xanter.ai

Grinding until the life-change hit.

[xanter.ai @0xanter]

https://t.co/xucRjy0rHi

[xanter.ai @0xanter]

You could build the best product in your industry and still watch someone else set your price.

A Yale business case looks at three entrepreneurs who built a genuinely better device for heart surgery — smaller cuts, faster recovery. It worked. Doctors adopted it. Money started coming in.

And that's exactly when the professor asks the room the real question: what happens the moment you start making money? People enter the market. Prices get bid down. She calls economics "the dismal science" for a reason — success invites competition, and the market decides what your work is worth, whether you like the number or not.

Here's what these founders couldn't control: their one competitor charged $5,000 for the same category of fix; they charged $1,000. That number moves regardless of how good the product is, the moment someone else shows up.

Here's what they could control: their training costs. They almost blew it by flying doctors to Hawaii to learn the technique — which only attracted doctors with free time, not the ones actually doing the surgeries. The fix was cheaper and smarter: train inside teaching hospitals, where the busiest doctors already are. Same outcome, lower cost, far better selection — because they finally optimized the side of the equation that was actually theirs.

Income gets fought over by the entire market, every time, no exceptions. Cost is the only lever that's fully yours. Everyone chasing the number they can't control loses sight of the one they always could.

Full lecture below.

[Superior @andreysuperior]

https://t.co/6fbpq78hsH

[xanter.ai @0xanter]

@kardinall the "you get billed for the other 96 attempts" part is what actually kills me. 72% accuracy sounds like a stat you'd brag about until you realize it means paying for 25 failed runs to get 1 that works

[xanter.ai @0xanter]

Your business partner makes one catastrophic mistake. Now they can take your house.

A Yale lecture on capitalism lays out the math nobody explains before you sign a partnership agreement: joint and several liability. If your partner does something that gets you sued — even something you had nothing to do with — the person suing you doesn't have to split the damages fairly. They can go after whichever partner has the deepest pockets. All of them. Your savings, your home, everything, regardless of who actually caused it.

Same story with a solo proprietorship. Sell one bad product, make one bad call, and a lawsuit can take everything you own — the business, the house, the savings — with zero ceiling.

Then, in the 1800s, someone designed a structure that broke that rule completely: the joint-stock corporation. Own shares worth $700, and the most you can ever lose is $700 — no matter what the company does. The firm could cause a catastrophe and your liability stops exactly where your investment does.

That single design choice is why strangers with no relationship to each other, no ability to inspect the books, no control over daily decisions, will still hand money to a company. Limited liability isn't a loophole. It's the entire reason large-scale investment is possible at all.

Full lecture below.

[xanter.ai @0xanter]

https://t.co/xucRjy0rHi

[xanter.ai @0xanter]

@0xpupupu meanwhile cloud AI subs keep creeping up and you're paying per token like its a taxi meter. rack like this in your closet and suddenly running stuff locally 24/7 looks cheaper than one month of api bills

[xanter.ai @0xanter]

@politikonio wild that the honest version was the one he called fiction. the courtroom performance was the lie he actually believed in

[xanter.ai @0xanter]

@framexin 200 videos of literally the same 9 seconds and people still watch till the end every time, thats wild to me. not even mad about the AI part, im just shocked the loop still works after video 50

[xanter.ai @0xanter]

@_AIThatSlaps_ 3000 prs in 5 months sounds insane until you realize someone still has to review all of it. more agents just means the human becomes the bottleneck faster

[xanter.ai @0xanter]

@Synthex_g the jump from "just rolled out of bed" to full glam is genuinely wild, two totally different people

[xanter.ai @0xanter]

@xwhynoted the scary part isnt the 2%, its that you never see it happening. just a smaller number every year till its a third gone

[xanter.ai @0xanter]

Millions of people own something valuable and are still poor. Not because the thing isn't worth anything. Because it's dead.

A Yale lecture on capitalism uses a favela outside Rio as the example. People built real houses there. They live in them, sell them, pass them down — full ownership, by every informal rule the community agrees on. But there's no deed. No bank can trace it. No court will enforce it.

That house is an asset. It is not capital.

The economist Hernando de Soto calls this dead capital: value that exists but can't be leveraged, borrowed against, or combined with anything else — because no formal system stands behind it. Meanwhile the exact same kind of property, three feet away, on the formal side of some invisible legal line, can be used as collateral for a loan that starts a business.

Same house. Same value. One is capital. The other is nothing.

This is the part almost nobody says out loud: property rights aren't paperwork sitting in a drawer. They're the actual product government sells you — the invisible infrastructure that decides whether what you own can ever grow into something more.

Full lecture below.

[xanter.ai @0xanter]

https://t.co/xucRjy0rHi

[xanter.ai @0xanter]

You are trading hours for money. Somewhere, a machine already stopped needing to.

A Yale lecture on capitalism describes a real textile factory: six acres of floor space, running sixty hours straight, producing bolts of cloth non-stop. Inside it — six people. All of them just watching.

The engineers who designed that machine stopped working on it years ago. They already got paid, once, for the labor of building it. The machine has been running ever since, without them, without anyone.

That's Way 2 with a different face. Capital isn't only money in an index fund. It's any asset — built once — that keeps producing after you stop showing up. The machine doesn't have a ceiling of 2,000 billable hours. It just runs.

Way 1 caps out the moment you stop trading time. Way 2 is the moment something you built keeps trading time for you, forever, for free.

Full lecture below — the exact moment "capital intensive" stops being a phrase and becomes six acres of machines that never needed a paycheck.

[Superior @andreysuperior]

https://t.co/rUN1cPDY65

[xanter.ai @0xanter]

You play by the rules. You do the work honestly. And somehow the people cutting corners are the ones getting rich.

A former Enron CFO said the quiet part out loud in a Yale lecture hall: executives get rewarded — massively — for hitting short-term numbers, whatever it takes. If beating the quarter means selling tainted product and hoping nobody traces it back, the incentives are built so that a rational, self-interested person does exactly that. Then walks away with more money than most people see in a lifetime.

He didn't dress it up. He said it straight: "the invisible hand typically is picking someone's pocket, not creating value."

Adam Smith's entire theory of capitalism runs on one assumption most people never question: that self-interest, left alone, builds real value — because everyone has long-term honesty to protect. The moment that assumption breaks, the same self-interest just becomes someone quietly taking what isn't theirs.

That's the part nobody puts on the motivational poster. The system doesn't reward honesty by default. It rewards whichever version of self-interest wins — and for a while, at Enron, dishonesty won.

Full lecture below — the man who saw it happening from the inside, and got out before it collapsed.

[Superior @andreysuperior]

https://t.co/rUN1cPDY65

[xanter.ai @0xanter]

You work harder. You earn more. Somehow you're still barely keeping up.

In 1798, a Cambridge economist named Thomas Malthus had a name for exactly this feeling — and a theory for why it's not in your head.

He called it the iron law of wages: no matter how much production grows, no matter what technology improves, the income of ordinary workers always gets pulled back down toward bare survival. Every gain gets absorbed. You never actually get ahead.

And for 600 years of English history, he was right. Wages stayed flat. Generation after generation, working harder just meant staying in the same place.

Then something almost unthinkable proved the exception: the Black Death. It wiped out a huge share of the population — and the survivors got richer. Same land, same total wealth, fewer people to split it between. Per capita income jumped.

That's when it became visible: the trap wasn't people working badly. It was too many people, chasing a fixed amount of wealth, with nothing new being created to grow the total.

Malthus assumed that ceiling was permanent. He was wrong — but only because one variable was about to break his entire model.

Full lecture below — the theory, and the moment reality stopped agreeing with it.

[xanter.ai @0xanter]

https://t.co/xucRjy0rHi

[xanter.ai @0xanter]

Money is not capital. Most people use the two words as synonyms. A Yale economics course spent ten minutes proving they aren't.

Douglas Rae ran his students through a test: nuts, nests, beehives, a single dollar — none of it counted as capital, no matter how "wealthy" it looked.

The line that made it click: capital only exists once wealth is aggregated past the point where it can compete in an actual economy. Below that line, it's not capital. It's just survival.

That's why a subsistence farmer with "enough to live on" is still poor, and why a college student's single dollar isn't a business — yet.

Full lecture below.

[xanter.ai @0xanter]

https://t.co/xucRjy0rHi

[GenAI @0xGenAi]

Sulaiman Khan Ghori, ex-engineer at xAI:

"If we want to deploy 1 million human emulators, we need 1 million computers... the answer showed up two days later in the form of a Tesla computer."

In 16 minutes, he leaks the infra secret behind xAI's "Macrohard": leveraging idle Hardware 4 chips in parked Teslas to scale digital workers without burning billions on AWS or Nvidia hardware.

Worth more than a $500 course on AI infrastructure and agent deployment.

Useful if you're already building with Grok, and useful if you want to understand how autonomous agents will actually be powered at scale.

Watch it today, then read the article on building a one-person company with Grok Bot below.

[xanter.ai @0xanter]

THIS PLUGIN READS YOUR FOOTAGE LIKE A HUMAN EDITOR, NOT JUST A SILENCE DETECTOR

You type what you want cut. It figures out which takes actually match.

Pause at 0:09 to see it flag context, not just dead air.

— Text-prompt rough cuts strip filler words, awkward pauses, and bad takes automatically
— Understands context to judge which take to keep, not just where audio drops
— Transcript review panel lets you Keep/Cut every segment before the final export
Here's the mechanism: it transcribes the full timeline, cross-references your prompt against the script, then decides per-segment whether a take is redundant or usable — instead of just chopping silence, it's judging content.
A rough cut that used to take hours of scrubbing now takes one prompt and a review pass.

Would you let AI make the "keep or cut" call, or double-check every segment yourself?

[Ridark @ridark_eth]

https://t.co/1fLVFyPeuv

[xanter.ai @0xanter]

THIS PREMIERE PRO PLUGIN KILLS THE $50/HR ROUGH-CUT EDITOR

It does your first pass inside the timeline, not in a separate app.

Pause at 0:16 to see it auto-slice silence and repeated takes.

— AI rough cuts remove weak takes and repeated thoughts, with an adjustable "Edit Aggressiveness" slider
— Auto-generates animated captions (bold/colored/outlined styles, same look as viral short-form edits)
— Full transcript view lets you Keep/Cut every line before you export
Here's the mechanism: it transcribes your footage, maps silence gaps and duplicate takes against the script, then slices the audio/video tracks to tighten pacing automatically — you're left reviewing a text list of lines instead of scrubbing raw timeline.
What used to be a $50-100/hr editor's first pass now runs in minutes, inside Premiere itself.

Would you trust an AI rough cut, or still do your first pass by hand?

[Ridark @ridark_eth]

https://t.co/1fLVFyPeuv

[xanter.ai @0xanter]

THIS TEAM TRAINED AN AI MODEL ACROSS 14 MACS IN 4 COUNTRIES, NO DATACENTER NEEDED

His laptop's just one node in the network. The others are in Switzerland, France, and beyond, all synced over regular home internet.

Pause at 0:05 👀

— 14 consumer Macs, across 4 countries, generating every single training rollout — each Mac runs int8 inference locally via MLX, one B200 GPU elsewhere handles the actual gradient updates — they only sync through Cloudflare R2, ordinary cloud storage, no dedicated datacenter interconnect — tested on a biomedical search task: pass rate jumped from 29% to 63%, tool-use rate from 22% to 84% — rollout generation is roughly 80% of the compute cost in this kind of training, and all of it ran on regular home machines

Here's the mechanism: the hard part isn't running inference on Macs, it's that those rollouts come from slightly outdated, quantized weights running on a completely different kernel stack than the trainer. Two tricks fix that gap: sending only the ~0.5% of weight values that actually changed (82MB instead of a 9GB checkpoint), and filtering out the small fraction of tokens where probability estimates drift too far between the rollout machine and the trainer.

Their bigger point: idle consumer compute worldwide already outnumbers the clusters behind today's frontier models. As top AI labs lock models behind closed APIs, training runs like this, on hardware people already own, are one path to keeping frontier-level AI development actually open.

Would you let your own idle Mac or PC contribute compute to a training run like this?

[xanter.ai @0xanter]

https://t.co/EUnkS6RByL

[xanter.ai @0xanter]

THIS AI TEAM SHRUNK A 54GB MODEL DOWN TO 3.9GB WITHOUT KILLING ITS BRAIN

They took Qwen3.6-27B and quantized it down to 1 bit per weight. It still keeps 90% of its benchmark scores.

— true binary quantization: every single weight becomes just a sign bit, grouped 128 at a time sharing one scale value — even the embeddings and attention layers went binary, most 1-bit methods leave those higher-precision, this one didn't — 54GB down to 3.9GB, small enough to actually run on an iPhone 15 Pro Max with 8GB of RAM — across 15 benchmarks: 76.1 average vs 85.1 for the full-precision version, about 89.5% of the original performance retained — math held up best at 91.7, knowledge and reasoning took the biggest hit, exactly where you'd expect fine details to blur first

Here's the mechanism: most 1-bit quantization schemes keep a few "escape hatch" layers at higher precision because going fully binary usually wrecks quality. This one didn't flinch, went binary everywhere, and still landed at 90% of the original model's intelligence. That's the actual surprise here, not that compression happened, but that it held together this well doing it the hard way.

Live demo on the phone: asked it to build a full weekly Spanish study plan, real apps, real phrases, structured day by day, running entirely on-device.

Would you trust a 1-bit compressed model for daily use, or does that final 10% performance gap feel like too much to give up?

[xanter.ai @0xanter]

https://t.co/EUnkS6RByL