’s reminiscent of poker or Liar’s Dice but we played a bunch over break with several groups and it universally loved. Even the 9 year-olds were super into it. Take 2 minutes to learn but then it’s very rich.

My favorite game review channel is Shut Up & Sit Down:

You really don’t need to buy the game to play it. Here’s the same game played with whatever cards you have around the house:

Finally, we played a giant round of a game I wrote about last year:

Left Center Right (1 min video)

This game is pure degeneracy and takes less than a minute to learn. Asian grandmas and 5-year-olds alike will lose their minds over it. Huge party hit this holidays. It’s actually an old game, but new to me. It has zero skill so when I heard how it works I immediately poo poo’d it but playing it in a group of 15 for a little cash is amazing.

If you want to make it skillful just create an open outcry side-market on who the winner is. Let’s say “Ann” is playing…Ann futures settle to 0 or 100 depending on if Ann wins so you can bid, offer, or trade any integer price between 0 and 100 based on your assessed probability of Ann winning. It’s a faithful simulation of mock trading (and really similar to the StockSlam game I was playing a couple years ago).

This year, we played a single round of LCR on Christmas Day that took close to an hour. 17 people with a $20 buy-in. Winner got $340. Asian aunties were rabid. Call me whatever you want, but “these people” love gambling.

[Because of the buy-in, we used poker chips instead of singles for tokens. I was told this took away from the experience since “grandma wants to see the cash”. Noted for next time.]

 

Stay groovy

☮️

Moontower Weekly Recap

Posts:

slop

Merriam-Webster’s Word of the Year for 2025 is “slop”.

First of all, what a difference a year makes. In late 2024, my friend Fonz texted me one morning thinking that the prediction market for who would be Time’s Person of the Year was really mispriced. Sam Altman was the favorite. I agreed with him immediately. “Prediction market audience is nerds. Anyone living in the real world knows would expect TSwift.” I mean 2024 was the year regular people paid a meaningful portion of their annual post-tax income to see the Eras tour live.

Well, a year later, prediction markets are normie enough that South Park lampooned them and “slop” is being discussed by…the dictionary. An institution with the heat signature of ancient dirt.

AI promises productivity. For a given amout of time, either

  • more output holding quality constant, or
  • higher quality.

The concern here is that AI slop is the efficient spam leveraging of plagiarism. I used Gemini to get the “L” in that acronym. There’s probably a word for the semantic/syntactic recursion, but I’m out of tokens so I’ll just sit here and spare trees but not know stuff.

I kid. I’m not out of tokens. I’m paying $100 a month to spew carbon instead of looking at sponsored results.

Ok, this time I am kidding. I still use Google. To look up store hours.

Anyway, AI slop is everywhere. My kid came into the room excited because he saw some genetic engineering video that had him think dragons could be real. Bruh, we expect this from grandma but you’re growing up in world where you know better than to trust your eyes.

[Pause for a community sigh. I’m good now. You good? Onwards.]

Before we get to my opinions, I want to share an extended thought from Brent Donnelly’s It’s not just X. It’s Y.

[Brent’s fantastic daily market letter is paywalled but I asked him to unlock this one because I really appreciated this section which will have broad appeal, so thank you Brent!]

Brent:

Caveat Lector

The biggest problem I have encountered with the recent trend towards decentralized content (Medium, Twitter, Substack, etc.) is that the writing has often not been edited for clarity, legibility, or accuracy. While it’s cool to begrudge the gatekeepers, there is value in having a reliable editor who will filter for bloat and garbage and egregious factual errors so that you don’t have to do it yourself. Random non-experts have been offered various platforms where they can disseminate objectively wrong information in essay format. If those essays tap into the right vibes, they will go viral. They can be dense and full of obvious factual errors. That will not matter.

This problem has become exponentially worse now with AI. I boosted a tweet late last week, and upon further review realized that I cannot tell whether or not it’s AI.

[Kris: I also boosted that Tweet despite knowing that parts of it were written by AI. I know Jared and that post is in keeping with this beliefs and the fact that at least the latter half of it is written by AI, I’ll address below. By the way, there are a few sentences which scream AI even if you are just skimming. I didn’t excerpt the full section by Brent but he includes some solid tips for recognizing AI writing. Anyway, back to Brent…]

So I deleted my boost. I put the tweet into an AI checker and Gemini; the checker said the tweet was 100% AI, whereas Gemini wasn’t totally sure.

Compounding the problem is that all people, including good writers, write in a voice that is an amalgamation of:

  1. Their own conversational voice
  2. Society’s accepted parameters around the style or type of writing they’re trying to produce, and
  3. The voice of everything they have ever read. The more people read AI text, the more their honestly-generated and original writing is still going to sound like AI. Just like Kurt Cobain kinda sorta accidentally copied the chords from More Than a Feeling because he was listening to a lot of Boston albums in the Nevermind days… Writers will accidentally sound more and more like AI unless we’re careful.

As a consumer of financial journalism and of writing in general, I am now at the point where I assume everything is AI and then work backwards to figure out if it’s not, based on the author and the publisher. I know if I’m reading Ben Hunt, Jared Dillian, or Noah Smith (for example), it’s not AI. If I’m reading yet another piece of financial nihilism on Substack or Twitter, it probably is.

[Kris here again. I especially liked this section which recommends ignoring financial nihilism pieces, which are now associated with virality. Oh the adverserial attention game is more boring when you recognize it. I must note that applying Brent’s rule to kyla scanlon would be a false positive—she’s been on that beat for a while with a sharp, grounded perspective. You might even say that the application of Brent’s rule to her would make her a victim of her own success. Back to Brent…]

This slop problem is good news for legit publications like Bloomberg because at some point, many people like me will find the effort to filter out the AI-generated garbage too onerous and migrate back to properly gatekept content. Much like if the FDIC got rid of deposit insurance, everyone would put their money at JPM. It’s too much work for everyone to have to vet everything all the time. Gatekeepers have bias and risk, but they also have utility. They have fact checkers and professional writers. Decentralization is overrated.

This move back towards gatekeepers is evident in the rise of The FP and the surprising success of the NYT in recent years. People don’t want random, unedited rants full of factual errors. But that’s what we’re getting from Substack and Twitter. And it’s going to get worse. I am noticing AI-generated slop all over the place, even in company press releases. Check out the unending stream of gibberish press releases coming out of SMX, for example. As AI would say: This is not just an inconvenience—it’s the critical new reality.

Here’s my approach to content consumption in 2026:

  1. Assume long form articles on Substack and Twitter are AI-generated unless there is reason to believe otherwise. When in doubt, filter it out. I don’t have time to extensively vet every single author and article. Best to over-filter quickly, not ingest a ton of stochastically parroted slop. Substack and Twitter are not inherently bad, but I need to be vigilant.
  2. Prioritize content from legitimate gatekeepers like Bloomberg and Reuters and anything that’s worth paying for. If it’s free, it’s suspect. If I am willing to pay $10 / month, it’s probably not.
  3. Ignore financial nihilism. Cynicism and nihilism were cool in high school, and they sound smart on Substack and Twitter. But they lead you nowhere. This doesn’t mean you should never be bearish. It just means that no amount of wishing we were still in the 1990s or 1950s will bring us back there. Successful traders are open-minded and forward-looking.
  4. Delete Twitter off my phone. I will use X at work, and that’s it. It’s an incredible timesuck and mental health wrecker mostly promulgating hate, falsehoods, nihilism, and negativity. Minimum viable dose only.
  5. Mute aggressively on Twitter. Mute users, mute words, mute conversations. If something bugs me on Twitter, I mute it; I don’t engage with it. Let them tell you the dollar has lost 97% of its value. Don’t waste your time correcting people who dish out obviously wrong information or who are writing fan fiction about imminent bank collapse, silver prices in Tokyo, or repo. Just chuckle and mute.
  6. Filter out all permabears, angry people, permabulls, nihilists, and captains of clickbait. Know the bias of every author you read and filter accordingly. What is useful and what gets boosted are two different things.

Finally, I will try to be the best gatekeeper I can possibly be. All my writing is edited and fact checked, but I have still made the mistake of boosting AI-generated content a few times and I still make factual errors. I will make a strong effort not to do so in future, or to advise readers as soon as I’m made aware of a mistake or AI boosting.

Thanks again to Brent for letting me share that.

Manager mode

I mentioned that I boosted Jared’s tweet despite a significant portion of it being written by AI. This is not confirmed but it doesn’t matter because what I’m about to say is only interesting insofar as I’m giving cover to AI writing.

I personally don’t care if what I’m reading is written by an AI if the message was the author’s intent (so long as it wasn’t plagiarized*) just as I don’t really care if the president didn’t write his own speech.

*Brent is concerned about plagiarism while recognizing the Cobain problem. Well, the Cobain problem is insidious and everywhere. I keep a file of “turns of phrases” I like. Am I to believe that my mind hasn’t recycled some of that indirectly? I always give credit in this letter when I can remember that there was credit to be given (and my notes are very good at keeping track because credit is a priority). I don’t think I have ever plagiarized. But I wouldn’t sell the tail option on that because we all know thoughts can be recombined inputs without us being aware of it. Just look at the opening of Andrew Courtney’s latest piece where he basically writes a post and scraps it because I once covered the same topic. We chatted about this. Sometimes it’s hard to know where you begin and your inputs end. But like porn, you probably know plagiarism when you see it. Fyi, the piece Andrew actually pivoted to is excellent, but I’m biased because I feel the same as he does.

Instead, I think of AI output that you share as YOUR agent. If you let AI write for you and present it as your own, you are responsible for those thoughts. You don’t get to enjoy the benefits of production without accountability. You can’t disclaim what you say with “I didn’t write that”. If you platform a bot, it’s on you.

As far as I’m concerned Jared approved a PR and is responsible for what’s in prod. So long as he maintains accountability, this is not only a valid workflow, but it’s the new default whether you realize it or not.

A few months ago I shared Venkatesh Rao “sloptraptions” post Prompting is Managing

It argues that concerns of using AI as a crutch misunderstand the nature of using it effectively. Interacting with LLMs is not a form of individual cognitive “doing”. It’s a shift into supervisory control, where the user’s brain state mirrors that of a manager overseeing a junior. Rao argues it is actually the standard cognitive signature of management necessary for high-level coordination.

I don’t feel brain-dead when I’m orchestrating multiple agents across different projects. Instead, it feels like writing outlines for articles or pseudocode for projects. It’s a different form of executive function. It feels managerial. It’s not my favorite kind of work, but it’s productive and necessary.

Hmm…that sounds an awful lot like the management aspect of any job. It’s tedious but sits at the heart of leverage.

I believe it was Agustin Lebron who said most jobs of the future are going to be “shit umbrellas”. Bots will do the actual work, but they can’t be held accountable. Humans will be paid to absorb decision risk rather than actually doing things. That sounds right.

Before AI, people spewed plenty of slop. They were held accountable. Either by the law, the market, or the judgement of their peers. The accountability will stay even if the transmission syntax and medium change.

To address AI making us lazy, I already posted slow is smooth and smooth is fast. It’s a concern but you’re hardly defenseless (that post resonated, it might have been the most popular non-trading one I wrote last year).

Scott H. Young wrote a post Will AI make us stupid? dealing with similar themes. He explains how we’ll bifurcate according to how they employ these tools. This is how I see it:

There are things that we don’t NEED to do because computers are better, but the act of doing them anyway changes you. You will need to actively decide what to continue doing and what to outsource. You will not always make the right decision.

Here’s Scott’s view:

Learning requires an investment of effort, and AI will make us stupider if that effort is avoided.

At the same time, not all effort in learning is helpful. Much of learning involves cognition that does not directly contribute to understanding and knowledge. Think of learning like powering a motor—all learning requires a source of energy to make progress, but not all energy is transformed into forward motion. Depending on the vehicle, much of it might be wasted as heat and noise.

Therefore, while I think AI is probably going to result in an incredible “dumbing down” of our self-education in the average case, it is probably also going to enable more careful students and teachers to facilitate learning much better than before. Because while AI can simply solve a problem for you, it can also generate worked examples, practice problems and feedback, and guide problem-solving dialog.

Getting this balance right is hard, and I don’t think it’s simply a matter of laziness. Even intelligent students are often wrong about what effort actually matters when it comes to learning, and many teachers are no better. This has always been the case, but AI has raised the stakes as the ability both to enhance learning and to bypass it entirely have expanded.

Thus my personal prediction is that in domains that are already largely under the powers of modern AI, such as languages, programming or chess, we’re going to see a divergence in human abilities. The average person will rely on the AI more, robbing them of the ability to learn the underlying skills. More sophisticated students will use AI to learn better, removing inefficiencies that were unavoidable in pre-AI learning environments.

Some evidence of this is already emerging…