HomeVideos

What The F**k

Now Playing

What The F**k

Transcript

213 segments

0:00

You know, I always thought it was going

0:01

to be the LLMs that make the end of the

0:03

world, but um I guess I was

0:05

significantly wrong. I just got done

0:08

reading this article and it's making me

0:10

feel all sorts of ways. Feelings I'm

0:12

just not used to feeling. Okay. Human

0:15

brain cells on a chip learn to play Doom

0:19

in a week. Now, you're probably thinking

0:20

this surely has to be some sort of

0:23

clickbait article. It's not like actual

0:25

human brain cells somehow wired up and

0:27

firing and actually playing a video

0:29

game, right? No, it is actual human

0:35

brain power being used to play Doom.

0:38

Okay, I have lots of things to say about

0:40

this. The first one obvious is that they

0:43

created a chip powered by human brain

0:45

cells. And what did they do? They made

0:48

the brain learn and play Doom over and

0:50

over again. What kind of weird purgatory

0:53

existence is this? You made something

0:56

that resembles some weird outline of

0:58

life and then forced it to play Doom for

1:00

all eternity.

1:02

You sickos. I know I'm like half joking

1:06

and all, but also it's like kind of

1:08

messed up, right? Like this is this is

1:10

just kind of weird. The version of Doom

1:12

they're playing doesn't look like real

1:14

Doom. This is some sort of mod andor

1:16

just like Teimu version of it. The fact

1:18

that you made a hellish experience for a

1:20

brain ship and then made them play Te-Mu

1:22

Doom, that is just not okay with me.

1:24

Okay, I'm not happy about this. But

1:26

first off, I figured we'd do a little

1:28

bit of background. So, in 2021, an

1:30

Australian company, Cortical Labs, used

1:32

its neuronpowered computer chips to play

1:35

Pong. And it looks like this. I honestly

1:38

I [laughter]

1:39

maybe I don't have enough brain cells to

1:41

understand what that even looks like,

1:43

what that's supposed to mean. If you

1:44

look at the game image, it's actually

1:46

doomed to play by itself. It's just

1:47

playing against a wall. Now, the chip

1:49

consisted of clumps of more than 800,000

1:52

brain cells on these micro electrode

1:54

arrays that can send and receive

1:56

electrical signals. And somehow that was

1:58

able to learn how to play Pong. Now, the

2:01

one that actually learned how to play

2:02

Doom used about a quarter as many

2:05

neurons as the Pong demonstration and

2:07

played Doom better than a randomly

2:09

firing player. Far below the performance

2:11

of the best human player. good, because

2:13

that means the best human player could

2:15

be matched by 200,000 brain cells.

2:17

However, it learned much faster than

2:19

traditional silicon-based machine

2:20

learning systems and should be able to

2:22

improve its performance with newer

2:24

learning algorithms. Who the hell forces

2:28

brain cells to learn how to play Doom?

2:30

Their entire reality is going around

2:33

hunting monsters with weapons and

2:35

picking up ammo and trying to find the

2:37

yellow, red, and blue key cards or

2:39

skulls where ultimately when you reach

2:42

the end of the first act, you have to

2:44

die. They actually created some sort of

2:46

Dante's Inferno brain torture machine

2:49

where you're just forced to be the Doom

2:51

guy and hunt demons. It's just actual

2:53

hellish experience. Think about it. That

2:55

poor brain just had to just process

2:57

nothing but violence all day. its only

2:59

existence was in violence and BFGs and

3:02

plasma rifles. But if you didn't have

3:04

that thought, your probably next thought

3:05

is why in the world would anybody

3:08

actually ever create this? What is the

3:10

purpose of having a brainbased computer

3:14

chip? Well, there's kind of like two of

3:16

them and this is where we get into like

3:17

the funny territory. So apparently the

3:20

realworld application that they're going

3:21

for is they want these controlling like

3:24

bio wetwware chips to control robotic

3:28

arms. They unironically want to create

3:31

the terminator.

3:32

>> The CPU is a neuron net processor a

3:34

learning computer

3:35

>> because apparently these brain cells

3:37

they just learn better than traditional

3:39

ML styled learning algorithms. The part

3:41

that's like the most terrifying of it

3:43

all. Besides for the fact that they

3:44

literally want to take brain chips and

3:47

shove them into robots and allow them to

3:49

walk around the world is that these

3:51

chips are actually controlled with

3:53

Python. Yes, that's right. The Cortical

3:55

Labs has developed an interface that

3:57

makes it easier to program these chips

3:58

using the popular programming language

4:00

Python. An independent developer Shawn

4:02

Cole then used Python to teach the chips

4:04

to play Doom, which he did in around a

4:06

week. You know what this means, dude? We

4:09

are going to be vibe coding robots. I

4:11

mean, we are we ourselves [laughter]

4:14

don't know.

4:18

Oh, we're going to be taking Python and

4:20

running just absolute critical

4:22

infrastructure software. Oh my.

4:25

[laughter]

4:26

Oh no. That means we're actually

4:27

entering into a world where pretty much

4:29

effectively everything that's on the

4:31

internet is potentially just made by a

4:33

robot and fake and then the real life is

4:34

just going to be filled with brain

4:36

chips. Like that's that's what's

4:38

happening. Somehow we're using brains to

4:41

make the robots walk around. It feels

4:43

deeply unsettling, you know, and I don't

4:45

know. And honestly, like just on on an

4:48

aside, this is really just kind of

4:50

feeling, you know. I I I must say that

4:52

this does not make me feel very good

4:54

because there's a lot more questions of

4:57

life as you add more and more brain

4:58

cells. What are we doing?

5:01

I don't I don't know if this is very

5:03

good. This just feels like a bad idea.

5:05

But I think my favorite part of the

5:07

entire thing isn't just the fact that

5:09

they're going to go freerange this on

5:11

some robot arms while vibe coding some

5:12

Python. It's that they don't even really

5:15

know how it works. Steve Ferour at the

5:17

University of Manchester, UK, agrees

5:19

that Doom is a significant level up from

5:21

playing Pong. But he says there's still

5:23

a lot we don't understand about how

5:24

these neurons are playing the game, such

5:26

as how the neurons know what is expected

5:28

of them, or how they can see the screen

5:31

with no eyes.

5:37

Oh my gosh. [laughter]

5:39

I mean, that in in of itself is just

5:41

like just that's an insane statement. We

5:45

have technology where we're just like,

5:47

well, actually, we don't know how they

5:48

learn or how they function or how they

5:50

see. We just know that when you give it

5:52

the data, it just works. And this

5:55

actually reminds me of one of my

5:57

favorite books of all time when I was

5:58

back in my master's day on intelligence,

6:01

which effectively one of the experiments

6:03

is that they took a ferret and they took

6:05

its eyeball eyeball cords and put it

6:08

where its ear holes should be and it

6:10

took its ear holes and put it where its

6:11

eyeballs should be. And guess what?

6:13

Couldn't even tell the difference. The

6:15

little ferret brain was able to take all

6:17

of that input and just be able to go,

6:19

"Oh yeah, I totally I totally get this.

6:21

Actually, nothing. That doesn't change

6:24

me at all." And now we're taking a maybe

6:26

what is equivalent to little ferret

6:28

brains and making them play Doom. Just a

6:31

true hellish experience. Your only

6:33

waking memories, your only knowledge of

6:35

life is seen through the old pixelated

6:38

Doom guy's perspective of just trying to

6:41

avenge his rabbit. Yeah, that's so

6:43

that's all I really wanted to yap about.

6:45

Um, dude, this has to be by far one of

6:49

the most interesting and also very

6:52

deeply uncomfortable stories I have ever

6:54

read. The name is Please don't vibe code

6:56

a machine and then put robot baby brain

6:59

intelligence on it and put it on a micro

7:01

electrode or array, whatever that means,

7:03

and then send it after me and my family.

7:05

Okay? Please, I'm begging you. Don't do

7:07

it. Okay? I Hey, I'm sorry for making

7:08

fun of vibe coding. I'm sorry for not

7:10

beating Doom enough. Okay? Hey, please,

7:13

Mr. Robot. Don't do it

7:16

again. Hey, do you want to learn how to

7:18

code? Do you want to become a better

7:20

back-end engineer? Well, you got to

7:22

check out boot.dev. Now, I personally

7:23

have made a couple courses from them. I

7:25

have live walkthroughs free available on

7:27

YouTube of the whole course, everything

7:29

on boot.dev, you can go through for

7:32

free. But if you want the gamified

7:33

experience, the tracking of your

7:35

learning and all that, then you got to

7:36

pay up the money. But hey, go check them

7:38

out. It's awesome. Many content creators

7:40

you know and you like make courses

7:42

there. boot.dev/prime dev/prime for 25%

7:44

off.

Interactive Summary

The video explores a scientific development where human brain cells grown on a chip learned to play the video game Doom in just one week. The narrator discusses how these 'bio-wetware' chips, developed by Cortical Labs, use hundreds of thousands of neurons to process information and learn significantly faster than traditional silicon-based machine learning. The technology involves interfacing the cells with Python and has potential applications in robotics, though it raises unsettling ethical questions about the nature of life and the mysterious way neurons can 'see' and interpret data without traditional sensory organs.

Suggested questions

5 ready-made prompts