Anyway, so PewDiePie probably has done

some distillation is my guess.

[laughter]

Is that Is that what we're about to find

out? Peudes is just distilling poor

anthropic and Daario right now is just

dying on the inside.

>> Seek 2.5 way bigger model than mine.

Facebook's flagship model llama for

Maverick. [laughter]

>> Does anyone even consider Facebook at

the race anymore? when you're like

Facebook's, you know, flagship model,

I'm just like, oh, okay. Yeah. Like, is

this equivalent to Carpathy's nano GPT?

Like, what are we looking at right here?

Model B Lama for Maverick destroyed. And

most importantly of all, my model

outperforms Chad GPTs 4 [laughter]

in like November or something. That

works. This sounds like impressive, but

it's way less impressive once we

>> No, first off, it depends on what he

actually means by that. But if if you do

out compete 40, that's actually super

impressive. So, I don't know what's

about to happen in this video, but hey,

I'm in. That's cool. In like November or

something that

sounds like impressive, but it's way

less impressive once we get into it, and

we will get into it. It also sounds way

less impressive considering the fact

that I almost burned down my house twice

for this project.

It also sounds way LESS IMPRESSIVE. I

LOST MY GODDAMN VICINITY FOR THIS. NOW,

I have not created my own AI. I have

merely taken an AI model and trained it.

It's like stealing a child on the street

instead of birthing one myself. It's way

more effective that way. Plus, it would

cost million. What? Okay. I mean, that's

it's true. I guess stealing a child is

way more like time effective than than

creating one yourself. True. True.

Absolutely true. It cost millions and

millions of dollars in infrastructure,

which I do not have yet. Am I talking

about birthing or But it's important for

you to understand just how naive I was

going into this cuz I knew nothing about

machine learning, training, AI, and

coding that I mentioned. My model is a

coding model. I can make a coding model.

Sure, Felix. Great idea. But I also know

I wasn't that crazy going into this, and

I'll explain why. But mainly the fact

that I wanted to do this was because I

wanted to learn.

>> Is he doing what I think he's doing?

Whoopsies.

Is

Yeah, look at this.

He put the He put the He put the little

microphone on his fingy. snapped it on

to the old figure like most people hold

it. He just snapped it on. The man's a

revolutionary genius. I swear he just

does things and I just I don't even know

what to do with it. He's just uh also

just to be completely fair, him saying

he knows nothing about coding but going

to build a coding model. Uh I mean

technically that makes him on par with

every other AI ML researcher right now.

Great idea. But I also know I wasn't

that crazy going into this and I'll

explain why. But mainly the fact that I

wanted to do this was because I wanted

to learn. The reason I'm here with this

video and the series of events has just

been me approaching this philosophy of

just yes this might be difficult but I

will learn from it step by step. You

know that takes you places. And that's

why I'm so excited to announce today's

sponsor which is boot.dev. boot.dev Dev

is a website. I teach it to you how to

code. But for real, I did their Linux

course and it's there is a nonzero

chance I might accidentally show up by

accident in this video. Very exciting.

Okay, very very exciting. Also, by the

way, just so you guys know, my audio

again is kind of goofy. Every time I

pause it, it requires me to like go back

a little bit cuz there's like a few

seconds of silence. So, if you're

wondering why I'm repeating it,

fantastic. I don't know what's

happening. When you actually understand

how something works, it changes things

for real. It's none of this Dualingo

bing fake learning. Okay, it's fun. It's

engaging and it works. I'm super excited

to do their other courses as well.

Create your own AI agent in Python

course. I'm going to say something that

makes me sound like such a boomer dad,

but I am at this point. So, to heck with

it. Instead of focusing on gaming too

much lately, I just been focusing on

leveling up myself. But it's true. It's

such an amazing feeling and I want you

guys to experience it as well. And I

think boot.dev is an amazing venture to

do so. So check it out in the

description and I will remind you later

as well. Okay. Now, funny enough to

dude, he's right. I hope you took that

little QR code. Do you see that little

QR code? I'm going to give you guys one

more chance. Okay. Well, the problem is

that hopefully that doesn't screw it up.

That hopefully that title [laughter]

>> true. It's such an amazing feeling and I

want you guys to experience it as well.

And I I can't believe Peudes is is a

better modern philosopher than most of

tech Twitter. You don't need to learn

[ __ ] Peudes is like actually as you get

older, it turns out it's really

meaningful and awesome to actually

become an expert in something and to

feel like you're actually improving at

something. I can't believe, dude, we

live in just the strangest world. Peudes

is now the modern philosopher. Funny

enough, to train my model, AI model, it

ironically it works very similar to how

you would train on yourself on boot.dev.

There's an instruction of a problem. You

get a framework on how to initiate or

what to use and then there's a validated

answer at the end of it. Basically, I g

around 100,000 samples like that [music]

and then you feed that to the model

which slightly nudges its parameters

like it slightly probably nudges your

brain [music] and bada bing bada boom

you have trained your model. It's kind

of like this. You're AI. Look at this.

Are you learning?

Are you getting this?

>> You will look at these. You will look at

100,000 more. Play. Pay attention.

>> This might take a while. The model. Hey,

to be fair, humans are super efficient.

Like if you really think about it, it

takes a baby from zero to to like one to

figure out how to walk. So it's only

seen however many hours of real life

from zero to one. And and for the first

couple months, its eyeballs didn't even

work. And it doesn't have a developed

nervous system that far into this entire

situation. It didn't even have muscles

and it figured it out. It takes AI like

billions upon billions of hours to be

able to do anything. You know, humanity

hyperefficient when you really think

about it. This little fellow right here,

hyper efficient learner. Okay,

>> hyper efficient. This might take a

while.

>> The model that I used was Gwen 32B,

which is already amazing at coding, but

I needed it to be amazing at coding.

>> He's also doing instruction tuning. By

the way,

>> the whole reason I decided to do this

and where I landed, cuz I discovered

something. There are many ways to

benchmark and test your model. I

discovered there's one called Ader

Polyglad. If you remember last video, I

used the agent ader to code my own web

UI. This is

>> Am I allowed to show this on Twitch?

What do we What do we do with this? Am I

Am I about to get banned? Is legendary

gooning lord Dan Clansancy about to come

in here and ban me? I'm sorry, Dan.

>> This is a respectable benchmark [music]

that has coding in six different

languages. That's six more languages

than I know.

But what was interesting about this is

[music] that state-of-the-art models

like Chat GPT perform like garbage on

this benchmark. 18.2 2% uh what my model

that I was planning to train on performs

8% [music]

trash but it performs 16% if you use a

different format called whole format

instead of diff format. Basically

there's two different formats but one of

them is important. How do I explain

this?

>> It's basically like this. You draw a

picture. Okay, imagine you draw a

picture and you want to add a cloud. But

instead of just adding the [ __ ]

cloud, you redraw the entire picture

with the cloud. It makes no [ __ ]

sense. But basically, a lot of

>> Did the original one The original one

did not have a

>> the cloud.

>> The original one did not have a a

dingleberry. He added more than just a

cloud in that. Okay. Hey, hey,

there is some bonus content in this one

that I was not I did not think we were

allowed to have bonus content of instead

of just writing the whole thing. It's

just stupid. It wastes compute and it

wastes time and I never used it as well.

So I thought hey if I can just fix the

format I will make my model 16%. And I

will almost beat Chad GPT. The goal beat

chat GPT easy because what I had on my

side at my disposal my arsenal was

Chinese AI research. You'd think China

would be on the more censorship side of

things. At least that's what I thought.

How wrong I was. It's completely the

opposite. Deepseek China Chinese AI

basically just [music] released their

model for anyone to run and a whole

document with their entire training

process in great just don't ask about

the three T's. You know what I'm

[laughter]

[laughter]

detail? It's amazing. There's so much

information from these research

documents combined with the open source

community. There's so much to try and it

just makes you really want to try it

yourself. At least that's how I saw it.

Even though a lot of it was super

advanced and I didn't understand

anything, eventually I did. I [laughter]

think

it's also hilarious reading these

documents. Chinese AI researchers are

unironically comedians cuz they write

the most unh [ __ ] They say [ __ ] like,

"Oh, we trained our model on 248

GPUs." You're like, "What? That's like

$60 million." And and you keep reading

and they're like, "We emphasize this is

an economical APPROACH." I'M LIKE,

"ECONOMICAL? WHAT'S THE non-economical

version?" But I think that kind of puts

things in perspective and why a lot of

companies in the west, [music] they

don't want to share information. So,

with Chinese AI research on my side and

a boot dev core,

>> to be completely fair, 24,48 2 to the 11

uh AH200s is an extremely cheap amount

of GPUs to be able to train. And the

reason why is because they uh you know

the whole uh

hey chat GPT

yo dog, how how does quantum mechanics

work? Yo, ain't give me that knowledge.

[laughter]

I that's that's what happens. ready to

do this thing.

Now, what do you need to train AI? Data.

Now, how do you get data without

sacrificing your soul in the process?

Well, there is options, believe it or

not. You can mine the stack, which is

this 60 TBTE data set that you're

[music] allowed to train on. Good thing

I kept my hard drives around. You can

use publicly available data sets. Yep.

You can also mine git, which is a little

more of a gray zone cuz you have to

check for licenses, which yeah, you just

you just hit all the MIT licenses. It's

actually something I did want to see is

like how good could you make nano GPT if

you just use uh MIT licenses and like

hey can you make an okay inline

autocomplete from MIT maybe which I'm

sure everyone is doing right [laughter]

these big tech companies they're they're

being super ethical and great about it

I'm sure that's why they're not sharing

any information they can also synthesize

your data glorious synthetic data by the

way is PewDiePie becoming the next uh

Rossman Is this what we're Are we in

real time? Oh, hold on. Time out. Time

out. Are we in real time watching the

guy who gave us Ligma? The Ligma Chad

to like Luke Smith pipeline, Vim Diesel.

Is this what we're watching like right

now in live action? Like, where are we

inside the cification of it all? Is Is

this what's happening? I think that's

what's happening. Wow. Don't taste so

good. I tried every single method there

is. It's been a mess. This is scraping

git or enriching my data. This is

scraping for more data. This is running

testing on the data. And this is

augmenting the data. And this is my

eight LLMs. This is the level we've

reached. This project was kind of like I

was like in the middle of a freeway and

there's a bunch of cars direct that I

had to direct constantly because [ __ ]

had to move for this to finish.

Everything took so much time and my GPUs

had to constantly be cooking and I had

to do all this makeup cosmetic surgery

to the data cuz all these developers

writing [music] poor lazy first of all

why do you even publish some of that

lazy whoa whoa whoa whoa whoa whoa whoa

whoa whoa whoa pudes what do what do you

mean commit I don't know that's dude

that's that's obvious what that one

means okay it means that they went on

stack overflow or asked an LLM them

about something. A bunch of code was

generated. It worked and they didn't

understand what happened. The second one

is also evidently clear. This guy's

working on CI right now. Do you have any

idea what CI stands for? It stands for

Dante's Inferno. Okay, it's just it's

awful. Okay, that's how you spell Dante.

It starts with a C. It's it's actually

insane. And so he's just praying, please

work. This is my one chance. This is my

one chance to actually have it happen.

And then this last one is also very very

obvious. Okay, some guy started fixing a

bug and then he accidentally had an idea

for a feature and then found three more

bugs along the way. And so a lot of [ __ ]

happened and you can't just like say all

that. So you're just like, yo, there's

just like just just changes. Okay, we

made changes. I don't want to talk about

it. Just a lot of stuff happened. Like

this is really good commit commit

messages. I know what happened on all of

them. Lazy instructions. Don't make a

commit if you're just going to add a

comment. But finally, I had my data, but

I still knew I needed more. And I also

wanted to try out synthetic generation

of data, which is basically you

typically take the strongest model that

you have and you ask it, hey, look at

this. Make more in this format. And my

god, it was a beautiful thing. You get

the perfect data exactly the way you

want it. It's amazing. But the problem

is, and maybe you already know this, AI

is wrong all the [ __ ] time.

[laughter]

>> Okay, it's basically like this. I tell

AI,

>> you know, also what is crazy. Can I just

can I throw something out there? Uh, in

today's world,

why just why oh why if you say AI is

wrong all the time on the internet,

people are like,

uh, hallucin hallucinations have been

largely solved. What are you talking

about? Well, dude, if you're getting No,

no one gets hall. That That was like a

2024 problem. Okay. No, no one gets

hallucinations. Like, what are you

talking about? I get hallucinations all

the time. It just makes up [ __ ]

constantly. Like, what do you what do

you mean?

[laughter]

My drawing skills have really I'm sure

glad I learned how to draw. [laughter]

>> I was just thinking like what happened

to his drawing skills?

by showing him a burger. And then AI

[music] makes what looks like a burger,

but I open the lid and he put razor

blades in there. So my harness is a

burger eater [laughter]

to check that if the real burgers are

made instead of fake burgers. I think I

explained that pretty well. You lied to

me. My synthetic approach, I know that

most people don't care. I used oss

instruct from magic coder and evol

instruct which is an amazing method.

It's basically like that cloning dancing

guy technique. You get feed a code

snippet and then you say hey make it

into this format and also make it do

another one but make it more advanced. I

don't want to get into this technical it

doesn't matter. this video would take

forever. But the problem I had was I was

not getting enough of it and I don't

trust

>> also like how do you know like if you're

not a coder, real talk, if you're not a

coder, how do you know you're even

generating the right thing? Is this why

chat GPT puts like if statements and

stupid guard checks around everything?

I don't know if you guys noticed that,

but chat chat jippity 35 codeex is just

like, oh, hey, is that a function that

takes in a number? Hey, I'm going to do

a quick type check. Hey, is that a

number? Are you sure your number? Hey,

if you're a number, are you like a whole

number? Oh, uh oh, you're a decimal

number. Hey, if you're a decimal number,

are you at least like greater than a

hundredth? It's just like, dude, what

are you doing? I just asked you to do

the absolute value. You don't we don't

need to do all this. Just just put put

the bags in the put the fries in the

bags, bro. Like, it's not that hard.

It's crazy how much if checks it does.

So, I validated it. I kept thinking the

problem was my test harness, so I needed

to FIX THAT. WHAT AN IDIOT.

THAT JUST MADE IT so I passed more

garbage. So when the time came for me to

finally train my model after months,

months, I was so excited. I ran the

benchmark and I had made an AI model

that finally

>> is worse.

>> That was worse. I had made it worse.

I probably should have quit then, but I

am way too stubborn for that. I just

can't. I had I don't have it in me. I

can't do it. Nice. This is when you

realize AI is kind of like magic, right?

But it also is garbage data in, garbage

data out. There was a lot of issues with

my data. When I had fixed my harness, I

had just let more garbage data passed

through. There's also all these other

issues like empty white spaces and

classic coding issues that I just wasn't

aware about. So that's why I had [music]

made it worse. But I gave it another

attempt. And this time, no more

mistakes, no more dillydling. lock in

again. The model is worse. [laughter]

Oh my gosh.

Oh, the brother locked in and actually

made it significantly worse.

[laughter]

Oh, that has to hurt a little bit. That

has to hurt a little bit. There's a meme

here. There's there I just feel it. I

just feel like there's a meme here.

Right here.

Something like that. There's something

out there. Okay. Okay, I don't know what

to do with it. I'm just going to let it

go out into the ether. Somebody will

come up with something way funnier, but

we're going to call it We're going to

call it even. Okay, we're going to call

it even. It was my harness yet again.

I don't know why I got so stuck on this

[ __ ] thing. I didn't even need

synthetic data. I just wanted it to

work. I just wanted to work so bad.

Finally, I had fixed the [ __ ] thing.

And the benchmark came in 16%. Sometimes

15, sometimes 14, but 16 was the

highest. The mod is not going to solve

the

>> Why not just I'm c I I Is it just

because he wanted synthetic data to win

so bad? Why not just GitHub MIT that

crap? GitHub MIT that just get it. Just

get give me all the MI.

>> Honestly, it's not going to solve the

same problem every time. It it just

doesn't work like that. There is a level

of randomness to it to the benchmark

performance, but 16 was a ceiling. And

that makes sense because that was a

ceiling in the official whole format or

whatever. And I had fixed it. So it's in

the diff format, but I had not made a

[ __ ] difference. I had not made the

model smarter. And remember, I need to

beat 18%. To say that I beat Chad GPT,

otherwise this means nothing. So my plan

of attack was to add reasoning to my

data. Adding reasoning to your data is

basically making it write out some

thinking before it solves the problem.

Instead of doing two plus two in your

head, you go, "Oh, okay. So, let me

reasoning to your Why is that mirrored?

Why is he always trolling us?" I swear

every image in all of his videos,

there's just like some small amount of

trolling. This is not some sort of like

right to left language. This is just

English inverted. Look, T h e, right?

Like just right there. It's just just

messing with it's just he's just trying

to get my brain

>> up there. Some thinking before it solves

the problem. Instead of doing 2 plus 2

in your head, you go, "Oh, okay. So, let

me break this down. So, I'll have two

apples and then if I add two more apples

and then I count them all together, then

I will have four." This is really

effective for complicated issues to

break down the problem into parts and it

really can improve the performance of

the AI but when it's simple questions

and it still does it it can be very very

irritating as well. You've probably seen

this actually if you use a stronger AI

model. A lot of times you ask it

questions, it goes, "Oh, absolutely.

Let's break this down into parts. First,

we're gonna you're like, "That was a

simple yes or no question. What do you

mean?" But it's a really effective way

to make your model solve problems more

accurately. And a lot of these smaller

open source models that I train on

struggle with this because they just

haven't been trained on it enough. So

that was my plan of attack to make my

model smarter. and I found a study that

showed that 10% in performance. I'm like

10% that's way more than I need, baby.

Let's just clone this git repo and get

going. The only problem was, of course,

that the ungodly level of computation

that was needed for this. And that's

when things were starting to smell

funny. I realized my room had this weird

aura to it all of a sudden. I swear it

didn't used to smell like that. So, I

decide to reboot my computer and

Lightning strikes all across it. Smoke

starts pouring out. My whole life

flashed before my eyes. I turn off the

computer, but the damage was done. One

of my GPUs [music] had died. Rest in

peace. I tested each one one by one. And

it seemed like everything else was okay.

It was just this one problem. And

looking at my purchase history, that one

GPU came from a different factory. You

have to understand what a hack job this

setup that I have. By the way, I love

that is like what a hack job his setup

is. And it's it's like the single most

beautiful

home setup I have ever seen. Yeah. It's

like my huge hack job. Oh, by the way,

you just excelled beyond everything I

have ever seen anybody just do by

themselves. Like that is beautiful.

That's a beautiful case. Does a great

job.

Okay. I mean, that one's less that Okay,

that's less of a good job.

That one's not as nice. But I mean, but

this one right here, that one's really

good.

It's very nice. [laughter]

Not so nice. That's not so nice. That's

not so nice. I don't know about that

one.

>> Bifurcated. It's undervolted to death.

These GPUs run on 450 watt. I run it on

175 watt just so my house wouldn't

[ __ ] crash every time. And then these

are hacked Chinese 4090 GPUs. This whole

thing is held together by prayers.

Literally in Japan, they sell these

prayer infused it badges. Where is it?

It's in front of my computer. Official

Japanese priests.

I probably have now blessed my computer

and I was ready to give it another shot

only for the smell to appear again.

After sniffing my GPUs like a maniac, I

concluded my GPUs was not the problem

this time. And eventually I found it. I

don't think it's supposed to look like

that. Again, I was using cable that was

graded for 1,500 watt. I was running

over 2,000. Change cable. We were good

to go. You know, I was planning on uh I

I had we had to do some power work at my

house and so I actually had uh some

extra wattage brought over to this room

cuz I was like, "Okay, you know, I

should probably maybe I should think

about building my own GPU rack."

I I don't know, bro.

I don't know. I don't know if I want all

of this. This kind of It just kind of

worries me about the whole the whole

situation. I don't really know what I'm

doing. Uh I don't want to end up

smelling I don't want to end up smelling

GPUs. You don't want that smoke. I don't

know if I want that smoke. Burning down

your house is really dangerous. True.

True. It's true.

It's actually really true. Like the the

more you burn down your house, like the

more likely you are to die. Okay. Hey, I

just want to throw that out there.

That's like it's it's you can't escape

it. Kept crashing still.

>> Oh, it's still thought it crashed. It

has crashed. Oh, [ __ ] What a pile of

You'd think training would be just this

straightforward thing, but it's really

not. It's taking too long to train. I

need more computer. Okay, what am I

going to do? It's not my fault. I know

what you're thinking, Felix. Are you

building another computer?

>> No, I am just extending the one I have.

[laughter] Of course. Of course. I'm an

epic minimalist.

>> He's Dude, Felix, I sorry, Mr. Mr. uh

Mr. pie. Uh, by the end of this video, I

swear he's going to be renting his own

place, being like, I actually had power

issues and I needed to be able to get

more power, so therefore I had to go and

rent my own facility cuz there's no way

I'm going to be able to rent this in my

in my like standard housing unit. I need

I need all the powers I can get. I feel

like that's where we're going. My

bathroom and I drilled a hole.

It's super heavy on your computer. And

every time it crashes, I have to [ __ ]

defibrillate it back. Bring it back from

a coma. It's not pleasant. It's And with

everything that had happened in the

past, extremely stressful. And I really,

really, really started doubting what the

[ __ ] I'm doing here for these 2%. I

started calling upon Deepseek API

because it's practically for free. And

eventually I had 15,000 samples which is

way less than I had aimed for but these

were the top of the top the creme the

cram samples. So hold on let me get this

straight. Okay. Okay guys we're we're

going to run we're going to run some

numbers here. Okay let me jump over here

really quickly. Let's open up [ __ ] GNU

image manipulation program. Uh in which

some people are offended by its name.

You can shut up if you are. Uh so with

[ __ ] let me get this straight. There's

something called open code. uh or sorry

not open code. Sorry, sorry, sorry,

sorry. I got DAX on the mind. I don't

know what happened there. There's Open

AI. Then there's something called

DeepSseek. Right now, what DeepS did is

that they went and requested a bunch of

synthetic examples from Open AI, stole

it,

and then used that to train their model.

So now Peudes, Mr. Mr. Pi over here, all

right, is doing the exact same thing to

DeepSeek. And now I'm no mathematician

but if a equals b and b equals c

therefore a equals c which in this case

means that pewdiepie is actually taking

from openi

okay

brain draining one of the great American

companies. Okay they're stealing. Okay.

Hey PewDiePie. You can't steal. Stealing

is wrong. Okay. Here's the thing. Okay,

hold on. So, if you just go on GitHub or

you take some books, that's one thing.

Okay, but if you go to a company that

has stolen the entire internet and then

proceed to steal their content, that is

unjust. They spent money on the on the

new stuff, therefore that's bad. I'm

just throwing that out there and that's

unethical. Beautifully crafted

step-by-step reasoning the world has

ever seen. I did my supervised

finetuning three epochs and I ran the

benchmark and it scored

>> 17.

Are you [ __ ] kidding me?

>> But as I mentioned, the performance is

kind of random. So I kept running the

benchmark. I had eventually given up,

but I had done one more just for the

[ __ ] You know how it's so funny about

what he's saying that people aren't

putting together. How much do you want

to bet the benchmarks that OpenAI also

published? You want to guess? You want

to guess what they look like? This is

the Opus uh 46 tracker.

Yo, dog. Why does it do really good and

then the next day it goes from 60% to

51%. 60% 55%. 58 54 52 54. And this is

all in the exact same benchmark.

Nothing's even changing.

Look at the look at the variance.

And then, oh my gosh, you jump over to

to to Codeex.

You know what's so funny is everybody

last week on Twitter was just like,

"Dude, Codeex sucks."

You go here and you're like, "Oh, that's

because it's been sitting at like a 60s

pretty much average the entire time."

And then proceeds to go down to a 45%

and everybody on Twitter talks about it.

Crazy. It's actually crazy that you can

go, "Oh, yeah. I see that. Look at the

weekly trend." [ __ ] the bed. [laughter]

It's nuts that like I can actually line

this up to reality.

I know. Anyways, holy [ __ ] This one is

having like a god run. It's at 40%. What

is happening? It drops to 30. I'm like,

"Hold, please hold." It drops to 25. It

drops to 20. And it finally finishes

all the exercises in the benchmark. It

is done at 19.6%.

I have beat chat.

It felt so goddamn good. STOP. STOP.

DON'T LISTEN to this guy. The benchmark

is invalid. I did not check for

contamination before the benchmark.

Basically, if you grab data all across

the internet, there's a high chance that

you're going to grab data that might

already exist in a benchmark somewhere.

So, you have to check for contamination.

I didn't check for contamination. I

didn't want to check for contamination,

but my stupid conscience was like, I

should probably do it.

>> I was backing up, going through my data

for the gazillionth time, and I realized

there was some contamination. It's not a

huge deal, but it's like if I just am

honest, whatever. To me, it's not good

enough. And I want to clear out my data

and I'm going to retrain again,

benchmark again. Oh my gosh, it's

painful watching this because I know for

a fact that

PewDiePie is more ethical than the guys

running those state-of-the-art models.

Like, oh my gosh, what are we doing? Oh,

why? Why is ligma man more ethical? This

is just my whole reality is breaking

before me. It's just it's hurting. Okay,

I'm in physical pain. All right.

>> And I think it will still give me the

same result cuz it's such [music] small

contamination. I'm kind of worried cuz

I'm running out of time. [laughter]

>> Have I done [ __ ] all this entire time?

Have I achieved nothing this entire

time? And the whole thing was a hoc

pogus. That's what I thought. So this

time I went all out. I trained on my

entire data set. Previously, I trained

on a small subset of what I thought was

my best data because it takes forever to

train. It takes forever. It takes days,

weeks. And since I reached the score of

19.6, I was like, uh, I'M DONE. BUT NOT.

THEN I made another discovery. I was

training on the wrong model. A major

update. I'm watching I'm watching my

video. this guy. I'm giving feedback to

my editor and I'm like, "Hold on, hold

on. What is that?

That is not the coder version."

>> YEAH, DUDE. OH MY GOSH. I NOTICED THAT.

THAT WAS EARLIER in the video and I was

just like, "Oh, weird. It's Quen too.

It's Quen whatever." And I was like,

"Huh, okay. That's interesting that he's

using Quen whatever. Not a big deal."

AND THEN HE HE'S TALKING ABOUT IT. OH, I

WISH I WOULD HAVE SET OUT. IT WOULD have

been so many IQ units. Oh, no.

>> Oh, I [laughter] the bag so hard.

>> Have I been training on the wrong

version this entire time?

>> Oh,

so maybe this has all been a blessing in

disguise because I feel like with these

two in my things in mind, I I should I

should beat Chacha. I should. But we

will know in a couple days. Did I beat

Chachi?

Well, I trained on the coding model and

the first score

4.4%.

I don't know what the [ __ ] happened. I

can probably think of five things, but I

don't care to figure it all out. It's

going to take forever. I just retrained

again and the model score 25% baby.

I have not just beat Chachi. I beat

Chachi twice. They're

that's really good.

That's a big improvement. Wow. Okay.

>> August shitty version as well. I was so

relieved. I was like, "Thank God."

>> You know, there is one problem though

that he's not saying, which I I I feel,

you know, I feel like especially a guy

with such uh, you know, such integrity

as him is that he's not mentioning it at

all,

which is Jippy 40. though being worse

will make you feel incredible. I mean,

the sycopency of 40 was so good that

like you could ask it anything and it's

going to love you and it's going to make

you feel loved and seen and it's going

to cause an entire group of people on

Twitter that if you say that thorough is

ridiculous and people that are addicted

to AI and are causing their entire life

to be ruined will go after you and your

family and will have absolutely insane

mental breakdowns live on the internet.

That's how good that model is. Okay? You

can't tell me that a model's you can't

you just can't you cannot convince me

that there has ever been such a great

model as 40. No one has created a

stronger group of sick infancy than

Jeepy 40. It was incredible. I was like,

"Thank God." But I made another

discovery. Okay, it's not over. A third

of the benchmark was not even running.

C++ and JavaScript just wasn't being

tested properly. Oh, that's okay. Okay.

No, nobody likes those languages

anyways. Okay. Nobody even cares about

it. Honestly, Pew, I think you're I

think you're fine. I think we can all I

think I can speak for everybody when I

say that if you never touch JavaScript

or or C++, you're living a good life.

Okay. A good good life. I run it again

and the final score

36% baby.

This means I also beat Google Schmoogles

thing as well. And I think I beat I beat

4.1 or something. I don't know. Chachi,

it's a massacre out here. You're being

destroyed. It's embarrassing. Open AI

stock just demolished. Just quit

already. This is what pops the AI

bubble. This [laughter]

That's actually really impressive.

Generated a bunch of synthetic data and

did that well.

That's really good.

That's like really really good. Is is is

Peudes is Peudes about to start

accepting investments and creating the

first Japanese-based AI company with

Japanese hardware prayer tags on every

GPU?

Is that where we're going? Is this the

universe we're about to land in? I was

like, there are still issues. I'm going

to do some post training. 1,500 samples,

splinky blinky, five epochs, and I ran

the benchmark again. Pure

decontaminated, let it be known. Purer

than the fountain of youth. 39.1%

baby.

Another destroy. I think this one I beat

Google Smoogle. I Where's the stupid

benchmark? Yeah, Gemini. Gemini Pro. You

guys pay for that. I did not think this

would even ever come close. But here's

the embarrassing part about all of this.

>> That does have to hurt Google a little

bit. Okay, this guy right here very

disappointed.

He has spent so much time.

Now look at him. Now look at him.

Glasses entirely too blue.

Doesn't even know what to do with his

life anymore. He's just very very upset.

Five. But Gwen 3 is out and it scores

40%.

So unless I beat 40% this means nothing.

[laughter]

And I'm I don't I'm out of time. I don't

have [ __ ] time. I need to send this

video to my editor. Yes, there was one

more thing. A model being good at one

benchmark is stupid as [ __ ] Okay, I

need to test this model on other

benchmarks as well. I want to test it on

Sweetbench, all these other coding

benchmarks. Unless you're improving that

it see future and said what benchmark he

was going to say and pulled it out.

>> I got that future site.

>> I don't even know. I am just out of

time. Okay, I don't have time to do it,

but I will. And if this model ever gets

to a point where I feel like it's good

enough, I would love to share it, but I

think if anything, I might just move on

to a different project in in the

background because it takes a long time.

And uh I'm kind of tired. That being

said, you've seen me fail a lot in this

video. I have become so accustomed to

failure, you have no idea. I've almost

given up on this project so many times.

There are so many times where I'm just

like, I don't know what I'm doing. This

is the stupidest thing ever. I have

graveyards full of just garbage,

debunkle, schmunkle, deformed data that

I have generated thinking this is the

best. [laughter]

I have gone through the whole alphabet

of failures. I was just so way in over

my head on this project. But I think the

number one thing I've learned, how do I

explain this? When you install Linux,

here's what happens. Linux the creator

becomes your godfather inevitably. And I

was watching one of a random video of

him talking and he was talking about how

he's doing this project and he was

failing. But that's okay because that's

how you learn. Uh some people think that

failure is a bad thing and I happen to

be one of those people who actually

enjoy doing

things I'm not good at

because it's how you learn.

>> Damn, Twitter could really use that

advice, dude. the mo the modern AI crowd

that's just like

it's all about taste now. You can taste

these nuts. Um so it's just like man

this is so good. I again I cannot

believe we're living in a timeline where

like the CEO of Y Combinator feels

incredibly less wise than PewDiePie.

Like what the he what is this world

we're living in? This can't even be

real, right? Am I Am I wrong here? Like,

this it's just No, no, no, no. I refuse

it. But yet, here we are. I don't know,

man.

I love the fact that Peudes is going on

this journey and doing this because

honestly, there are so many people that

just need to be encouraged to see like

failure is actually really good.

You should be measuring your growth not

in just successes, but more importantly

in how many things you've been failing

at. Like honestly, people don't realize

like just how valuable that fail fail

fail succeed path is. It's incredible.

It's just oh my god.

>> I'm like he's speaking to me right now.

Oh my god. But I really feel like that's

the main thing I've learned from all

this because there's so much to learn

from failure. Learn from it and iterate

and keep working. I think if you have

expectations of how things should go for

yourself, you're just going to get

disappointed and you're going to want to

give up. So, expect to fail, embrace

failing. That's the message I want to

send out to you kids. Last thing, this

is a coding model and I think a lot of

people are looking at coding models like

are they going to replace and I saw

Linus say this as well and he was

basically saying coding models if

anything it will just bring in more

people interested in learning how to

code again. That was online as tech

tips. very very good message. Absolutely

loved it. He said that the, you know,

the genie's out of the bo the bottle.

Pandora's opened up the box. They're

here to stay. But does he think No. He

just thinks it's going to end up

[clears throat] aligning closer to a

tool.

Good take. Good take, Lionus. Good take,

Lionus. Like, I would never have learned

wanting to learn how to code if it

wasn't for AI coming into the picture.

So, on the final note, learn something

new. Check out boot.dev. I highly

recommend it. It's a great course. Pick

out whatever you want. Challenge

yourself into something difficult that

may be above what you think you're

capable of. That's it.

>> Now, scram.

>> I'm just kidding, bro.

This month, we're traveling a lot. And

what always happened whenever we travel?

The inevitable connecting to public

Wi-Fi. Free Wi-Fi is a trap. Get the

reference. If you connect to someone

else's Wi-Fi, you might as well just

give up your credit card and banking

[music] information and all your

information they ever have. You should

always use a VPN. NordVPN. Say it with

me. Nordmeos

board connect always. I'm even connected

right now.

>> Now that we're going that direction,

>> I made a little module for me.

>> I wasn't ready.

>> Cuz whatever I do online, if I want to

download is my goddamn business. It's my

[clears throat]

It's my goddamn right. So, thank you

NordVPN for making it possible to free

yourself, protect yourself online. If

you go to nordvpn.com/piepie, you get a

huge deal on a 2-year plan, plus bonus

extra nordvpn.com

online privacy. Seriously, this is the

best deal for NordVPN you're going to

find. So, take advantage. Thank you,

Nord, for sponsoring this video. That's

nordvpn.com/pie.

>> Very lovely. Hey, that was really good.

again. I can't believe how good, dude.

He's just so good. It's just so good.

Absolutely good. The name

this is the primogen.