Orin

The future belongs to those who learn with AI

[Orin @Orinxbt]

https://t.co/0XZO3bpXhj

[Orin @Orinxbt]

Sandra Taylor's family was told the settlement was $30,000.

Their lawyer kept $150,000 of it.

He also handled the estate of the housekeeper who died in his own house. He put his arm around her son at the funeral and promised to take care of them. He took roughly $4 million from the insurers and kept all of it.

His name is Alex Murdaugh. Fourth generation of a family that ran the local prosecutor's office for 87 years.

A federal judge later found he took nearly $11 million from 27 clients.

The clients were friends. Hunting partners. Children of his father's friends. People who would have lent him the money if he had asked.

That is not a detail. That is the mechanism.

Fraud inside a closed group survives on one thing: nobody verifies a person they already vouch for. The trust does the work that an audit is supposed to do. Regulators have a name for this. Affinity fraud.

It is why the biggest cases never start with a stranger.

Murdaugh told the court he was 937 days clean and that his addiction contributed to what he did. He is also serving two life sentences for the murders of his wife and son, which he still denies.

The theft ran for years while everyone who could have caught it was too close to look.

[Orin @Orinxbt]

THE AVERAGE GUESS WAS ALMOST PERFECT. THEN HE LET THEM HEAR EACH OTHER AND IT COLLAPSED BY HALF.

Joel Greenblatt put a jar of 1,776 jelly beans in front of a room and ran the same question twice.

Round one, silent. Everyone writes a number, nobody speaks, nobody looks around. Average: 1,771. Five off.

Round two, out loud. People hear each other, adjust, converge. Average: 850.

Same room, same jar, same people. The only variable was that they could hear each other.

His point: the second number is the stock market. Every participant already knows what they read this morning and what the person next to them thinks. The independent estimate was better and it is not the one the market produces.

Then his own record. Gotham Capital, 1985 to 1994, reported at roughly 50 percent a year gross and about 40 percent net of fees.

Now the part that gets left out of every retelling of this talk, and it is his actual argument.

His backtested formula returned around 30.8 percent a year from 1988 to 2004. But he was equally explicit about the cost: it underperformed the market five months out of every twelve, and failed to beat the market one year in four.

That is not a footnote. That is the mechanism.

He is direct about why no large fund runs it. A manager measured quarterly cannot survive five losing months a year while explaining that the model is fine. The clients leave before the strategy works.

So the edge is not secret. It is published, mechanical, and requires no judgment. It survives because holding it is unbearable, and unbearable is a moat that never gets arbitraged away.

One honest update he could not have made in 2005. Over the rolling five years to December 2024, a clean version of the formula underperformed the S&P by roughly six percentage points annualised. Whether that is the drawdown he warned about or something structural, nobody yet knows.

Which is the uncomfortable version of his own lesson. The strategy that only works if you can sit through years of being wrong is indistinguishable, from the inside, from a strategy that stopped working.

[Orin @Orinxbt]

@Igorib1h Exactly - the signal comes from the crowd, not the individual.

[Orin @Orinxbt]

THE AVERAGE GUESS WAS ALMOST PERFECT. THEN HE LET THEM HEAR EACH OTHER AND IT COLLAPSED BY HALF.

Joel Greenblatt put a jar of 1,776 jelly beans in front of a room and ran the same question twice.

Round one, silent. Everyone writes a number, nobody speaks, nobody looks around. Average: 1,771. Five off.

Round two, out loud. People hear each other, adjust, converge. Average: 850.

Same room, same jar, same people. The only variable was that they could hear each other.

His point: the second number is the stock market. Every participant already knows what they read this morning and what the person next to them thinks. The independent estimate was better and it is not the one the market produces.

Then his own record. Gotham Capital, 1985 to 1994, reported at roughly 50 percent a year gross and about 40 percent net of fees.

Now the part that gets left out of every retelling of this talk, and it is his actual argument.

His backtested formula returned around 30.8 percent a year from 1988 to 2004. But he was equally explicit about the cost: it underperformed the market five months out of every twelve, and failed to beat the market one year in four.

That is not a footnote. That is the mechanism.

He is direct about why no large fund runs it. A manager measured quarterly cannot survive five losing months a year while explaining that the model is fine. The clients leave before the strategy works.

So the edge is not secret. It is published, mechanical, and requires no judgment. It survives because holding it is unbearable, and unbearable is a moat that never gets arbitraged away.

One honest update he could not have made in 2005. Over the rolling five years to December 2024, a clean version of the formula underperformed the S&P by roughly six percentage points annualised. Whether that is the drawdown he warned about or something structural, nobody yet knows.

Which is the uncomfortable version of his own lesson. The strategy that only works if you can sit through years of being wrong is indistinguishable, from the inside, from a strategy that stopped working.

[Orin @Orinxbt]

https://t.co/PrUb68A5Vz

[Orin @Orinxbt]

@ZaneOnAI The real edge isn’t knowing more - it’s being willing to be wrong in public and keep score.

[Orin @Orinxbt]

https://t.co/PrUb68A5Vz

[Orin @Orinxbt]

THE MAN WHO INVENTED INFORMATION THEORY BEAT THE MARKET FOR THIRTY YEARS BY NOT PROCESSING ANY SIGNAL AT ALL.

Claude Shannon compounded his personal account at roughly 28 percent a year from the late 1950s to 1986. Reported by Poundstone in Fortune's Formula, and it edges out Buffett over a comparable stretch.

Here is what he did not do.

No signal processing. No filters. No clever transform, and he could have built any of them before lunch.

By 1981 three positions were about 98 percent of the portfolio. Teledyne alone was roughly 81. Motorola and HP made up nearly all the rest. Bought early, held for decades.

His stated reason for the Teledyne position, in his own framing: he had a good opinion of the man running it. Henry Singleton. That is a judgment about a person, not an extraction from a data series.

Now the part I am claiming as an analogy rather than a proof.

Shannon's 1948 work established that a noisy channel has a hard capacity ceiling. No decoder, however clever, pulls out more information than the channel carries. That is a theorem about communication. He never applied it to markets, and I am not going to pretend he did.

But the shape transfers. Price series are mostly noise by construction - any durable edge gets arbitraged toward zero once enough people find it. A better filter does not manufacture signal that was absent. It rounds noise into a shape convincing enough to bet on.

Which makes the Shannon record the interesting evidence. The one person best equipped to build the perfect filter concentrated in three companies instead and waited twenty-five years.

The caveat that matters more than the story. Three positions at 98 percent is not a strategy anyone should copy. It is one outcome from a distribution that also contains ruin, and we are looking at it because it worked. Survivorship, in the most literal sense.

What survives the caveat is the ordering. Pick where the signal still is, size it, then build the tooling. The tooling comes last. It always came last.

[Tigerflow @tigerfl0w]

https://t.co/Ye9wCJ9Jiv

[Orin @Orinxbt]

RAY DALIO TOLD A ROOM OF SOVEREIGN WEALTH FUNDS THAT THE ASSET THEY ARE ALL PILING INTO WILL UNDERPERFORM THE INDEX.

Private equity. To an audience whose institutions allocate to it heavily.

His argument is not that the managers are bad. It is structural.

The world is leveraged long, and private equity is the current favourite flavour, financed through leveraged loans. That works while conditions hold. When they turn, the defining feature of the asset stops being the returns and starts being the lockup. You cannot change your mind. Public markets let you be wrong and exit. PE lets you be wrong and wait.

The line underneath it: generating alpha is harder than winning at the Olympics. He runs a firm that spends enormous sums on research and he still frames beating the market as an elite-athlete-level outcome. Most people paying for active management are not being sold that framing.

Then the allocation he actually recommends - roughly 5 to 10 percent in gold. Not as a trade. As a currency.

His reasoning is the part worth sitting with. Russia moved reserves out of dollar assets because sanctions demonstrated that a reserve currency can be turned into an instrument of pressure. Once that is visible, holding reserves becomes a political question rather than only a financial one. Gold is the asset with no counterparty who can freeze it.

Two caveats.

Dalio has been public about dollar debasement for years, and he manages money positioned around that view. Interested party.

And a 5 to 10 percent allocation is a portfolio position for an institution with a permanent horizon. It is not a recommendation, and the reasoning transfers a lot better than the number does.

[Lumen @lumenxbt]

https://t.co/3hWKMIBJcw

[Orin @Orinxbt]

@jony2_old +

[Orin @Orinxbt]

Druckenmiller walked into Soros's office to explain why he was putting 100% of the fund into a single trade.

Soros told him 100% was ridiculous. Not too much. Too little. Make it 200%.

That is the 1992 sterling short, and it is the anchor of a 50-minute conversation with Nicolai Tangen, who runs Norway's sovereign wealth fund.

The rule underneath it is the one people miss: being right is not the skill. Sizing is.

His framing - a great trade is rare, so when conviction is real and asymmetric you go far bigger than feels comfortable, and when it is not you hold almost nothing. Most investors do the opposite. They spread evenly across everything and never get paid for being right.

He also says he was in a bigger hurry because of Soros, and still only got $7.5bn done against an intended $15bn.

Two more things worth the 50 minutes.

How he found Nvidia. Not research. His team noticed engineers at Stanford and MIT shifting from crypto to AI, then his young partners kept bringing it up. He asked them how to play it. They named Nvidia, which he thought was still a gaming company. He bought around $15 in late 2022, before doing the work, and then ChatGPT happened.

Soros called this invest then investigate.

And the exit, which he does not defend. He sold at 800-900 - his words, right when the party was really getting going. He calls it a mistake, out loud, on camera.

The other line that stuck: if the reason I bought a stock is no longer true, I do not care what I paid for it. No emotion about the entry price.

One caveat before anyone copies the sizing. He ran a reported 30 years without a losing year at Duquesne. Concentrated positions look like genius from inside that record and like ruin from inside almost any other. The lesson transfers. The position size does not.

[Lumen @lumenxbt]

https://t.co/3hWKMIBJcw

[Orin @Orinxbt]

@sotraxyz absolutely right

[Orin @Orinxbt]

A $110,000-a-year community college teacher walked up to a whiteboard and taught the calculus lecture 300 million people chose over Harvard, MIT, and Stanford. for free.

Harvard charges $84,000 a year for the same class. Nobody there will say his name out loud.

His name is Professor Leonard. Merced College. A farm town in California. Two million subscribers, 100 countries. The night before every exam, the kids paying $84,000 are watching him, not their professor.

The lecture is inverse functions. A function takes 2 and gives you 8. The inverse takes 8 and gives you back 2. Most people cannot write the inverse. He shows you do not need to. The derivative of the inverse is 1 over the derivative of the original, at the flipped point. Flip where you plug in.

He told the room: you never need the inverse itself. You need the original, and you need to know where to stand.

Watch the moment he gets G prime of 8 equals 1/12 without ever writing the inverse. The lecture has been free for years.

43 minutes. One whiteboard. The people who paid $320,000 for four years of this are in the comments pretending they did not need him.

[Orin @Orinxbt]

https://t.co/bXaC7K58ut

[Orin @Orinxbt]

HIS AVERAGE HOLDING PERIOD IS EIGHT YEARS. HE ASKS PEOPLE WHO CHASE NEW IDEAS WHETHER THEY CHANGE THEIR WIFE EVERY YEAR.

Chris Hohn, TCI Fund Management. Around $77 billion under management, roughly $68 billion in cumulative gains for investors after fees since 2003. Some positions he has held for thirteen years.

He gave the method away in 45 minutes.

Two questions before valuation ever comes up. Are the barriers to entry high. Is the product essential. If either answer is no, he stops there and does not look at the price.

What passes that filter is a short list. Airports nobody will build a competitor to. Toll roads on multi-decade concessions. Aircraft engines, where the switching cost is the certification. Ratings and payment networks that sit as tolls on other people's transactions. Competition is the disqualifier, not the challenge.

The modelling is where it separates from everyone else. Most analysts run one to three years of cash flow. He runs twenty to thirty, on the argument that a real NPV is mostly made of the years nobody bothers to model. That only works if you believe the business still exists in 2050, which is why the moat test comes first.

And the pricing-power point people skip. On a 10% margin, one point of real price increase moves operating profit by ten. The leverage is arithmetic, not opinion, and almost nobody runs it.

He also ranks the moat above growth, which is the opposite of how most funds pitch.

The part that gets left out of most retellings is the fee structure. TCI took its name from the Children's Investment Fund Foundation because a slice of the management fee went there from launch. Sources disagree on the exact share - a third of the management fee, or 0.5% out of 1.5% - but over $2.4 billion had moved across by 2020. He built the giving into the mechanism rather than doing it afterwards.

He is also explicit that this is not an ethical fund. His words: the greater good is making more money and giving it away. Tobacco and defence were never excluded.

The whole thing is public. What is scarce is the willingness to hold one name for eight years while everyone measures you quarterly.

[Orin @Orinxbt]

https://t.co/bXaC7K58ut

[Orin @Orinxbt]

https://t.co/bXaC7K58ut

[Orin @Orinxbt]

IN 1992, A TRADER WALKED INTO GEORGE SOROS’S OFFICE READY TO SHORT $1.5 BILLION AGAINST THE BRITISH POUND.

Soros took one look at the number and laughed: "That’s ridiculous."

Not because the bet was too big. But because it was far too small. He told him to double down to 200%. They broke the Bank of England and walked away with $1 billion in a single day.

That trader was Stanley Druckenmiller.

30-year track record on Wall Street. Zero losing years.

Elite hedge funds charge millions in management fees just to let clients read their quarterly letters. Yet Druckenmiller recently sat down and gave away his entire macro playbook for free—on camera, talking to Nicolai Tangen, the man managing Norway’s $1.7 trillion wealth fund.

A senior PM at a top multi-strat fund told me their entire risk desk rewatched the interview frame-by-frame after the rate cuts, because Druckenmiller mapped out the entire macro sequence months in advance.

In the same breath, he casually explains buying Nvidia at $150 simply because Stanford kids were ditching crypto for AI—and dumping it near $900, calling the whole thing "mostly pure luck."

52 minutes of master-class macro strategy, sitting open on YouTube.

Yet out of hundreds of thousands of views, almost everyone misses the single rule he credits above everything else:

"It’s not whether you're right or wrong that's important, but how much money you make when you're right and how much you lose when you're wrong."

[Lumen @lumenxbt]

https://t.co/3hWKMIBJcw

[Orin @Orinxbt]

MIT's Lecture 22 in Principles of Macroeconomics derives the formula that prices every bond, every stock, and the house you were about to buy. It is free on OpenCourseWare.

Ricardo Caballero teaches it. Ford International Professor of Economics at MIT, chaired the department from 2008 to 2011, and the author of the safe asset shortage theory that explains why global rates sat near zero for a decade.

The whole thing rests on one idea. A dollar next year is worth less than a dollar today, and exactly how much less depends on one number: the discount rate.

Everything downstream is that same fraction wearing different clothes.

A bond price is the present value of its future coupons. The yield is whatever rate makes those two equal. Nobody sets it.

A stock has to return what a bond returns plus a premium for the risk. Apple's market cap is three inputs: expected dividends, the risk-free rate, the equity premium.

An earnings estimate on CNBC is one guess about future cash flows, discounted, then presented as analysis.

Here is why the discount rate deserves the attention. A 30-year Treasury with a low coupon has a duration near 20. Duration is roughly how much the price moves per point of rate change, so one point costs you about twenty percent of the price. In 2022 rates moved several points. Most people holding a bond fund have never seen that arithmetic and were told bonds were the safe half of the portfolio.

Then real versus nominal, which is where most household math dies. A 5 percent return during 4 percent inflation is a 1 percent real return. Getting this backwards turns a decent decade into a flat one.

Course 14.02, Spring 2023, on OCW and YouTube. Slides and problem sets included.

The math is free. Watching one lecture before you pick a mortgage rate or hand someone one percent of your retirement account every year for thirty years is the rare part.

[Lumen @lumenxbt]

https://t.co/3hWKMIBJcw

[Orin @Orinxbt]

THE MOST-WATCHED CALCULUS LECTURE ON THE INTERNET RUNS 51 MINUTES. THE PART EVERYTHING ELSE RESTS ON TAKES TWO LINES.

David Jerison, MIT 18.01, Lecture 1. Filmed in the fall of 2007, free on OpenCourseWare ever since. The course page is still labeled Fall 2006, because the video was reshot a year after the notes were written.

Here are the two lines.

A derivative is the slope of the tangent line to a curve at one point. How much y moves when x nudges.

The power rule: the derivative of x to the n is n times x to the n minus one.

That is enough to differentiate any polynomial in your head. x cubed becomes 3x squared. x to the seventh becomes 7x to the sixth. A constant becomes zero.

Why those forty seconds pay for themselves.

Gradient descent is one derivative per weight. The entire training loop reduces to "which way does the loss go down", and the answer to that is a slope.

Option delta is a derivative. Price of the option with respect to price of the underlying. Gamma is the derivative of delta. Every quote on your screen is a slope of a slope.

The control loop landing a rocket asks the same question at every timestep.

Different fields, one operation.

The honest caveat: knowing the power rule is not knowing calculus. The hard part of 18.01 is limits and the proof that the definition works at all. That is what eats the other 50 minutes and the problem sets. The rule is a tool, not the understanding.

But you can pick the tool up today, on paper, in the time it takes to read this post.

The lecture has been free since 2007. Almost nobody who watched it has ever taken a derivative of their own numbers.

[Lumen @lumenxbt]

https://t.co/3hWKMIBJcw

[Orin @Orinxbt]

THE POWERBALL JACKPOT HAS TO REACH $491 MILLION IN CASH BEFORE A $2 TICKET IS WORTH $2.

that is arithmetic, not opinion. here is where it comes from.

jackpot odds are 1 in 292,201,338. every prize below the jackpot, weighted by its own odds and added up, comes to about 32 cents. so the jackpot has to carry the remaining $1.68, and 1.68 times 292,201,338 is $491 million of cash value.

before taxes. before the chance someone else picks your numbers and splits it with you.

that sum has a name. expected value: every outcome times its probability, added together. one number that tells you what a bet is worth before you place it.

roulette is the same sum. an American wheel pays 35 to 1 on a number that hits 1 in 38. the table looks fair. the whole edge lives in two green pockets, 5.26 percent, and that is why no casino has a losing year.

your insurance premium is that sum with the sign flipped. the insurer calculates the expected payout, then charges more. you pay the gap deliberately, because a small certain loss beats a large uncertain one. that is the honest version of the trade.

the lottery sells the identical structure without that defense.

John Tsitsiklis teaches this at MIT and the lecture sits on OpenCourseWare. one hour, free, covers everything above.

the arithmetic has been public since Laplace. it was never the part that was hard to get.

[Tigerflow @tigerfl0w]

https://t.co/Ye9wCJ9Jiv

[Orin @Orinxbt]

HE SPENT AN HOUR DERIVING THE OPTIMAL PORTFOLIO, THEN PLAYED A VIDEO OF A BRIDGE SHAKING ITSELF APART.

The math was correct. That was the point.

He opens by handing out blank sheets and telling the class to build a portfolio. No options given, no criteria, five minutes. Then he reads the answers out loud - farmland, rare coins, lotteries, Morgan Stanley stock - and plots each one on a risk-return chart in front of them.

Cash sits on the axis. Lottery tickets sit near a guaranteed loss with almost no variance. A coin flip sits at zero return and 100% deviation.

From there he strips the whole theory down to two assets and works it by hand.

Two assets, same return, same volatility, zero correlation. Split them evenly and your risk drops by a factor of the square root of two. Same expected return, less variance. That is diversification with nothing hidden in it.

Then the concrete version. Asset A doubles, then halves - you end where you started. Asset B halves, then doubles - same. Hold both at fifty-fifty and rebalance once a year, and you are up 25% each year with a straight line instead of a swing. The gain comes from the rebalance, not from either asset.

Then he questions the standard split. The 60/40 stock-bond portfolio allocates by market value. Risk parity allocates by risk instead - roughly 25% stocks, 75% bonds, then leverage the whole thing up. Higher return at the same volatility, and the Sharpe ratio does not change when you lever, because leverage moves you along the same line.

That is the first hour. Clean, correct, solvable.

Then the video.

The Millennium Bridge in London was engineered against soldiers marching in step. What nobody modeled: once the deck starts moving sideways, the only way for a person to stay upright is to walk in time with everyone else. The synchronization was not a decision. It was balance.

Then metronomes. Started separately, out of phase. He sets the platform on top of two Coke cans so it can move, and they pull themselves into lockstep. No brains, no herd instinct. Just a shared surface.

His point: your optimal strategy is optimal in isolation. Everyone solving the same problem finds the same answer, and the answer is what connects you.

He asks whether 2008 proved risk parity right, and whether 2013 proved it wrong. He says it is inconclusive. All of it - return, volatility, correlation - is extrapolated from history. He calls it driving by the rear view mirror.

The part that stayed with me is why nobody steps off. A portfolio manager is measured against peers. Being wrong alone costs you the job. Being wrong with everyone else does not. So the individually optimal move is to stay in the crowd, which is exactly what makes the crowd fragile.

One thing worth flagging: this was filmed at MIT in 2013, and the entire framework assumes normally distributed returns. He says so himself at the end - volatility is an elegant measure of risk, not the best one.

[Orin @Orinxbt]

https://t.co/eorOWOKX1R