If you weren't on the internet for the

last week, you may have missed it, but

Midjourney has decided to enter into the

medical arena. That's right. They're

throwing their hat into medicine. And

what they're effectively building is a

tube in which you go into and it

vibrates water and thus measuring the

vibrations is able to take a decently

accurate image of all the soft stuff

inside of your body. By the way, this

video very beautiful. Like I just even

me trying to even me trying to record

this YouTube video, I'm like watching

them like wow. They look so futuristic.

I've seen this this like on aliens. They

also released a blog going into the

details of how they plan on rolling this

out. What is their strategy? And a super

cool technical video actually breaking

down how this works, the data they had

to transfer and everything. Now, you're

probably thinking this is going to be me

making fun of an AI company. No, that's

not the plan of this video. No, I'm not

going to tell you about the efficacy of

the medical arena and whether

ultra-sonic scans are better than MRIs

or compete with MRIs or anything like

that, okay? Hey, if you're coming to me

for medical advice, you you messed up in

your life. I'm not even going to talk

about the fact that you probably are

going to have to get your MRI results

publicly on a Discord server. IN FACT,

I'M NOT EVEN GOING TO TALK ABOUT THE

FACT that they named it full body

ultra-sonic computational tomography,

okay? Or for short.

for short.

What the is going on here?

All right. for short. What I do

want to talk about are the numbers, the

actual data numbers, what they're

proposing because the technology that

they are proposing they're going to be

building over the next couple years is

literally space age technology alien

stuff. So, let's talk about it and why

honestly, it's the most it's the

craziest part of this entire pivot.

Midjourney going into medical, yeah,

that's kind of un- unexpected.

Midjourney saying what technology

they're going to build, nope. I don't

believe it for a second. But before we

begin, thank you to our sponsors.

Hey, do you like waiting? Yeah. Neither

do I. With Blacksmith, you get speed, 2x

faster hardware, 4x faster cash

downloads, and 40 times faster Docker

builds. So, stop waiting around. Use

their migration wizard to easily

transition your repos onto Blacksmith

today. Thank you, Blacksmith, for

sponsoring this video, and you, check

out the links down below. Hey, welcome

back. If you want to support the

channel, click the link in the

description. All right, so the blog

gives more details. So, it starts off by

saying, "Hey, Midjourney is going to

become a spa in which they're going to

be opening up the first one in San

Francisco in 2027, where they'll have

hot tubs, saunas, cold plunges, and of

course, a cozy rooms with pools of

golden light in which you can softly

scan your body."

Some weird. Also, that golden room, I

recognize that golden room, okay? That

is construct off of Halo. I've seen I've

actually I was birthed my I was birthed

in this room, okay? I know it quite

well. All right, so let's get into some

of the more technical parts about this.

Uh we first need to start off with their

just general goal. They want an

ambitious goal by 2031 to have a fleet

of over 50,000 scanners worldwide with a

scanning capacity of a billion scans a

month. So, that of course means if there

is 50K of these things and you have 30

days in a month, plus you have 24 hours

in a day, plus you have 60 minutes in a

hour, plus you have 60 seconds in a

minute, putting that all over 1 billion

means that at 50,000, you're going to

have to be scanning somebody every 129.6

seconds. So, already at that point,

that's just not going to happen, okay?

Every single machine which takes 60

seconds to scan somebody is going to

have to scan out, scan out, scan out.

So, but this is an important number to

keep in mind, because this is kind of

the basis of all the other numbers which

increasingly just get more and more

absurd. So, they're kind of their goal,

even if they were to double this instead

of having 50,000, they had 100,000,

still makes it so that you're scanning

somebody every approximately 4 minutes,

which just feels like that can't even be

done.

And that's 24 hours a day at any moment.

People are being scanned in every single

place continuously. Isn't it kind of

weird knowing that that's going to be

like the least crazy number that you

hear all day?

Like this entire video from here on out,

sorry, it's it's crazy land time. Okay,

everybody, we're going to get a little

bit more nuts now. In their technical

video breakdown on minute 3:32, right

here they say they generate 306

terabytes of raw data. Slightly before

that, they say they reconstruct all the

data across 21 servers. They claim that

each slice of the body takes about 40 GB

of data to move through the system. If

you're curious what a slice looks like,

it's this right here where you took the

body image and each one is this really

small thin slice of your body, so you

can slowly see every single bit of

information. They're claiming it's like

a half millimeter to 0.1 millimeter,

something like that. People are saying

maybe it's 2 millimeters per slice, but

it they're they're thin slices. Now,

this is where we're going to start

seeing some conflicting numbers start

happening. It also says that the system

captures about 17 GB per second. Their

goal is to capture several hundred

slices of your body. Now, if they would

have said a couple hundred, that'd have

been 200. If they would have said a few

hundred, I would assume that's 3 to 500.

Several hundred, that I mean, you can't

use several and not mean at least five.

Can we agree to that? Can we all agree

to that? So, it has to be many many of

slices of your body. So, I said all

those numbers off pretty fast. So, we're

actually going to start breaking it down

a little bit more. So, first off, they

did say they do a 17 GB of processing

per second. So, that puts us at

approximately 1 TB of information

processed for your 60-second scan, thus

leading to maybe about 500 images. There

are some estimates saying 300 images,

but let's just say it's about 500 you

know, images. That is going to be 1 to 2

GB per image. And that seems to make

sense because these are going to be

high-resolution images of your inside.

They need to be medical grade. You get

the idea. There are going to be some big

images. Now, here comes the point where

all the numbers are put together and

things are a little bit confusing. So,

first off, you have the machine that's

like the tube machine. Then you're going

to have your 21 computers way over here

that are going to be the 21 servers in

which can have that information. There's

going to be some sort of obvious

connection betwixt the two in which the

data from this machine needs to be

processed over here because you can't

just take the raw analog data shooting

out of these crazy little vibrating

sensors and then just make images out of

them, right? There has to be image

processing. Now, it says something along

the lines of 800 TB of raw data. Let's

just throw that number out cuz that

number is just it's just it's it's just

crazy, okay? Even if you had like the

world's fastest switch at 1.6 TB per

second, that's 66 minutes to get the raw

data over from here to here if you had

21 connections, each image being sent to

each individual 21 computers, that's 3

minutes of transfer time even on the

world's fastest network switching, all

local assuming no latency and absolutely

no frame dropping and absolutely 100%

utilization. So, what I think they

actually mean is that somewhere on this

machine the 800 TB, the 40 GB per image

somehow gets reduced and then sent

through the system to get crunched.

Maybe it's the 40 MB or GB per image

gets sent through the system to the 21

machines. You could imagine that each

machine receives maybe 40 GB at a time

to process that one image. And to me

that could potentially make sense. If

that were true, that would be about 20

TB worth of information, which is going

to take a decent amount of time to

transfer even if you have many

connections split up between every

single machine. And remember, we need to

be pumping out a scan every 129 seconds.

The transfer rate alone is going to eat

into that 129 seconds. Now, let's just

pretend that data transfer is zero and

we just simply think only about actual

frame production. If you remember the 1

billion row challenge in which you have

to calculate a median across a billion

rows, or maybe it's the mean, I can't

really even remember. Effectively, the

challenge comes down to memory mapping

12 GB file at once, doing a bunch of

integer parsing, and then boom,

calculating out the value you're looking

for. And some of the fastest times I've

seen are right around 1 and 1/2 seconds.

I'm sure there's some ones that are even

sub 1 second, but they all tend to be

right around in this range are the super

fast ones. Now, remember, the raw data

for the image is going to be 40 GB and I

have a a general assumption that you're

going to be doing a little bit more

processing and parsing on these raw data

images than you would be doing on the 1

billion row challenge. I'm just I you

know what, I'm just going to assume it's

slightly more complicated, but you do

have 21 servers, which means if I'm

being generous and you have 500 images,

but you have 21 servers, but they each

take 3 seconds, you're at about 71

seconds to be able to process

everything. So, that already is way too

much time. Okay, you got to be you got

to be processing so fast to be able to

keep up with this billion scans a month

demand across 50,000 machines. This is

going to be eating into a huge amount of

your time. But somehow, that's not even

the most absurd part, okay? If you have

1 billion scans per month,

and you have approximately 500 1 GB

images that you need to be sending up, I

d- I don't know what that extra letter

is there. That is going to be

approximately like 16 exabytes.

Obviously, we need to you know, there's

going to be some G zipping, things are

going to get a bit smaller. Let's just

say we can get down to four exabytes.

Today of under today's amount of

internet, this would represent

approximately somewhere between 12 and

1/2 to 25% of the total internet

transfer today. Just to transfer medical

images. Even if we were be using the

fastest Nvidia server rack costing $10

million plus 40 acres of storage plus a

gigawatt of power to be able to store

all of these images across a single year

of operating, 25% of the internet and

you're going to tell me you're going to

be able to do a billion of these scans

in a month. Like I think all the numbers

are wrong. You're not going to hit a

billion scans a month with 50,000

machines. Okay, you're going to need

like 500,000 machines. Second off, just

the pure sheer power that machine's

going to take locally is going to be

nuts. The amount of internet traffic

you're going to have to be able to

utilize is going to be just insane.

Every single day you're going to be

Netflix by itself. You you I mean four

exabytes is what Netflix does every day.

You will be Netflix just off of medical

images. Not to mention just the insane

amount of data warehouses you're going

to have to have plus the powers. You're

going to have to have a dedicated

nuclear reactor just to power the hard

drives. And look at this, none of these

calculations considered really like

speed of reading from RAM or speed to

writing to disk or what happens when you

have more memory and it needs to start

doing a little bit of flippity floppity

out of the RAM space onto hard drive

space. Like there's all sorts of real

world problems that would exist that

processing this amount of information is

not going to be done in 60 seconds or

less. It is not going to be gone in 60

seconds with Nicholas Cage, okay? It's

going to take many, many minutes. And I

assume you won't be able to run the

processing of information while stacking

up more. I I genuinely have no idea how

they're going to get there, but I do

want to say something that I think is

pretty neat. Is that Midjourney, the

image company,

is attempting to do more for medicine

than OpenAI and Anthropic and all the

other places currently that I've seen.

They're at least trying to It appears

that they're at least attempting to do

something because the whole kind of

stated mission is that if you can

measure yourself every single month, if

any of your organs or anything starts

changing shape or density, you'd be able

to identify it pretty quick, hopefully

leading to like a quicker, you know,

root cause analysis of some sort of

issue, thus potentially saving many

lives or extending lives. The obvious

The obviously hard part that I just

can't imagine is how can this be cheap?

How can you get a billion people? Like

how do you make technology that is

affordable to a billion people with the

numbers that we were talking about? Like

these numbers, to be able to set this

up, is going to be a hundred billion

dollars. You're going to cost as much as

an AI laboratory to be able to operate

just due to the sheer amount of volume

of data and the kind of equipment you

need to be able to process that much

load. Because they were saying that a

spa can hold like 10 units. So 50,000 is

going to be like 5,000 spas across the

world. And each one of these are going

to have to have a 10 million dollar

server farm in it. I'm not a math

magician and I don't think anyone can

multiply 5,000 by 10 million. Those

numbers are both way too big to multiply

together. Therefore, there's probably no

way anyone can afford that. Okay, hey,

we're all broke boys compared to that

number. But at the end of the day,

still,

Midjourney, appreciate you

doing something a little bit different.

Uh the numbers, I don't believe. Hey,

I'm glad to be proved wrong, but I'm

happy that you're doing this. The name

is the Primeagen.

Also, please don't let this be

vaporware. Please don't let this be the

greatest rug pull of all time, okay?

Come on. Don't let it happen. Don't let

it happen.