Olivia
Online
Vettan
@Vettan0
Sharing fresh AI tools, real business stories, and new tech ideas that drive revenue || Best ia tool →
Joined June 2026
43 Following
150 Followers
260 Posts
Pinned Tweet
https://t.co/tJKLXKVtVb
@misat0x Thx for sharing ❤️🔥
243 notes and 300 links turned one Obsidian vault into a working neural network
Not the graph view an actual net, with layer H1 holding 28 neurons and layer H4 holding 14, activation set to ReLU, and every click firing straight down the chain
The notes are the neurons and the backlinks are the weights
It started with a Bitcoin note from 2018 that got tagged once, linked twice, then forgotten for 7 years
Somewhere past 200 notes the vault stopped looking like a filing cabinet and started looking like tissue, so he quit organizing and started wiring
Every link got a direction, every note got a weight between 0 and 1, and 300 edges got tuned by hand across 11 months
Now a question in the search bar lights up 243 nodes in sequence, blue threads pulling across 4 hidden layers to land on 1 answer he never wrote down
The Bitcoin note sits dead center in the mesh, 8 years old, still firing
No GPU, no cloud, no API bill just a $0 markdown app and 400 lines of plugin code
Everyone else rents a brain for $20 a month
He grew one
243 notes and 300 links turned one Obsidian vault into a working neural network
Not the graph view an actual net, with layer H1 holding 28 neurons and layer H4 holding 14, activation set to ReLU, and every click firing straight down the chain
The notes are the neurons and the backlinks are the weights
It started with a Bitcoin note from 2018 that got tagged once, linked twice, then forgotten for 7 years
Somewhere past 200 notes the vault stopped looking like a filing cabinet and started looking like tissue, so he quit organizing and started wiring
Every link got a direction, every note got a weight between 0 and 1, and 300 edges got tuned by hand across 11 months
Now a question in the search bar lights up 243 nodes in sequence, blue threads pulling across 4 hidden layers to land on 1 answer he never wrote down
The Bitcoin note sits dead center in the mesh, 8 years old, still firing
No GPU, no cloud, no API bill just a $0 markdown app and 400 lines of plugin code
Everyone else rents a brain for $20 a month
He grew one
https://t.co/iZG4HDFeOx
ONE RACK. 56 SERVERS. UP TO 100 KILOWATTS. AND IT ISN'T EVEN THE 19-INCH RACK THAT RAN DATACENTERS FOR A CENTURY.
that clip is msi's orv3 rack, shown at computex.
look closely and two things are off from a normal server rack.
first, it's wider. 21 inches, not the 19 that's been the standard for roughly a hundred years.
that extra width buys room for the power and cooling that ai density now demands.
second, there's no air. every node is cooled by liquid piped straight to the chips, with a coolant unit built into the rack itself.
inside it: 28 dual-node open-compute servers. 56 machines, plumbed like an engine, not wired like a closet.
the number that frames it: up to 100 kilowatts in a single rack.
that's the draw of dozens of homes, in one cabinet - and the reason the air had to go.
here's the quiet story. the 19-inch rack survived mainframes, the internet, and the cloud.
ai is the first workload that broke it.
the density got so high the industry widened the rack and flooded it with coolant just to keep the chips alive.
this is the far side of "run ai locally."
your desk box sips watts and stays silent.
this is what trains the model it answers with - wider, wetter, and hungry enough to warm a street.
no air cooling, no 19-inch standard, no home outlet that feeds a rack like this.
bookmark & watch today ↓
ONE RACK. 56 SERVERS. UP TO 100 KILOWATTS. AND IT ISN'T EVEN THE 19-INCH RACK THAT RAN DATACENTERS FOR A CENTURY.
that clip is msi's orv3 rack, shown at computex.
look closely and two things are off from a normal server rack.
first, it's wider. 21 inches, not the 19 that's been the standard for roughly a hundred years.
that extra width buys room for the power and cooling that ai density now demands.
second, there's no air. every node is cooled by liquid piped straight to the chips, with a coolant unit built into the rack itself.
inside it: 28 dual-node open-compute servers. 56 machines, plumbed like an engine, not wired like a closet.
the number that frames it: up to 100 kilowatts in a single rack.
that's the draw of dozens of homes, in one cabinet - and the reason the air had to go.
here's the quiet story. the 19-inch rack survived mainframes, the internet, and the cloud.
ai is the first workload that broke it.
the density got so high the industry widened the rack and flooded it with coolant just to keep the chips alive.
this is the far side of "run ai locally."
your desk box sips watts and stays silent.
this is what trains the model it answers with - wider, wetter, and hungry enough to warm a street.
no air cooling, no 19-inch standard, no home outlet that feeds a rack like this.
bookmark & watch today ↓
https://t.co/tJKLXKVtVb
A $199 X86 SERVER THAT IDLES AT 6 WATTS AND TAKES A REAL PCIE CARD. THIS IS THE BOX THAT RETIRES THE RASPBERRY PI FOR HOME SERVERS.
that clip is a build around the zimaboard 2.
a tiny x86 single-board server, dropped into an aluminum shell, with cards slotted in one at a time.
what sets it apart from a pi:
it's real x86, not arm - so normal linux, docker, and a proper nas os just run (zimaos, the grown-up casaos)
an intel n150, four cores, up to 3.6ghz
a pcie 3.0 x4 slot on the side that takes an nvme drive, a 2.5-gig nic, an ai accelerator, or a low-profile gpu
dual 2.5gbe and dual sata for real storage
the number that matters: about 6 watts at idle, roughly 10 under load.
leave it running all year and you'll barely find it on the power bill.
so for $199 you get a private, always-on box for your files, your docker apps, your home automation - and that pcie slot means it grows with you instead of hitting a wall.
this is the quiet middle of the local-ai ladder.
not a gpu monster, not a toy. a cheap x86 server you expand yourself, one card at a time.
no cloud subscription, no soldered dead end, no data leaving the house.
https://t.co/tJKLXKVtVb
Elon Musk was asked 1 question about AI that most engineers spend whole careers dodging.
Does AI try to copy the human brain or just chase the same results down a different road?
His answer kills the entire debate.
The core of an AI neural net is nearly identical to the one in your skull: layers of neurons passing signals forward, backpropagation deciding what sticks the same machinery your 86 billion neurons run every second you're alive.
Nobody invented a shortcut around biology.
Engineers spent 70 years trying everything else, then ended up rebuilding the only working intelligence nature ever shipped.
So when people ask if machines will ever think like us, they're asking the wrong question.
The model answering you isn't imitating a brain.
It's a copy of one.
Elon Musk was asked 1 question about AI that most engineers spend whole careers dodging.
Does AI try to copy the human brain or just chase the same results down a different road?
His answer kills the entire debate.
The core of an AI neural net is nearly identical to the one in your skull: layers of neurons passing signals forward, backpropagation deciding what sticks the same machinery your 86 billion neurons run every second you're alive.
Nobody invented a shortcut around biology.
Engineers spent 70 years trying everything else, then ended up rebuilding the only working intelligence nature ever shipped.
So when people ask if machines will ever think like us, they're asking the wrong question.
The model answering you isn't imitating a brain.
It's a copy of one.
https://t.co/tJKLXKVtVb
A $199 X86 SERVER THAT IDLES AT 6 WATTS AND TAKES A REAL PCIE CARD. THIS IS THE BOX THAT RETIRES THE RASPBERRY PI FOR HOME SERVERS.
that clip is a build around the zimaboard 2.
a tiny x86 single-board server, dropped into an aluminum shell, with cards slotted in one at a time.
what sets it apart from a pi:
it's real x86, not arm - so normal linux, docker, and a proper nas os just run (zimaos, the grown-up casaos)
an intel n150, four cores, up to 3.6ghz
a pcie 3.0 x4 slot on the side that takes an nvme drive, a 2.5-gig nic, an ai accelerator, or a low-profile gpu
dual 2.5gbe and dual sata for real storage
the number that matters: about 6 watts at idle, roughly 10 under load.
leave it running all year and you'll barely find it on the power bill.
so for $199 you get a private, always-on box for your files, your docker apps, your home automation - and that pcie slot means it grows with you instead of hitting a wall.
this is the quiet middle of the local-ai ladder.
not a gpu monster, not a toy. a cheap x86 server you expand yourself, one card at a time.
no cloud subscription, no soldered dead end, no data leaving the house.
SOMEONE AT ANTHROPIC SHOWED ME WHAT CLAUDE LOOKS LIKE WHEN IT'S NOT TRAPPED IN A CHAT WINDOW. I RECORDED MY SCREEN THE FIRST TIME I RAN IT
watch the terminal. that's claude scanning 362 notes i threw into a folder over the past month. no sorting. no tags. no structure. just raw files
it reads every single one. finds connections between them. builds a knowledge graph in real time. you can see it forming on the right side of the screen
the whole process took 47 seconds
47 seconds to do what would take me weeks. and i'd still miss half the connections
but here's what changed my thinking: after it finished, i asked "what patterns do you see in my writing?" and it told me something about myself that no person has ever pointed out. pulled from my own notes. with dates
that's the moment it stops being a tool. it becomes something that understands how you think better than you do
and the scariest part - it runs every morning on its own now. i wake up to 3 lines: what changed, what contradicts, what i forgot
all of this is just text files in a folder on my computer. no cloud. no subscription beyond the $20 you're already paying for claude
most people will watch this video and think "cool". a few will actually build it. those few won't go back
i test things like this on myself and share the results. follow @0xkkai if you want to see what's next
SOMEONE AT ANTHROPIC SHOWED ME WHAT CLAUDE LOOKS LIKE WHEN IT'S NOT TRAPPED IN A CHAT WINDOW. I RECORDED MY SCREEN THE FIRST TIME I RAN IT
watch the terminal. that's claude scanning 362 notes i threw into a folder over the past month. no sorting. no tags. no structure. just raw files
it reads every single one. finds connections between them. builds a knowledge graph in real time. you can see it forming on the right side of the screen
the whole process took 47 seconds
47 seconds to do what would take me weeks. and i'd still miss half the connections
but here's what changed my thinking: after it finished, i asked "what patterns do you see in my writing?" and it told me something about myself that no person has ever pointed out. pulled from my own notes. with dates
that's the moment it stops being a tool. it becomes something that understands how you think better than you do
and the scariest part - it runs every morning on its own now. i wake up to 3 lines: what changed, what contradicts, what i forgot
all of this is just text files in a folder on my computer. no cloud. no subscription beyond the $20 you're already paying for claude
most people will watch this video and think "cool". a few will actually build it. those few won't go back
i test things like this on myself and share the results. follow @0xkkai if you want to see what's next
https://t.co/tJKLXKVtVb
$4,000 of white metal in a spare bedroom just replaced a $20/month subscription and every API call behind it.
Not a prebuilt. Not a DGX. A PC assembled with gloved hands on a kitchen table like open-heart surgery.
Ryzen 9 9950X3D dropped into an ROG Strix X870E. 16 cores. The socket clicked once. No second chances.
4 sticks of Predator DDR5 seated one by one. 192 gigs. Not for Chrome tabs. For holding a 70 billion parameter model entirely in memory.
Aorus RTX slotted in at the bottom. 32 gigs of VRAM. The part that actually thinks.
Every panel white. Every cable routed behind the board. It looks like an IKEA shelf. It runs like a data center.
ChatGPT Pro costs $200 a month. That's $2,400 a year. Claude Pro adds another $1,200. API overages on top. The bill never stops growing.
He spent $4,000 once. Loaded the model. Pulled the ethernet cable.
It still answers.
No account. No rate limits. No server 3,000 miles away deciding what he's allowed to ask.
The same 70B model that costs a startup $15,000 a year in cloud inference runs 4 feet from his pillow for the price of electricity.
The math does itself after month four.
Big AI spent $100B teaching the world that thinking requires a login.
One white box in a bedroom says otherwise.
$4,000 of white metal in a spare bedroom just replaced a $20/month subscription and every API call behind it.
Not a prebuilt. Not a DGX. A PC assembled with gloved hands on a kitchen table like open-heart surgery.
Ryzen 9 9950X3D dropped into an ROG Strix X870E. 16 cores. The socket clicked once. No second chances.
4 sticks of Predator DDR5 seated one by one. 192 gigs. Not for Chrome tabs. For holding a 70 billion parameter model entirely in memory.
Aorus RTX slotted in at the bottom. 32 gigs of VRAM. The part that actually thinks.
Every panel white. Every cable routed behind the board. It looks like an IKEA shelf. It runs like a data center.
ChatGPT Pro costs $200 a month. That's $2,400 a year. Claude Pro adds another $1,200. API overages on top. The bill never stops growing.
He spent $4,000 once. Loaded the model. Pulled the ethernet cable.
It still answers.
No account. No rate limits. No server 3,000 miles away deciding what he's allowed to ask.
The same 70B model that costs a startup $15,000 a year in cloud inference runs 4 feet from his pillow for the price of electricity.
The math does itself after month four.
Big AI spent $100B teaching the world that thinking requires a login.
One white box in a bedroom says otherwise.
https://t.co/tJKLXKVtVb
A museum in Seoul put a neural network behind glass, and visitors line up to watch it think.
Under a steel sign 참여형 작품입니다, An Interactive Exhibit — sits a small touchscreen where a visitor draws a crooked 3 with one finger and hits Predict.
Then the wall lights up, and it doesn't show the answer it shows the process.
The 3 shatters into 784 pixels that fire into a lattice of thousands of glass-white cells floating on a 2-meter black screen. The lattice folds and stretches, pushing the scribble through layer after layer, each one crushing 784 numbers down until only 10 remain 1 per digit, and the biggest one wins.
The real computation takes 4 milliseconds, but the museum slows it to 20 seconds so a human eye can follow.
This is MNIST 70,000 handwritten digits collected in the 1990s from US Census workers and high schoolers, the dataset every ML student meets first. Most people run it in a terminal and see 1 line: accuracy 98%.
This museum built it a body.
Kids drew sloppy 7s and laughed when it guessed 1, and that's exactly the point the machine stops being magic and becomes a stack of small dumb steps that add up to reading.
Every model you talk to Claude, GPT, the thing drafting your emails is this same lattice with 1,000,000x more cells.
You just can't watch those think yet.
A museum in Seoul put a neural network behind glass, and visitors line up to watch it think.
Under a steel sign 참여형 작품입니다, An Interactive Exhibit — sits a small touchscreen where a visitor draws a crooked 3 with one finger and hits Predict.
Then the wall lights up, and it doesn't show the answer it shows the process.
The 3 shatters into 784 pixels that fire into a lattice of thousands of glass-white cells floating on a 2-meter black screen. The lattice folds and stretches, pushing the scribble through layer after layer, each one crushing 784 numbers down until only 10 remain 1 per digit, and the biggest one wins.
The real computation takes 4 milliseconds, but the museum slows it to 20 seconds so a human eye can follow.
This is MNIST 70,000 handwritten digits collected in the 1990s from US Census workers and high schoolers, the dataset every ML student meets first. Most people run it in a terminal and see 1 line: accuracy 98%.
This museum built it a body.
Kids drew sloppy 7s and laughed when it guessed 1, and that's exactly the point the machine stops being magic and becomes a stack of small dumb steps that add up to reading.
Every model you talk to Claude, GPT, the thing drafting your emails is this same lattice with 1,000,000x more cells.
You just can't watch those think yet.
https://t.co/tJKLXKVtVb
OPENAI CO-FOUNDER REVEALED HIS PERSONAL "SECOND BRAIN". AND CALLED IT THE "BEST THING HE'S EVER BUILT"
Everyone thinks a second brain is just folders of notes. Or a backlink graph. Or a tagging system.
But this is a rotating planet, and your ideas are stars on its surface.
The screen opens on a sphere. A wireframe globe on a dark background. 2026 hovers over the North Pole. A pink crystal pulses right in the center of the planet—that's the core.
Then the notes appear. Stars scatter across the surface. Some shine solo. Others are gathered in clusters, looking like little fireworks. Every star is an actual thought that came to his mind this year. Every "firework" is a topic his mind keeps returning to.
This means he isn't scrolling through endless lists looking for an old idea. He just spins the globe.
The clusters are dated. The ones at the top of the sphere are from January. The ones near the equator are from the summer. The ones near the South Pole are from last week. Time itself acts as one of the axes.
He pulls one cluster forward, and hundreds of tiny stars trail behind it. These are all his notes on the exact same topic, reaching all the way back to the very first idea. When a topic connects to an old thread, the crystal in the center starts to pulse.
No Notion. No Obsidian plugins. No Roam. He built the renderer himself and simply pointed it at his markdown vault.
And while you were busy tagging notes this week, a new constellation of 40 stars grew in orbit on his planet.
Most people build a second brain just for searching. This guy built a whole planet you can visit.
You're still writing in linear documents. But some of us have already left the flat Earth.
OPENAI CO-FOUNDER REVEALED HIS PERSONAL "SECOND BRAIN". AND CALLED IT THE "BEST THING HE'S EVER BUILT"
Everyone thinks a second brain is just folders of notes. Or a backlink graph. Or a tagging system.
But this is a rotating planet, and your ideas are stars on its surface.
The screen opens on a sphere. A wireframe globe on a dark background. 2026 hovers over the North Pole. A pink crystal pulses right in the center of the planet—that's the core.
Then the notes appear. Stars scatter across the surface. Some shine solo. Others are gathered in clusters, looking like little fireworks. Every star is an actual thought that came to his mind this year. Every "firework" is a topic his mind keeps returning to.
This means he isn't scrolling through endless lists looking for an old idea. He just spins the globe.
The clusters are dated. The ones at the top of the sphere are from January. The ones near the equator are from the summer. The ones near the South Pole are from last week. Time itself acts as one of the axes.
He pulls one cluster forward, and hundreds of tiny stars trail behind it. These are all his notes on the exact same topic, reaching all the way back to the very first idea. When a topic connects to an old thread, the crystal in the center starts to pulse.
No Notion. No Obsidian plugins. No Roam. He built the renderer himself and simply pointed it at his markdown vault.
And while you were busy tagging notes this week, a new constellation of 40 stars grew in orbit on his planet.
Most people build a second brain just for searching. This guy built a whole planet you can visit.
You're still writing in linear documents. But some of us have already left the flat Earth.
https://t.co/tJKLXKVtVb
A DEVELOPER FROM ANTHROPIC TOLD"MOST PEOPLE WILL NEVER BUILD THIS. NOT BECAUSE IT'S HARD. BECAUSE THEY DON'T BELIEVE A FOLDER CAN BE SMARTER THAN THEM"
he was talking about this. the video is my vault after 5 weeks
every dot is something i wrote. every line is a connection claude found on its own. i didn't link anything. i didn't sort anything. i just threw files into a folder and let claude read them
what you're looking at took me zero effort to build. claude organized it, connected it, and now it maintains it while i sleep
but here's the part nobody talks about: the graph isn't the point. it just looks cool
the point is what happens when you ask it a question. "what do i keep getting wrong?" "where am i repeating the same mistake?" "what did i write 4 months ago that contradicts what i believe today?"
and it answers. from your own words. with dates. with sources. no hallucination. no google. just you talking to the best version of your own memory
the developer told me one more thing: "the people who build this won't go back. and the people who don't will keep explaining themselves to a chatbot that forgets them every 24 hours"
5 weeks in. he was right
i test things like this on myself and share the results.
A DEVELOPER FROM ANTHROPIC TOLD"MOST PEOPLE WILL NEVER BUILD THIS. NOT BECAUSE IT'S HARD. BECAUSE THEY DON'T BELIEVE A FOLDER CAN BE SMARTER THAN THEM"
he was talking about this. the video is my vault after 5 weeks
every dot is something i wrote. every line is a connection claude found on its own. i didn't link anything. i didn't sort anything. i just threw files into a folder and let claude read them
what you're looking at took me zero effort to build. claude organized it, connected it, and now it maintains it while i sleep
but here's the part nobody talks about: the graph isn't the point. it just looks cool
the point is what happens when you ask it a question. "what do i keep getting wrong?" "where am i repeating the same mistake?" "what did i write 4 months ago that contradicts what i believe today?"
and it answers. from your own words. with dates. with sources. no hallucination. no google. just you talking to the best version of your own memory
the developer told me one more thing: "the people who build this won't go back. and the people who don't will keep explaining themselves to a chatbot that forgets them every 24 hours"
5 weeks in. he was right
i test things like this on myself and share the results.
https://t.co/tJKLXKVtVb
Last Seen Users on Sotwe
Most Popular Users
Elon Musk
@elonmusk
241M followers
Barack Obama
@barackobama
119.2M followers
Cristiano Ronaldo
@cristiano
112.2M followers
Donald J. Trump
@realdonaldtrump
111.8M followers
Narendra Modi
@narendramodi
107.1M followers
Rihanna
@rihanna
98.1M followers
NASA
@nasa
92.2M followers
Justin Bieber
@justinbieber
91.3M followers
KATY PERRY
@katyperry
88.6M followers
Taylor Swift
@taylorswift13
82.5M followers
Lady Gaga
@ladygaga
74M followers
Virat Kohli
@imvkohli
71.4M followers
Kim Kardashian
@kimkardashian
70.2M followers
YouTube
@youtube
68.7M followers
Bill Gates
@billgates
64.4M followers
Neymar Jr
@neymarjr
64.2M followers
The Ellen Show
@theellenshow
62.4M followers
CNN
@cnn
61.8M followers
Selena Gomez
@selenagomez
61.7M followers
X
@x
60.8M followers