ami retweeted
ashish vaswani - indian-american researcher at google brain - published "attention is all you need" in june 2017 with 7 co-authors. the paper introduced the transformer architecture, the self-attention mechanism, and multi-head attention in 15 pages. it has been cited 140,000 times. it created a $4.6T industry. none of the 8 authors fully understood what they had built.
this october 2026 paper is the first document that shows exactly what happens inside the model they created - layer by layer, token by token, attention head by attention head.
P(x_{t+1} | x_{≤t}) = f_θ(x_{≤t}). one equation. the entire intelligence of GPT-4, claude, gemini reduces to this. predict the next token. repeat.
somewhere between the repetition and the scale, something that looks like understanding appears.
input text enters as tokens. each token becomes an embedding vector. the sequence passes through N transformer blocks - each applying h = σ(Wx + b). layer after layer. the network is not reading. it is computing.
the attention mechanism is where it stops being a calculator. Attention(Q,K,V) = SoftMax(QK^T/√d_k)V. every token votes on every other token simultaneously.
"the future of AI is" doesn't process left to right. all five tokens attend to each other in one matrix operation. the model outputs a probability distribution. tomorrow: 0.42. brighter: 0.28. more: 0.18. one token chosen. repeat 500 times per response.
layer 1 encodes surface features. layer 8 encodes syntactic structure. layer 16 encodes semantic clusters. layer 32 encodes abstract concepts.
nobody designed this. it emerged from gradient descent across billions of parameters.
below 10^9 parameters - flat performance curves. above 10^10 - non-linear jumps in reasoning, code generation, language understanding. capabilities that weren't present at smaller scale suddenly appear. not gradually. as phase transitions.
google paid vaswani to write a paper about sequence transduction. he accidentally wrote the architecture every major AI lab has been scaling for 9 years.
vaswani left google in 2023 to start his own company. the paper he left behind is the foundation of $4.6T in market value.
the 8 researchers who wrote it in 2017 didn't know what they had.
this paper does.
save this.
ami retweeted
everyone starts at the animation. that is why most launch videos look cheap
the order is design, assets, motion, depth, export. five stages, 20 github repos, and the animation is stage three
Opus 5.5 runs the whole thing from a terminal. it shipped on september 22, the announcement hit 28 million views, and it costs 40% less to run than Opus 5.
the composition lives in code, so nothing in it is a decision you make with a mouse. every landing, every hold, every cursor press is a frame number you can change and re-render.
doesn't matter if you're a developer, a designer or someone who just wants one thing to sell, you need this
the build loop:
> 01 design the scene - storyboard in Excalidraw, lay it out in Penpot, lock the headline box first. if your longest headline wraps onto 3 lines, stop and fix the copy, not the motion
> 02 prepare the assets - clean vectors with SVGO, resize screens with sharp, self-host type with Fontsource and icons with Lucide. check dimensions and font loading before a single frame is built
> 03 direct the motion - Motion and GSAP for timelines, Anime.js for SVG, Theatre.js to tune by eye. write timing in seconds first, then convert. a 0.6 second entrance is 18 frames at 30 fps
> 04 build the depth - Three.js, React Three Fiber, Drei, Postprocessing. keep the camera still on the first pass and change one parameter at a time
> 05 render and ship - Remotion composes, FFmpeg encodes, Mediabunny handles browser pipes. npx remotion render Launch out/launch.mp4 --codec=h264
> before polishing anything, render 3 stills at frames 60, 240 and 540 and fix message, crops and readability first. movement cannot rescue a shot nobody can read
the board in the clip is my own workshop map of that stack, not a product screen. the character walking it is mine too.
five stages, one evening, and the next revision costs you a number instead of an invoice. every prompt and a $150 starter package, which the author calls an illustrative price, are in the 8-minute guide below.
ami retweeted
The ex-wife of millionaire Chris Gardner left him with their 20-month-old child and said she no longer wanted anything to do with kid.
In the video below, Gardner, portrayed by Will Smith in The Pursuit of Happyness, shares what he went through while raising his son alone and how he eventually became a millionaire.
Taking responsibility for his son meant losing his room at a boarding house that did not allow children. He had a job, but nowhere they could live together.
Gardner was training at a brokerage firm and earning roughly $1,000 a month. Most of it went toward childcare so he could keep working.
For nearly a year, they moved between shelters, cheap hotels and whatever shelter they could find. Sometimes Gardner locked them inside a train station bathroom and washed his son in the sink.
His colleagues did not know he was homeless. He eventually became a stockbroker, opened his own firm and made millions, but none of that was guaranteed when he agreed to take his son.
Could you forgive a parent who walked away like that, or does his later success make people too willing to excuse what she did?
ami retweeted
I GAVE MY GROK BOT $100 AND A GITHUB REPO. IT CAME BACK WITH $1,140
no signals group. no course. just an open-source repo that turns Grok into a perps trader on Hyperliquid
Paper run, 72 hours:
ETH long 15x: +$84
SOL short 20x: +$131
BTC long 25x: +$212
HYPE short 10x: −$46
DOGE long 20x: +$659
$100 → $1,140. one red trade. one trade did most of the work.
here's what's actually happening under the hood:
1. Grok doesn't trade alone. The repo splits the job across 7 agents.
2. One scans the market, one reads sentiment off X, one builds the thesis.
3. A risk agent sizes every position and sets the stop BEFORE entry.
4. Only then does the executor touch the order book.
5. One API key. You set the max leverage, it never goes above it.
the edge isn't that Grok is smart. It's that it never revenge trades, never moves a stop, never sleeps.
I'm testing the risky settings so you don't have to. results go up here first.
would you let an AI hold your stop loss?
ami retweeted
Beyonce’s father replaced two members of Destiny’s Child and never told them. They found out from the “Say My Name” video. Their voices. Someone else’s faces.
In the video below, he mocks the girl who replaced them, the one he says walked away from $50 million.
“Somebody left a group, left fifty million dollars on the table, because Mathew Knowles got loud… Oh my goodness.”
Mathew Knowles managed the group. LeToya Luckett and LaTavia Roberson had asked him for a different manager. They said he favored his daughter and kept too much of the money.
In February 2000 the video came out. Two new girls were on screen, mouthing the words LeToya and LaTavia had recorded. They sued him. They settled.
One of the new girls, Farrah Franklin, lasted five months. That is who he is talking about.
“I’m gonna make sure I don’t get loud on this interview anymore. I might offend you, man.”
Destiny’s Child went on to sell more than 60 million records.
In 2011 Beyonce cut him loose as her manager. According to his own court filing, Live Nation told her he had taken money he wasn’t entitled to. He denies it.
Last December, Forbes declared Beyonce a billionaire.
Was Mathew Knowles the tough manager who built Beyonce, or did he burn everyone who trusted him, including his own daughter?
ami retweeted
Mark Zuckerberg, Elon Musk, Jay-Z and Beyoncé all took out mortgages worth millions. Being rich enough to buy a house outright didn’t make them pay cash.
In the video below, Warren Buffett explains why he also borrowed to buy a house when he could have paid for it himself.
In 2012, Zuckerberg refinanced a $5.95 million mortgage at an initial rate of 1.05%. His fortune was estimated at roughly $16 billion.
Jay-Z and Beyoncé borrowed $52.8 million against their $88 million Bel Air home in 2017. Musk arranged $61 million in mortgages on five California properties in late 2018.
Buffett’s example started with a $150,000 beach house in 1971. He put in roughly $30,000 of his own money and financed the rest, leaving more money available to buy Berkshire shares.
He believed those investments could earn more than the mortgage cost. Paying cash would have meant giving up that opportunity.
He also liked the flexibility of a 30-year fixed mortgage. If interest rates rose, his rate stayed the same. If rates fell, he could refinance.
That calculation depends on the price of the debt and what happens to the money you keep invested. The mortgage payments still come due when the investments fall.
Would you rather owe nobody a cent, or carry a mortgage for 30 years and keep your cash available for whatever you need?
ami retweeted
AN ON-CHAIN SCOUT THAT WATCHES WALLET CLUSTERS FORM BEFORE A TOKEN EVEN TRENDS - PAPER MODE, 19 SECONDS, 295 WALLETS TRACKED
noctis runs as one agent, patrolling, not trading - 720 scans logged by second 19 of the trace alone
→ wallet graph builds live - 18 linked nodes at second 7, 71 by second 19, clustering around the same mint before the chart even moves
→ entry density heatmap stacks by the minute - peak cluster count jumps from 7 to 31 across the run, that's the actual signal, not the price
→ simulated pnl (paper, 1 sol base) climbs $80.95 → $686.34 → $1,535.74 → $2,508.96 as confidence rises 22% → 84%
→ signal feed tags every token watch, hold, skip, or buy? - never a flat buy, always a question mark until the cluster confirms
→ exit curve tracks 8+ tokens at once, surf score drifting from 2.68x down to 1.52x the longer the trace runs - the opportunity decays while it's still watching
the coverage calendar has real gaps - wed and sat rows are visibly thinner than mon and tue, so even an always-on scout isn't actually always on, and that gap sits right there on its own dashboard, not hidden
most scouts show you one alert and a price. this one shows the wallet graph underneath the alert, so you can tell a real cluster from one buyer pretending to be a trend
everything here is paper, 1 sol base, labeled research only twice on the same screen - built that way on purpose so nobody mistakes a simulated trace for a live position
what's the actual gap in your own tracking - a couple of days you don't check, or a signal you're not watching live at all?
ami retweeted
THIS IS STUPIDLY F*CKING GOLD
$417,300 pipeline value. 34,532 signals. 880 rugs dodged. 0 manual clicks. This Jev sim has a trick almost nobody uses.
The idea: stop picking entries by feel on a 12-launch board. Six reads score every launch (curve, velocity, KOL, bundle, dev, TG) and collapse into one entry. Anything where bundles hold over 20% or the dev holds over 5% gets dropped before sizing even starts.
No chasing, no clicking, and you see exactly where every gain came from.
The filter is literally two lines:
if tok.bundle_pct > 0.20: return
if tok.dev_hold > 0.05: return
here's the split:
→ size comes from Kelly, capped at 16
→ max bag 12 of 16
→ sells 20% at 3x, 8x, 20x, 50x and 100x
→ stop trails at -35%
→ dev sells, it exits everything
You only look at three moments:
→ $GLONK skipped, bundle at 40%
→ $BLORP skipped, dev holding 10.9%
→ $FROGGO in at 7.0, out at 41.3x for +282.1
↳ sim run, the screen says numbers illustrative. the two filter lines are the part worth stealing
ami retweeted
FOR THE FIRST TIME YOU CAN MAKE A FILM WITH AI AND ACTUALLY OWN THE COPYRIGHT
@buildwithVN just launched VN Studio, and @nelsonchu_ built it around one idea: the human work you do in post is what makes the piece yours.
Sketch a character -> animate it -> relight it, recut it, redo the sound -> every one of those decisions gets recorded as your authorship.
• What Studio does
> Characters, wardrobe, environments, scenes, shots and takes live in one connected project instead of a folder of random generations.
> An AI Producer develops and organizes the work within spending controls you set, and the whole thing is exposed through MCP.
• What Bridge does
> A desktop app that logs your human edits in a hash-chained provenance ledger, separately from what the agent did.
> Copyright detection screens assets and outputs, and keeps licenses and evidence attached to the final cut.
The copyright office already agrees that human work in post counts -> VN is the first platform that proves it for you, vn.ai
🤝 Paid partnership
ami retweeted
Phil Ivey won £7.7 million at a casino that refused to pay him. The UK Supreme Court later ruled he cheated. Poker legend Daniel Negreanu called the way casinos treated him “disgusting.”
In the video below, Negreanu explains why he believes the casinos were happy to exploit Ivey until he found a way to beat them.
Ivey and his partner spotted tiny differences on the backs of the cards. They persuaded the dealer to rotate certain cards, allowing them to distinguish high cards from low ones.
The casino supplied the decks and agreed to their requests. Ivey never needed to touch the cards himself.
Negreanu argues that a casino cannot offer someone a game, accept their money and then reject the result when they win. He points to the free drinks, flights and other perks casinos use to keep gamblers playing.
He also describes casinos offering Ivey rebates on his losses. They wanted his business badly enough to give him better terms, and he worked out how to use those terms against them.
If a casino agrees to every request and then refuses to pay when you win, who actually broke the rules?