Well, he already has drawn most of the stages he'll need, just has to have the network listen to the audio and choose the appropriate visual from the list. Animating/transitioning them would be a separate matter
I don't really know about neural networks, but I think that you'll need a lot more samples than just what's in the video, and it would probably be simpler to hard-code such a thing, but great idea!
Wouldn't it be more effective to use a database with passport photos? The photos you used have a lot of different angles and expressions, whilst passport photos almost always have the same angles and expressions. I think it would improve your friends' faces quite a bit.
Tortillian You could also combine biometric photographies with non-biometric yet frontal photographies of faces so the neural network first learns about how a face looks like in general and then learns about the individual expressions.
They don't need to be celebrities, though. I've seen plenty of people that aren't known that still pass the standards of beautiful. Also, let's say the A.I. took an ear from a picture that is at an angle, and the rest of the face from front-facing pictures, I doubt that generation would be beautiful.
Why do you show us your "own mediocre work"? Because making learning entertaining is really the hard part. Now I am interested in a subject I never would have thought about before.
You should make a neural network write one of your videos, make the voiceover and generate images to show at various parts. I dunno, I just think that'd be really awesome and creepy at the same time.
yo thanks for making these, your stuff is probably the only way id be able to sit through hearing how neural networks work and youve probably opened a lot of eyes to how cool this stuff is getting
There are several ways that you could extend and improve this neural network to make it more effective, due to the differences between gender, age, and skin color. In fact, you could filter the images with another neural network, inputing only people who have a similar skin color and gender, although age may be harder to factor in.
Henry Smith your lapis not bubble lol I bet you eat yoylecake every day and you probably give gluten a red food Color and put on top and then you put fries (after he grows his growtatos) then inside you take a page from book and put it in the yoylecake
I was going to make a comment about you being like the announcer (since your name is metal metty and your picture is Mettaton) but then I got the image in my head of Mettaton as the announcer and found that very funny ...or maybe you just ate a yoyleberry. Either way, hugely looking forward to the next machine learning video (man I love this community!)
To anyone replying to that guy, It's best that we ignore the trolls. Just focus on what makes this channel great like community and a love for computers/programming/science in general :D
Not really, because your house would take up less than a hundredth of a pixel and that doesn't really qualify as seeing. More like, you can see the area around my house where me and millions of other people live.
Considering the sample data consisted of people of different genders, from different perspectives and unique lighting conditions, the results were actually incredibly impressive.
@@philmikehunt7479 pho·to·syn·the·sis /ˌfōdōˈsinTHəsəs/Submit noun the process by which green plants and some other organisms use sunlight to synthesize foods from carbon dioxide and water. Photosynthesis in plants generally involves the green pigment chlorophyll and generates oxygen as a byproduct.
Hey so you cant look through all the images and clear the junk. But what about teaching a neural network to detect junk images and only use the good training data?
Wow. This is the the best video I've ever watched on machine learning. No video has taught me as much as this one has. This is extremely interesting, thank you.
Not really actually, what he showed in the video was only really one half of neural nets which is just forward propagation. The calculas comes in during back propagation which is the method used to train neural nets and the main function of calculas in it is the process of gradient descent. I think he explains it in some other videos.
12:42 top right looks like a female, lesbian version of a mix between Cristiano Ronaldo and Robbie Rotten while the one below it looks like melania trunp got hit in the face with a brick
@@official-obama I've developed a lot of knowledge over the past 4 years, and I'm now prepared to answer this question, at least in a rudimentary way. The human brain functions through many patterns and a high amount of interconnectivity, to the point that each neuron has an average of 1000 connections. Many people researching quantum biology, as found in photosynthesis for plants, speculate that neural connections (not just in humans) are room temperature quantum systems. The efficiency of these systems is far beyond what would be reasonably capable of non-quantum systems. This now leads me to the line of thinking that "logical" computers (ones that use simple bits instead of qbits) are incapable of hosting competent human-level AI with any decent scale (you'd potentially need a supercomputer set up just to run a single AI). If you were to design a system similar to the human brain, with proper pattern recognition, spatial reasoning, visual memory, and an emotional framework, you could build a properly intelligent machine, one that understands art and values it as is. This would be a much better alternative than simply trying to make a neural network build itself (as is done with machine learning traditionally) up to a template for whatever *seems* correct.
oh I thought this comment was on the music video nvm it would just be better to have a normal neural network but utilizing qbits instead of normal bit processing (quantum computers are more efficient for non-standard computations (i.e. stuff that isn't physics-based or intensely logical (we haven't developed a unified quantum-gravity theory and using quantum computers for a simple calculator isn't desirable))).
Maybe you should separate the celebrities based on their sex (male/female) to bet better results because males and females have different bone structure and thus have different beauty standards.
Keep describing thinks as if we know nothing, its the greatest to hear the same concept in so many different forms from so many different videos. Not trying to sound sarcastic.
Yeah great, that kind of thinking is why none of the chemistry textbooks I've read bother teaching the Octet rule. Brilliant, take the tools out of the hands of the students. That'll teach `em.
I hope you never use your powers for evil! LOL I am completely amazed by your videos. I’m an electronics engineer, which means I’m a hardware guy. I can write code, but it isn’t very efficient or polished. I understand everything you’re doing, but could never replicate it. I like the style of your videos, as well. Interspersing intellect with humor makes for enjoyable viewing. Thanks and take care!
![Felix "Unfair" | [Stray Kids : SKZ-PLAYER]](http://i.ytimg.com/vi/Oswujxm2Ag0/mqdefault.jpg)
You should train a neural network to detect pronounciation and use it to animate your characters' mouths.
Good luck for the training, you'll have to provide the lips movement for all your samples!
Well, he already has drawn most of the stages he'll need, just has to have the network listen to the audio and choose the appropriate visual from the list. Animating/transitioning them would be a separate matter
that's actually not a bad idea
I don't really know about neural networks, but I think that you'll need a lot more samples than just what's in the video, and it would probably be simpler to hard-code such a thing, but great idea!
yiss
13:03. Man do I love the friend at the bottom right.
Looks like a bird (like a Raven or crow) but with a human face
*Idfk wut a shet is dis*
That's the demon that ate my soul
Amnesia monster?
He’s my favorite man
That picture in the bottom right just can't decide what its background should be, oh my god
She’s tripping
What's worse is the one near the upper left that can't decide what kind of abomination to be.
I FOUND YOU MULTI!!!
12:19
Everybody: looks *”decent”*
Bottom left corner: HI GUYS
Hey what about 2nd from top 2nd from left
Lol it looks like thomas jefferson
I love how by 5 hours of training most of them are okay and then the bottom right is just unable to decide on a face shape
Some of them look like they're being taken by the abyss from Dark Souls lmao
it literally terrifies me whenever i look at the bottom right 💀
This reminds me how i watched Vsause: Understand 70%, pretend you know the rest. Great video though :)
AwsomeNess Studios *_VSAUSE_*
IDK Michael here
*_WHERE ARE YOUR FINGERS?_*
If it is only 30% you don't know, you should ask questions to get to understand the rest.
AwsomeNess Studios Wait a minute Michael is Cary’s brother Dun Dun Dunnnnnnn
3:40 donut-testing in sync with the music! Nice!
shelvacu it made me happy when I saw that!
shelvacu yes
I... actually didn't even notice that...
...how...
Sky's sketches
yeah, it was satisfying to watch. just imagine what it would be like if he put is just a little bit slower of faster, basically torture xd
Wouldn't it be more effective to use a database with passport photos? The photos you used have a lot of different angles and expressions, whilst passport photos almost always have the same angles and expressions. I think it would improve your friends' faces quite a bit.
Tortillian this is very true^^^^
Tortillian You could also combine biometric photographies with non-biometric yet frontal photographies of faces so the neural network first learns about how a face looks like in general and then learns about the individual expressions.
Would like to see this indeed.
True - but if the goal is to get the most "beautiful" people, using celebraties does make more sense than using a passport photo database.
They don't need to be celebrities, though. I've seen plenty of people that aren't known that still pass the standards of beautiful. Also, let's say the A.I. took an ear from a picture that is at an angle, and the rest of the face from front-facing pictures, I doubt that generation would be beautiful.
Why do you show us your "own mediocre work"?
Because making learning entertaining is really the hard part. Now I am interested in a subject I never would have thought about before.
Am i really the first one who reacts here?
I mean, after 1 year and 160 likes... WHAT?
@@headphoneusersbeware4202 yes, hi
29”&83 is your new nickname
@@headphoneusersbeware4202 4th
@@MAX-uh6kh yea... lol
You should make a neural network write one of your videos, make the voiceover and generate images to show at various parts.
I dunno, I just think that'd be really awesome and creepy at the same time.
11:12 Grade trying to fuck up the image generator. XD
Gert Kardol I also just noticed that
THAT NEEDS TO BE CENSORED!
+JRA Cubing and Gaming?
lol, nice catch
Gert Kardol i'm going to say that thanks
Cary is a beautiful people
We are ALL beautiful people on this blessed day :)
Jennings AsYetUntitled ok
Jennings AsYetUntitled So Gradeaundera is a Beautiful People either
I am beautiful pineapple
Hello to the efficiency champion
yo thanks for making these, your stuff is probably the only way id be able to sit through hearing how neural networks work and youve probably opened a lot of eyes to how cool this stuff is getting
10:23 Is this what Hadvar saw while I made my Skyrim character?
@Daniel Jakob ruclips.net/video/3O_A05PDc-M/видео.html
@@excinerus stfu
-@
This video is amazing! I loved the end result, and really can’t wait to watch more of your content. You’ve got youself a new subscriber
It said that this video was recommended to me...
So true... computers are getting smarter every day...
Minecraft Miner no every hour
Minecraft Miner a computer isn't smart.
Mecrom smarter.
Ruined ur 100 likes
hello it's me from 6 years later.
Your animations are amazing, always getting better, love the humor
6:23 - Nope Edges of edges of edges detects the most edgy memes
tru
Jan Pokorný Panie Janie, czy pan jest z Czech?
Why do you have name like from slovakia
Jan Pokorný very suspicious why do he have name like Slovakia
__spencer__ it's more like from Czech Republic. believe me, am a Czech guy
thank you for moving things in-time with the music. It really helps me get into stuff(:
God, you can tell how hard he works on editing and animating this videos that it almost hurts, great job dude!
15:13 That thing you said (the tutorial from scratch thingy) that would be awesooooooooooooooooooooooome .. please do it .. pleeeeeeeeeeeeeeease
I second this
please do.....just do ittttttt!!
hunar omer I agree
hunar omer Yessiree, that would be good wouldn't it?
hunar omer yes
Ya that sound great
But first, let's talk about parallel universes.
Epic Ass Channel lmao
Epic Ass Channel
Somebody call Jack Sullivan.
But first, let me take a selfie
An A press is an A press. You can't say it's only a half.
oh ok
There are several ways that you could extend and improve this neural network to make it more effective, due to the differences between gender, age, and skin color. In fact, you could filter the images with another neural network, inputing only people who have a similar skin color and gender, although age may be harder to factor in.
Thanks so much for sharing your work. I have a good time watching your videos
did anyone notice grade a under a at 11:12
bob 123 W00T
yes bro
Did you catch timtom?
me
-.-- . ... .-.-.-
Morse Code.
You should make a neural network figure out how to play ballz and see how far it gets!
Take [0] it will figure out the cheat...
were is this video at 13:30 from ? im feeling forced to rebuild it O_O
SSpKig its a five intersecting tetahedron
well i have a tetrahedramonopentagonodaenossquareon
How did he built it? SEND HELP
IrNitro D just search 5 intersecting tetrahedra origami
Thank you guys so much, I have been looking for this for so long
woah! you're going to show us how to program something like this? yoylecake! That's great!
Henry Smith your lapis not bubble lol I bet you eat yoylecake every day and you probably give gluten a red food Color and put on top and then you put fries (after he grows his growtatos) then inside you take a page from book and put it in the yoylecake
I was going to make a comment about you being like the announcer (since your name is metal metty and your picture is Mettaton) but then I got the image in my head of Mettaton as the announcer and found that very funny ...or maybe you just ate a yoyleberry. Either way, hugely looking forward to the next machine learning video (man I love this community!)
Henry Smith Hai Lapis Lazuli
Memester Mc Memes go back to filthy frank's channel, you trashy angsty ball of hormones
To anyone replying to that guy, It's best that we ignore the trolls. Just focus on what makes this channel great like community and a love for computers/programming/science in general :D
I like how the music synced with the training data changing :D
You can technically see my house at 8:35. How crazy is that?! :O
And you could actually see my house at 14:32 well, old house.
Im too far up :’(
Not really, because your house would take up less than a hundredth of a pixel and that doesn't really qualify as seeing. More like, you can see the area around my house where me and millions of other people live.
@@ObjectsInMotion More like, you can see the average color of that area actually.
Oh shit! You've said too much private information!
Considering the sample data consisted of people of different genders, from different perspectives and unique lighting conditions, the results were actually incredibly impressive.
Lmao when the filter found the first donut, my alarm went off and I thought it was from the video.
!!WARNING!! DONUT FOUND!!!!!! WATCH OUT
11:12 Lol the only picture they could find...
(Also MattShea is at the bottom 3 seconds later)
clayton8or_2 how did you see matt
tfl gamer I looked with my eyes
James Murray too
clayton8or_2 saw him too
I only found GradeAUnderA and AsKaGangsta
Could you call this photosynthesis?
Socrates That pun kills me slowly
no
@24 hour subway gang But it is synthesizing a photo
@@philmikehunt7479
pho·to·syn·the·sis
/ˌfōdōˈsinTHəsəs/Submit
noun
the process by which green plants and some other organisms use sunlight to synthesize foods from carbon dioxide and water. Photosynthesis in plants generally involves the green pigment chlorophyll and generates oxygen as a byproduct.
@24 hour subway gang The joke went over your head
Wow. I understand how this works now! I never saw a video that explains it as clearly as this. Great job!
at 12:01 the bottom left corner of the generated people looks like a smudged Hitler
Regisle DragonRider LOL
This comment made my day :DD
lol
lol
thank god it wasnt just me that thought that
This guy can make evolution and machine learning programs
I just made tic tac toe
Aiden Ocelot
I made a boo-boo on my tummy
I wanna die
I just made awful animations in Plotagon.
I just make CO2 waste
I can just print hello world lmao
Hey, we've all got to start somewhere! If it's something you're passionate about learning, it doesn't matter where you start from.
7:00 *Eye* detector: High probability of a *nose* being here.
This is extremely informative !!
Cary's content is always so interesting and he himself is inspiring, keep doing what you're interested in, Cary!! 🤩🤩
So nice to see you pass time making other beautiful things while waiting for your code to compile =)
Hey so you cant look through all the images and clear the junk. But what about teaching a neural network to detect junk images and only use the good training data?
David McDermott Kalman filter my dude
LOL
Nice gradeAUnderA at 11:10
Conga 11:11
Conga Also Tim tom
i saw that two i thought i was the only one
Mattshea caught my eye
Messed up the last pic lel
"Wtf why are you standing I thought I broke your legs a minute ago"
😂
This is low key awesome. Like it reminds me of matpat theory videos and how they are teaching me new stuff while also being enjoyable to watch.
*"hey guys these are all my friends but for kinda ugly"*
**SCROLLS UP**
**
**ILLUMINATII THEME PLAYS**
i have just done the monster celebrities illusion looking at the edge of tge "friends"
If you pause it right on 1:42, that means it’s saying white is black-
AI and neural networks have come so far in just 6 years 😮
Yeah, they have, it's scary
Wow. This is the the best video I've ever watched on machine learning. No video has taught me as much as this one has. This is extremely interesting, thank you.
This.
Is amazing.
Thank you for creating this.
It's beautiful.
What was he making at 14:08, and how do you make the thing.
multitetrahedron.
Training my computer to become Picasso
12:38 has the only celebrity on there I know (fourth image from left to right) it is Todd Howard, CEO of Bethesda
Haha dude i love your work and your sense of humor. Good going !
This is actually really impressive...
Good job, this is well over my head.
"and calculus and math"
HA, I knew the filter on filter thing sounded like differentials!
Not really actually, what he showed in the video was only really one half of neural nets which is just forward propagation. The calculas comes in during back propagation which is the method used to train neural nets and the main function of calculas in it is the process of gradient descent. I think he explains it in some other videos.
What does he start building at 13:04
Caleb Intal its an tethahedron
Caleb Intal a mathematically correct 3D star
Alguien entiende aca español ? por favor necesito ayuda !!
puedo ayudarte
2:03 lmao i exist on the spectrum as well
Hypergan's training looked like the kind of faces you'd see in a dream, really coll
At 8:09, there is a spongy on the shelves
It was fun watching the training data for people I know. 12:35 there is hunter avalone.
It sure was fun
11:12 grade a under a showed up xD. I dont think you want cartoon characters for your training data.
*COUGH COUGH* TimTom *COUGH COUGH*
Cabalex tim tom?
marshmallow guy in green shirt waving
Sweg Master lel it's the guy at 13:25 we were talking about
brende the dev it's the best characters you can make with mspaint
The output for the training process is straight out of a horror movie
Great content
Thanks for making this more visible to the public. It's so cool
12:38
2nd image down on the right side.
It's him.....
He's everywhere.....
Who?
Justin Young?
TODD
11:27 why's there a random lizard in the training data, third one down from the top first row.
You mean zucky?
WELCOME BACK MY OVER INTELLIGENT FREIND
What is a freind?
Rolando Cueva A dyslexic friend
SU Fan Hi ruby
HELLO HELLO *tips bandana* good day to you my sir
You are the one who has introduced me to this, thank you very much, your golden
Nice crankbrothers T-Shirt! And nice video, thank you fo making it understandable for unexperienced people
13:25 Is that TimTom? Lol I know him
I saw it too
AHEMMMM
Me 2
Keetz Who else would it be?
Ricegum
12:42 top right looks like a female, lesbian version of a mix between Cristiano Ronaldo and Robbie Rotten while the one below it looks like melania trunp got hit in the face with a brick
funny thing is, I've made this before but I never thought it was considered AI, or even suffisticated. Also, I was 14 and did not have any freinds.
Perfect and simplest explanation. Good work my friend.
This is the greatest ml101 video ever.... Lucky I found out your Chanel.
Imagine doing this this using a quantum processor
This this
But how could you make it do it?
@@official-obama I've developed a lot of knowledge over the past 4 years, and I'm now prepared to answer this question, at least in a rudimentary way. The human brain functions through many patterns and a high amount of interconnectivity, to the point that each neuron has an average of 1000 connections. Many people researching quantum biology, as found in photosynthesis for plants, speculate that neural connections (not just in humans) are room temperature quantum systems. The efficiency of these systems is far beyond what would be reasonably capable of non-quantum systems.
This now leads me to the line of thinking that "logical" computers (ones that use simple bits instead of qbits) are incapable of hosting competent human-level AI with any decent scale (you'd potentially need a supercomputer set up just to run a single AI). If you were to design a system similar to the human brain, with proper pattern recognition, spatial reasoning, visual memory, and an emotional framework, you could build a properly intelligent machine, one that understands art and values it as is. This would be a much better alternative than simply trying to make a neural network build itself (as is done with machine learning traditionally) up to a template for whatever *seems* correct.
oh I thought this comment was on the music video nvm it would just be better to have a normal neural network but utilizing qbits instead of normal bit processing (quantum computers are more efficient for non-standard computations (i.e. stuff that isn't physics-based or intensely logical (we haven't developed a unified quantum-gravity theory and using quantum computers for a simple calculator isn't desirable))).
5:14 on the right the image looks like a face. Was this intentional?
very uncanny during generation
Awesome video, great music! Learned so much. Keep it coming.
Holy shit. That hypergan people creation process looked oddly satisfying... also... you made people who dont exist... like damn....
Wait a sec! Is that the long-lost yoyle city in the background at 1:52?
Yes does make sense because he’s the guy who made BFD I
It's TIM!
thought his name was Tom
TimTom!
Munky Draws lol poped out from nowhere
13:35
Can anyone tell me what he's doing?
A star?
I love how you included grade A under A's Avatar in your training data.
this is such elaborate presentation, good job!!
I use SNAP! to randomize google images
Maybe you should separate the celebrities based on their sex (male/female) to bet better results because males and females have different bone structure and thus have different beauty standards.
He made a face generator and it worked and there was a slider that changed gender AND makeup
Poor TimTom
IJT's Scrap Pile poor gradeaundera
10:28 Mumbo Jumbo was the last person I expected to see there.
Wow. This taught me more about neural networks than videos actually trying to teach me about them.
6:12 are you trying to get the cubers to notice you? Corners and edges? Also we (the cubers) saw your 3x3 during the paper thing time lapse lol
AWSMcube what are the cubers
Pancake Comet we are people who love solving 3d puzzles. These are including but not limited to the classic 3x3x3 rubik's cube.
i was thinking the same :D
Did you see the megaminx
9:30 challenge accepted
ok i give up
i meant the programming
Its been 4 days. How far have you gotten?
13:25 izz timtom
you are skilled at simplifying everything complicated
i think this is the best explanation of neural networks ive heard tbh
So this is all about donuts?
2:38
**Ruby intensifies**
Keep describing thinks as if we know nothing, its the greatest to hear the same concept in so many different forms from so many different videos. Not trying to sound sarcastic.
Yeah great, that kind of thinking is why none of the chemistry textbooks I've read bother teaching the Octet rule. Brilliant, take the tools out of the hands of the students. That'll teach `em.
The highest donut score would be 6.8. The lowest one would be -4.5.
I hope you never use your powers for evil! LOL I am completely amazed by your videos. I’m an electronics engineer, which means I’m a hardware guy. I can write code, but it isn’t very efficient or polished. I understand everything you’re doing, but could never replicate it. I like the style of your videos, as well. Interspersing intellect with humor makes for enjoyable viewing. Thanks and take care!
Ever since i've found your videos I've stopped using pornhub.... Just like to let you know.
james senor ye, those are some sexy computers XD
james senor I
james senor i'm guessing you havent watched kurzgesagt then
james senor WHAT THE FUCK IS THIS SHIT HERE HAVE YOU BEEN MADTURBATING JAMES?