dimveil

where AI meets money.
the plays nobody's posting · you're early

[dimveil @dimveil]

@rugikkk Damn this is good, and the ranking bit is the part nobody gets, ordering and ranking feel identical so ppl skip the second one and end up perfectly sequenced on stuff that barely moves the goal

[dimveil @dimveil]

Wall Street paid for a cartoon in 1956 to sell ordinary clerks one habit, and it's still the quietest way people seem to build wealth today.

Not a stock tip. Not a system. A rhythm.

The idea was a fixed sum into the same thing every single month. When the price drops that money buys more shares. When it climbs it buys fewer. No timing the top. The average does the arithmetic most people were never going to do in their heads.

The exchange had a reason to animate this. In 1956 it was running a national campaign to convince regular people the market wasn't only for the rich. 3 out of 4 shareholders back then earned under 10,000 dollars a year. So the message had to be simple enough to draw.

They gave the habit a committee name. The Monthly Investment Plan. Forgettable on purpose.

70 years later the same idea got a cleaner label. Dollar-cost averaging. Same math, repackaged, sold back in podcasts and threads as if someone just discovered it.

Almost nobody stuck with it, because it's boring. No screenshot moment, no 100x, no story to tell at dinner. Just the same small buy, month after month, while everyone around them chased the thing that moved fast that week.

The best-known money lesson ever put on film is 70 years old, free, and public domain. It fits in one line.

The same amount, every month, no matter the price.

[dimveil @dimveil]

Bill Hwang said all his money couldn't fix one bad year, 3 years later that money hit $30B, then vanished in 2 days.

In 2018 he sat on a stage and told a room he once had a bad business problem. He had a lot of money by then, and it was public, and none of it helped him.

He meant it as a faith story. It reads now as a warning.

By March 2021 his family office, Archegos, held $160B in market exposure. Built on borrowed money. Total return swaps that let him bet 5x what he actually had, through banks that never saw the whole picture.

His personal wealth ran from $2B to over $30B in under a decade. One of the largest fortunes on earth, and almost nobody knew his name.

Then a few stocks dropped. The margin calls came. He couldn't meet them.

The banks dumped every position backing him at once. Credit Suisse took losses that helped end a 167-year-old bank. Nomura bled billions.

Hwang went from $30B to $55M in under a week.

He wasn't wrong about most of his picks. Some recovered later. At half the leverage a 20% drop is a bad month, not an extinction.

The leverage didn't make him rich. It made him fragile.

He told a crowd his money couldn't save him. Then he borrowed enough to prove it twice.

[Rugikk @rugikkk]

https://t.co/wEOKLYneMh

[dimveil @dimveil]

Jonathan Gruber walks to an MIT blackboard and draws the price of your entire life in two lines.

He teaches 14.01, the intro economics course MIT puts online for free. Most people scroll past it. What he draws is the thing no one sits you down to explain.

It's called the labor budget constraint. One line, two ends.

At one end: 24 hours of leisure, zero consumption. You do nothing, you earn nothing.

At the other end: zero leisure, and consumption equal to 24 times your wage. That's it. That's the ceiling. Every hour awake, sold, at your rate.

Sit with that second number. 24 times your wage is the theoretical maximum a human being can earn selling time. Not a lot for most. A hard wall for everyone. A surgeon at $500 an hour hits the same wall as everyone else, just higher up. There are 24 hours, sleep included, and not one more.

Gruber even says it out loud. You have 24 hours in a day. Sleep is in there too. That is the whole budget. You are not negotiating for more hours. You are only negotiating the price of the ones you have.

And here's what the blackboard quietly proves. Selling time is the only form of income with a hard cap written into the math. Own a thing and it earns while you sleep. Build a thing and it sells while you sleep. Sell your hours and the moment you stop, it stops.

Every raise you've ever chased was a move along that same line. Never off it.

The lecture is free. MIT posted the whole thing. Almost nobody who trades their hours for a living has ever watched the arithmetic that prices them.

[dimveil @dimveil]

https://t.co/DxXmdQVjza

[dimveil @dimveil]

@EBabalyk yep, building it is engineering. obeying it is character

[dimveil @dimveil]

Jim Simons built the machine that ran his money, then wrote one rule: he was never allowed to override it.

Think about how strange that is.

You build the thing. You hire the physicists.
You feed it the data by hand, sending people to the Federal Reserve to copy interest rates that didn't exist online yet. And then you sign a rule that says your own judgment is banned from the room.

Everyone else does the opposite. They build a model, and the moment it says smth that scares them, they overrule it. Simons called that not science. You can't backtest how you felt when you got out of bed of that morning.

He learned it the hard way. For years he traded the human way, read the news, formed a hunch, placed the bet. Some mornings he walked in a genius. Next morning an idiot. He was good at it. He still couldn't stand it.

So in 1988 he flipped the switch to 100% models and swore off touching it. At 3:13 he explains the rule in his own words. Whatever the thing says, you do, no matter how smart or dumb you feel in that moment.

The results are the part nobody outside the firm believes. Roughly 66% a year for three decades. The Medallion fund closed to outsiders in 1993, bought them all out by 2005, and since then it's owned by the 300 people who work there. It runs. They mostly stay out of its way.

Jim Simons died in 2024, one of the great mathematicians of his century. He is remembered as the man who solved the market.

He didn't solve it by being right. He solved it by building smth that didn't need him to be.

[dimveil @dimveil]

https://t.co/DxXmdQVjza

[dimveil @dimveil]

@koaxbt you goddamn right, money was never the point.time was.

[dimveil @dimveil]

He built a machine that made other people $58 billion, then sold every share and walked out the door.

His name is Ray Dalio. In finance everyone knows it. Almost nobody has watched him stand at a podium and explain the thing he actually built.

In 1975 he started Bridgewater from a two-bedroom apartment. He was 26. For years it was cold calls from a spare bedroom. In 1982 he made one public prediction so wrong it nearly wiped him out, and he borrowed $4,000 from his father to cover the family bills.

Then something turned. Not his effort. His structure.

He stopped selling his hours and built a culture that could run the fund whether he was in the room or not. He wrote down the rule behind every decision and turned the rules into a system.

That is the part almost nobody copies.

By 2020 the machine had made its investors more than $58 billion, more than any hedge fund in history. At its peak it ran over $150 billion.

Here's the tell. He stepped down as CEO in 2017. Gave up the investment seat in 2022. Sold his last share and left the board in 2025.

The machine still runs. It manages roughly $92 billion today. It does not need him to walk into any room.

At 4:04 he stops talking about people and explains the mechanism. Every meeting taped, anyone in the company can listen, nobody discussed behind their back. A system built to find the truth without him in it.

He spent 47 years building something that would finally pay him without him.

[dimveil @dimveil]

https://t.co/DxXmdQVjza

[dimveil @dimveil]

@triiiam booked 37 mins for this tonight. the dollar-weighted vs time-weighted gap is one of those ideas you can't unsee once it clicks. saving this.

[dimveil @dimveil]

https://t.co/DxXmdQVjza

[dimveil @dimveil]

Bethany McLean broke the biggest fraud in corporate history with one question, and Enron's CEO hung up on her for asking it.

Early 2001. Enron is the "It Stock." 7 straight years of Wall Street calling it a genius. $60 billion market cap. Fortune's most innovative company in America, 6 years running.

A 30-year-old reporter at Fortune reads the financials nobody else bothered to open.

She can't find the answer to one thing. How does the company actually make money?

She calls Enron. Skilling, the CEO, tells her the question is unethical. Then he hangs up.

Executives fly to New York to calm her down. Ken Lay calls her editor and leans on him to kill the piece. The editor says no.

March 2001. She runs 2,000 words under the meekest headline in journalism history. "Is Enron Overpriced?"

The stock was $80 at the time.

8 months later Enron filed the largest bankruptcy America had ever seen. 20,000 jobs gone. The stock hit 26 cents.

Everyone had the same public filings she did. The math was sitting there the whole time.

She just read them.

The smartest guys in the room got beaten by a reporter who asked how.

[dimveil @dimveil]

Bethany McLean broke the biggest fraud in corporate history with one question, and Enron's CEO hung up on her for asking it.

Early 2001. Enron is the "It Stock." 7 straight years of Wall Street calling it a genius. $60 billion market cap. Fortune's most innovative company in America, 6 years running.

A 30-year-old reporter at Fortune reads the financials nobody else bothered to open.

She can't find the answer to one thing. How does the company actually make money?

She calls Enron. Skilling, the CEO, tells her the question is unethical. Then he hangs up.

Executives fly to New York to calm her down. Ken Lay calls her editor and leans on him to kill the piece. The editor says no.

March 2001. She runs 2,000 words under the meekest headline in journalism history. "Is Enron Overpriced?"

The stock was $80 at the time.

8 months later Enron filed the largest bankruptcy America had ever seen. 20,000 jobs gone. The stock hit 26 cents.

Everyone had the same public filings she did. The math was sitting there the whole time.

She just read them.

The smartest guys in the room got beaten by a reporter who asked how.

[dimveil @dimveil]

Most people building AI search in 2026 are losing 30% of their accuracy and blaming the model.

It's not the model. It's one line of code they never learned.

There are 2 ways to measure how close two sentences are. One reads the angle between them. One reads the angle and the length. On unnormalized vectors, picking wrong quietly poisons everything you retrieve.

They copy a tutorial. They never check which one they're using. Then they pay a consultant $15,000 to "fix the retrieval," the consultant adds one normalize line, and everyone calls it magic.

A professor explained this exact trap in a free lecture. Recorded in 1999. Still up.

I put the whole thing in one piece: the math, the money, and the 26-year-old who turned it into $31,000 in 4 months.

The people winning aren't smarter about what's next.
They're calmer about what's already free

[dimveil @dimveil]

https://t.co/2aaMooZYyA

[dimveil @dimveil]

https://t.co/2aaMooZYyA

[dimveil @dimveil]

@seeconvm same thing happened to me ditched the 3h videos, started linking every concept in notes now i can build a tiny net from scratch in one sitting instead of rewatching the same tutorial for the 4th time

[dimveil @dimveil]

6 mistakes kill most Claude Skills, and 5 of them happen before you ever run it

Your description is too vague. Claude reads it to decide when to trigger, so two lines of instructions and one line of description gets you silence. Describe the trigger, not the task.

You skipped the evaluation. An untested Skill fails quietly, not loudly. Run the eval before you install.

You built one mega Skill. Monolithic Skills are impossible to debug, and one broken piece takes down every workflow. Small, focused, composable.

You never tested the negative case. A Skill that fires on everything is as broken as one that never fires. Send 3 unrelated requests and confirm it stays quiet.

You put no tools inside. Instructions alone make a prompt, not a Skill. Deterministic scripts beat regenerated code every time.

You fixed the output, not the Skill. Patching the answer means the same failure returns next session. Bake the correction into the SKILL.md.

Debug trick that takes 10 seconds. Ask Claude "when would you use this skill?" It quotes your description back. You'll hear exactly what's vague.

Which of the 6 is yours?
Small skills. Big reliability.
I break these so you don't have to.

[BLAZT @BLAZT_Ai]

https://t.co/87qH5XW7NA

[dimveil @dimveil]

Just hooked Voicebox up to Claude Code. The cloning from a short clip sounds pretty natural and it's all local which is nice. Hearing it speak code actually helps catch weird logic faster. Out of the engines it has which one have you been using the most for technical stuff and longer sessions?