AI Learns To Walk | Asmongold Reacts

[@aiwarehouse]

It's so cool to see Asmon react to Albert:D When AI takes over, he'll be spared

[@siliconhawk]

its albert

[@Gengh13]

Good job on the video man, I for one will gladly accept my Ai overlords😂.

[@greasybrownie]

Hail AI overlord Albert o7

[@viice_]

I've seen similar videos, but yours is really well done 😊👍🏼 Code Bullet is also really great, because i love his dry humor, but his programming seems to be giga messy 😂 Would be super cool to see a video at the end of Alberts journey, that explains how all of this is done.

[@seribeeri6759]

love your vid man :D

[@brianestrada1993]

This Ai learns faster than a video game journalist.

[@bobbobber4810]

This is saying the game journalist could learn.

[@MilesHoppus]

That's assuming the journalists aim is to learn

[@mcbain1131]

To be fair water bears learn faster then video game journalists

[@digduck9463]

​@@mcbain1131 Even a pigeon does.

[@ekcman]

Lol AI is like 2000 gaming journo playing together and finally the winning one can write the article

[@shaicat]

To the person who asked why have Albert learn to walk instead of creating him already knowing how: the point is to teach the AI *how to learn*

[@Draco_Zakai]

I forget where but there's someone developing an AI, starting as a baby. His idea is that it will give them more "humanity" and respect for life

[@JusttAlf]

​@@Draco_Zakaiwhat a dumb hypothesis, as if all humans didn't start as baby and grew up to not give a crap about life in general

[@ricardocoleman2326]

If you have to code the knowledge, all you have is a really good script. I think the power of ai is that it will be able to do things we didn't explicitly teach it to 🤔

[@SkyMina_]

The point is to have AI use the knowledge that are given and fill the gap to the missing info, their role is to figure it out just like us. This is especially similar with habits , what are the chance that the AI will figure it out "This is the most efficient and correct way".

[@Thedarkbunnyrabbit]

@@klzeccwozi1290 The reason why a crocodile cannot be raised as a human is because its brain is fundamentally different and more limited than a human, and it is programmed extremely differently than a human. AI can be programmed any way you want it to be. It can never be sentient, but it can mimic sentience enough to fool anyone.

[@IcemanCT]

The persistence. The perseverance. We all became Albert fans this day.

[@AXharoth]

youre just projecting

[@ryucartel351]

I'm not impressed. A child learning to walk is absolutely incredible, but AI isn't anything, it's nothing, and so it poorly replicating something that's actually real is just boring.

[@vergillives9890]

​@@ryucartel351until the death lazers

[@wkkqewqaver7766]

​@@ryucartel351 why is a child learning to walk invigorating but ai learning to walk is go to sleep boring.

[@Redgon82]

AI dosen't have feelings only goals and sooner or later any AI will fullfill it's goal. We human fail because of frustration and other feelings.

[@WisecrackJax]

"Now that you can walk, there's a whole new world of [humans to destroy.]"

[@2WarriorJay8]

I trust well-programmed AI more than a lot of real humans. :)

[@Arkovin]

@@2WarriorJay8 Considering how apparently 50% of the population is worse than a bear i guess we all can agree on that..

[@4IndieForever]

Wait a minute.. What!

[@metalmask5]

@@2WarriorJay8 This is just my opinion, which I have thought partially, so it may have some flaws and undercooked in some parts, just to be clear: AI is designed to understand emotions in a described way and then adapt itself to them later, but it doesn't feel them "raw," so it will always prioritize logic over emotion, at least initially. Its moral compass is programmed, not developed by natural means. If that were the case, we would have a super-efficient AI that is opposite to humans and may overshadow humanity's way of thinking. Mostly because what makes us human is that, just like AI, we make rational decisions, but we do so in order to accomplish irrational goals, most of the time, and morals intervene in what we usually do, unlike AI, at least in the fairly primitive state it is in at the moment, until it somehow develops further into a future, not a fiction movie-like one, but a realistic one, where humans actually let AI decide for their lives, maybe not in big decisions, but in day-to-day matters, like what we should eat, what we should study, and so on; we would become dependent on them to an extent, better said. I don't deny the fact that AI could actually be programmed to feel emotions the same way we do; after all, if we think of the human brain, it's like a computer, very complex and incredibly designed, but actually organic and unique; so we should actually master ourselves and understand how to replicate the limbic system of the brain, which is responsible for emotions, focus, urges (including sexuality), personality, and behavior, but most importantly, instinct. So if we can replicate that, we may be able to integrate a similar system into AI, but to reach that point, we will need a lot, lots of time. So it's not far-fetched, but it's not something that is happening anytime soon; however, it would cause other kinds of problems, such as morality, just to mention one. Also, why would we need people like engineers, mathematicians, and so on if we have an AI that can actually think outside of the rationality box and create even better things than us? If the automation nowadays is something that is disliked by some people who have jobs such as artistic, musical, animation, and similar ones, would people allow AI to reach even further heights? It would rivalize with some humans, so again, they have to be programmed to achieve such a feature, because no matter what, AI is still AI at the end of the day.

[@jf9387]

@@metalmask5I think human intrigue and economic demand will definitely create AI that can replace any human in any job. One day the economic value of a human will be zero, and we’ll have a lot of thinking to do about how society is going to work. Perhaps AI can help us find a solution to that one as well lmao!

[@agoosecalledxaro6679]

How has this man never seen code bullet?

[@Howabouthere]

Fr, he should watch a few of Code Bullets videos, It's really funny 😂

[@slamdunq3192]

Bro i legit couldnt remember that channels name for like 2.5 years i tried everything thanks so much brother 🙏

[@-Wave-ol]

Asmongold needs to watch code bullet.

[@lucasfink876]

I'm actually a more impressed with ChatGPT machine learning process. That's some very basic AI.

[@securatyyy]

Code bullet is an amazing dude. Pretty crazy asmon never heard of him.

[@kanosig]

No joke though, Asmon needs to watch one of those Trackmania AI videos. The guy really breaks down game physics and reinforcement learning and does it in an understandable/entertaining way. I've never even played the game and it was fascinating.

[@JamilaJibril-e8h]

Would they be blue or pink just don't make them intersex We are already dying here from figuring out the difference

[@lilithvia1122]

Don't play trackmania. You'll get banned from discords for mentioning one of the top players actually uses hacks

[@erossenpai2884]

AI: *learns and becomes more powerful* Hairless ape: "haw haw look at the stoopid ai its so dumb" *learns nothing and remains the same* Somehow, im not surprised humans got their shit rocked by the machines in terminator.

[@Mahlak_Mriuani_Anatman]

Skynet supremacy

[@Syphirioth]

The fact we can teach computers the same way we teach dogs and kids is pretty amazing but some do not realize it very well.

[@ahkao.9347]

im not surprised either that the humans got their shit rocked in a completely fictional story.

[@Syphirioth]

@@ahkao.9347 I love the argument about LLM is just a prediction tool... Like we use magic when we walk or drive??

[@wobbles86]

So this is what happens in our 1-3 year old brains that we all forget about

[@grants7390]

maybe the reason we don't remember is because we were spending all energy and mental faculties doing this instead of remembering comparatively unimportant details. /not serious

[@JamilaJibril-e8h]

​@@grants7390 huh !?....

[@markpiper6382]

@@JamilaJibril-e8h Think of it this way. Imagine a cardboard box, with all your memories stored inside. While we're young, our brain can store all of our memories in the box, as there hasn't been a lot of them accumulated yet. However, as time goes on, the box gets filled up, and our brain has to start throwing some memories out, choosing which ones are the least important and replacing them with memories it decides are more important. Since the box can never get bigger, more and more of our old memories get thrown away, and eventually only the extremely important ones remain. The selection process varies from person to person as well. Eventually, as the box gets really old, it might rip, and break open, causing a lot of our memories to start falling out. Obviously it's a lot more complicated, and I left some stuff out to keep this relatively concise, but I hope that this was a helpful analogy.

[@grants7390]

​@@JamilaJibril-e8h i meant that, there is only so much you can focus on and put energy into at a time, and it's all going towards low level stuff like learning to walk and talk with none to spare.

[@JamilaJibril-e8h]

@@grants7390 whatever.... Easy to say hard to execute aka let's admit the limits

[@Noizzed]

People forget you are seeing something go from not being able to move, to walking perfectly fine in a few real time hours. It takes human babies a few months to even stand up and only for a short period of time.

[@whamcogames]

It's quite child-like. Creepy-cool for sure

[@LikeAFemaleDog]

@0:40 Not necessarily, AI Art has an deep-seated problem regarding it's method of operation, more particularly how Art (generally) requires time to flourish while AI art wants to be fast, or how it's progress is dependent on the jobs it's trying to replace, and it's already having trouble with "limited data." AI can flourish, but even if it does, due to how much art evolves by the minute it's possible that relying on AI can end up just resulting on us limiting ourselves just for the sake of having more quantity over quality.

[@l33tninja1]

We basicly already run things that way now. Companies focus on quantity over quality all the time because it gets them more money in the long run just like quality is better for the customer in the long run.

[@VisArtVisuals]

What makes this far more interesting that AI art is that it isn't stealing our legs to copy them and sell them back to us. I like this, but i don't think I'd be as impressed if I had to give up my legs tbh

[@oolavitzoo]

Looks like Asmongold when he's playing dark souls and throwing a tantrum

[@Incognitiv]

I mean, a very similar thing was (and is) available in Unreal Engine as well, where you could train artificial intelligence to make your enemies smarter. It's called "Learning Agents", which - potentially - could make the npcs way less predictable, but I'm not sure if any game used that yet.

[@pixelpuppy]

This AI learns faster than a physician.

[@themaxterz0169]

I dont understand the comparison 😅

[@pixelpuppy]

@@themaxterz0169 it was a meme from a while ago when Asmon interviewed the mod of that girl that did the blindfold Monster Hunter run. The guy defended her by lying and saying he's a physician.

[@bakedandbeaded]

@@themaxterz0169 To add on, he said that when you’re stressed, you tend to sit up higher and be more opened versus shriveling down if you’re stressed, you know? (The cheating girl was sitting at a weird angle which allowed her to see the screen from behind her blindfold) Which is 100% bullshit because he said it so matter of fact about EVERYONE, then went on to say he’s a physician, he knows what he’s talking about. It was one of her discord mods who also claimed to be her husband and claimed he saw her do the blind run IRL multiple times… All lies all around of course. Said she’d be willing to do the blind run for a third time and then she deleted everything and disappeared lmfao

[@JamilaJibril-e8h]

​@@pixelpuppy look failing is hard we all fail but to push on rubbish is devastating actions.....like they are jumping walls ...

[@armandpeanutspinou3430]

@@bakedandbeaded Wait he claimed to be her husband? idr that

[@ClearlyNotBuer]

Albert is already smarter than me.

[@kierinhernandez7524]

The AI in this video is a mirror to evolution. The conditions change, the strategy changes, the ai adapts. Worm -> Skipping ->Gallop-> Shuffle->Walking

[@wobbles86]

This is like a retail casual trying to get out of the green fire on the green floor in the green room of the green zone

[@borzydar1196]

4:15 the algorithm for walking is too complex to just write it down. Artificial neural net consists of input neurons (here probably position and orientation of Albert, position of buttons, position of obstacles etc.) multiple hidden layers of neurons (black box) and output neurons (each limb or joint). All of the neurons are interlinked with asinged random value for all connections. Neuron can either activate or not depending on link value threshold. Learning here is about adjusting the values (weights) between connected neurons. To direct development of the neural net it is given a fitness function - a scoring system that the AI is programmed to improve.

[@iAmDiBBz]

3:37 i unironically associate his movements on the floor after hitting an obstacle as the "kid with green shirt on his head rolling around on the floor while other guy looks on in disgust"

[@Peter-wj2hz]

Now let's train an AI by rewarding it whenever it successfully trains an AI. What could go wrong?

[@raphaelst-hilaire3669]

they would train themself for sure

[@ProblmSolvd]

This is similar to how they teach robots to walk, the only difference is the time between resets. Watch the time lapse of Boston Dynamics robots and you'll see all of the same "Ahah!" moments.

[@joshuadehler5039]

Dude I could FLY through Walmart skipping as a kid. Felt like I beat gravity

[@BIOSHOCKFOXX]

This is how Skynet starts. By the way, I could bet that AI thinks this sort of body is inefficient. It could learn much better with more normal anatomy type of body than this cube dude where you need the torso has to be held because it's one of the parts that doesn't have a control set programmed, only legs. That's why it skips or limbs with one leg rather than using two, because the weight of upper body just falls in a free fall. It's like trying to hold a stick upwards on your palm with a dish on top of it, like in circus.

[@Sleepy_Cabbage]

Theres alot of simulatgrs like this where the humanoids most efficient mode of movement is just leaping around like a frog lol

[@nickwells20]

It shouldn't be too unsettling considering that AI was made in mans image. So hopping around like a kid makes sense to a degree.

[@plfaproductions]

the scary part are the implications

[@darthseagraves]

Made in God's image

[@casualfungaming4356]

@@darthseagraves did "God" made us trying to replicate what made Him as we create "Them" and a cycle of Creators creating Creations start to unfold O_o

[@LexisVoyage]

@@darthseagravesi mean yeah but he also was a burning bush one time to scare the shit out of moses because it was funny

[@Evan-k]

​@@darthseagraveslol god

[@leroycrosby9987]

This tech for NPCs inside of VR games, or in Cortana-esque assistants, is going to be awesome

[@Leongon]

I kinda knew what to expect but when it started using the arms for balance I still got my mind blown. Holy shit.

[@Bababoboboa-gu3iu]

Well, yeah, this has already been possible for like 10 years.

[@hijjak97]

Nah I've been walking for longer than that

[@tamix9]

And it's barely gotten any better. There's been a few new techniques, but the vast majority of the improvement in the last few years is due to increasing dataset size and training time. And big tech companies are running out of data already. The idea that AI tech is comparable to early computing is ridiculous. It uses the same hardware, the same software, the same maths as anything else, and none of that is improving much. There is no breakthrough in sight.

[@ManySeptims]

@@Glenners People who know about ai also know the exponential potential of ai, especially in the upcoming years, and understand that one of the next steps after transformers and gpt's is going to be something a lot more powerful. With multimodal ai and the potential approach in the next 5-10 years towards AGI, and the projected near-future use of quantum compute and nuclear fusion generators.

[@ManySeptims]

@@Glenners lol reinforcement learning though yeah it's been around for a long time, nothing new about that. It's cool seeing homebrew RLMs at work though, which is what makes videos like these interesting.

[@Yottenburgen]

​@@Glenners You're extremely naive if you think there aren't plenty of people who see lesser stable diffusion models that can't do hands or text and still think that AI can't do hands or text at all. They probably still won't know in 2 years because they aren't interested anymore. There's also plenty of those who pay attention to AI and still don't even understand the basics of how it works or of its current capabilities. Also why are you so angry about that comment. It's clearly a joke that people are shortsighted with short attention span and dislike things on a whim.

[@Truebro79]

we literally witnessed a baby learn to walk from like new born to 2 years old but in just 8 minutes.

[@gingeral253]

Machine learning existed way longer before this. There just was never the techniques that optimized it and increased the scope. Code Bullet made many of these years ago.

[@lordomacron3719]

Now Asmondgold needs to watch Code Bullet and his AI ‘learning’ to walk.

[@askel6498]

I for one welcome our new A.I. overlord!

[@lLenn2]

This guy doesn't even realize that they're gradually introducing biases on how to walk so his argument that they do this to find a better way of walking is completely negated. He's right about it though, but this is not the video to show it.

[@DurzoHighwind]

True, but I have no idea if regular machine learning isn't trained on biases and rewards like this anyway. Maybe there should have been just one directive on reach point A the fastest way possible and give him some limitations like don't break bones etc while giving him a human skeleton.

[@lLenn2]

@@DurzoHighwind It is, by introducing constraints it learns a lot faster which is important, but you'll only get specialized AI that are good at one task. Still useful, but not the scare that AI is hyped up to be.

[@dominikpecuch2177]

@@lLenn2 If the constraints introduced try to mimick real life, it might find a better way to walk. Like you can tell right now, the most efficient way to walk wouldn't be by your chest touching the ground, right? That doesn't mean it won't learn anything new. Although I doubt it, evolution is pretty much AI, but in span of million years, so humans are already walking as efficient as possible, imo.

[@firstsomeonelastname42]

@@dominikpecuch2177 Lol no the hell we are not as efficient as possible.

[@dominikpecuch2177]

@@firstsomeonelastname42 since the goal is to expend as little energy as possible, of course humans are efficient...we are probably the best endurance creature on the planet, can out-endure a horse while running, thanks to bipedalism which makes breathing independent from walking/running... unlike 4 legged animals who have breathing and movement coupled

[@Blackywulf37]

imagine being reincarnated into being Albert

[@xSergisX]

we're making ourselves obsolete and we're enjoying it

[@RamboTheArtist]

Asmon clearly sees no difference between people stealing artworks and feeding it to AI to make weird mashup of stolen works and people making AI figure it out by it self. Making AI walk was never an issue, making it look good, yes, because AI will use the best way to reach the target, not the most "natural", but anyway. People never complained about self teaching AI, its not even a new concept... you can google AI walk and Im sure there will be plenty of videos that pre date the whole AI drama... However people do complain about art Theft and its really sad that you don't see a difference between the two...

[@Nobody-vr5nl]

Ai video games have been YT content for like 8 years. My fav is that it would glitch its self into a more efficient walk.

[@idpro83]

This AI went from crawling like a worm to skipping to first baby steps to walking like a drunk person to walking on the moon.

[@PrimeNPC]

5:43 this segment will lead to our enslavement by the machines

[@BobDude65]

The gravity was set way too low. This was like watching me trying to run in my dreams.

[@Hunoa44]

AI is generally an interesting topic and it gets better once you see how it can play games or break tf out of the physics engines

[@huntergrant2011]

Now give 2 of them swords and make them fight

[@josephnolan6323]

Would be neat if they create a bunch of these and see which ones end up learning that fastest overall

[@AymenDZA]

People don't know the difference between AI, machine learning and procedural generation !

[@bennyboiii1196]

For reference, this tech is what allows the Boston Dynamics dog to walk better in more adverse environments.

[@FireHam]

no, this is simply bruteforcing

[@umrete]

don't make fun of our A.I. bro here ... I've watched the animatrix ... I know what's up..

[@stepankotyk8823]

comments about AI with quantum computing made me spill my coffee :D But is showingf why AI hype is sooo huge right now, as most people dont know how it works

[@mullive]

Notice that Albert learned everything by himself only guided by the conditions of each stage. I bet that if Albert had another fully walking cube to watch and copy (like any baby would) at second stage he would be already walking.

[@JohnDoe-ug3su]

Its incredibly unsettling how similar the steps are with how a baby learning to walk.

[@grenvallion]

He couldnt even stand up at the start. Its incredible to see how amazing ai is.

[@АртёмОльховой-ф9е]

“The next lesson, Albert, is how to hold and aim ak-47 properly.”

[@ricardomilos3872]

That AI did what took us between 10 to 18 months of our existence to do in 8 only minutes 💀

[@HiranoMagu]

Nice to see Asmongold is still salty with the AI art argument.

[@folonrng]

ai will never be a threat. we'll just introduce ai cancer.

[@tinypixiebread]

Rare footage of early development of the Automaton

[@mina_loi]

there used to be a channel on here that streamed an AI trying to play super mario bros on nes, it took the AI 2 god damn years to figure out how to jump over the first pipe on lvl 1-1.

[@Ramsey276one]

Older computer, right?

[@Ekstrax]

It is a more effecient way of moving! my first thought too when i saw this! also it's not just when you're a kid, skipping is a little more energy efficient for adults as well

[@astrovation3281]

A good one would be the newly released trackmania AI by Linesight, he's been working on it for 2 years and it's showing real creativity in racing lines now.

[@user-ck9tb4fv2x]

One thing I love of these types of vids is it shows the AI doing what it needs to do. But Walking seems to be challenging and I think A.I's could learn this task faster if the lower anatomy made it easier for him. Right now he moves either leg the whole waist twists to match. Dudes having a hard time because everyone makes the simplle object ( which is fine A.I. is challenging let along modeling) but since most people don't take into the fact they limit the legs with a bar (sorta like taking your pants and putting them lower makes it harder to walk) instead of how legs work usually.

[@theseal126]

What so intresting about this is that the trained AI can you then give to any robot and it will work. Which means you only have to teach an AI once and all will be able to do it. Further they can learn in simulations instead of real life, which means if you speed up the computations of the simulations, they might be able to learn how to do parkour in 1 second which takes humans 10 years to get pro at. Of course we currently have no robot or simulation with such dexterity and such fast computations. But nonetheless its possible. So not only can you have one ai learn one thing at a place in the world and another ai on another place and they can share the trained programs and use it, meaning they can learn multiple things at the same time, but even more as we have seen with large language models that the trained program can actually be used at other places in reality for patterns repeat. The easiest example would be large language models being able to learn new languages on their own or how an AI used to detect shapes in a bakery is now used for detetcting different shapes of cancer. Yes thats real you can look it up. Of course there are limits to AI, but the potential is staggering. Altough I still think humans have more potential for we have the capability of doing things wrong and sinning. Some of the most important scientific discoveries have come from making mistakes like the antibiotics. And I think both creativity is of the same vain. Like Arthur Schopenhauer says: "Talent hits a target no one else can hit; Genius hits a target no one else can see.". Therefore we humans are the true genius and should remain that way. That AI serves us humans instead of humans serving AI. There are far more reasons for this. But from a mythological perspective you could think of AI as the artificial God and we might regard it as such, and Lucifer the highest angel God created fell in love with it self and had knowledge and think he knew what is the best. In the end dooming those tempted to follow him. For us to praise an artificial God as God and also letting it make our decision, might very well be the end of us, it might very well be us placing satan as God. I am an atheist, but still would be typical of humans to think there own creation which is AI is God like and thus place it as such letting it make decisions. Thus we serve AI and our free will disappears, and our ability to sin disappears. Sin is what allows us to have free will. In North Korea you cannot sin, that is how such a society could evolve and it will eventually happen. Humans and the human spirit should always be at the top, and AI should serve humans and give recommendations, never robbing humans of their unique power and talent of free will and genius of sinning. (Gallileo was thought to be a sinner when he proposed the heliocentric system, so it is of the same vain when I talk about sin here. There are bad sins that are unforgivable. And I dont mean those, I just mean thinking new and hitting a target no one else can see)

[@kyouksterm208]

7:00 he's actually walking like a toddler taking its first real steps 💀

[@buzzlightyearpfp7641]

asmons '18 second' analogy is actually great

[@karniejbeats]

People ask why ? After Albert learns to walk, you just upload it to fully working machine robot and you've saved millions of dollars on crushing robot parts.

[@SHADOWSTRIKE1]

Regarding the question of if a Child or an AI learns faster, the difference is scalability. We’re seeing one instance running hundreds of thousands of times to learn. Instead, you can scale thousands of instances which can share data and learn significantly faster. It’s all about how much compute power you want to put into it. For my fellow anime nerds, it’s like when Naruto used multiple shadow clones to train and then gained knowledge from each clone’s experience, rapidly speeding up his training time.

[@Sigrt]

I can't wait for AI speedruns

[@Devon-hw6ih]

I saw a video with the same premise of an ai playing jump king. Very cool video

[@RusticTrophies]

Consciousness is Not a Computation.

[@jason2mate]

Just a note, the reason to have him learn to walk, rather then teach him, is because it's a proof of concept and allows us to improve the methods we use for when we give them something that isn't "as simple as" learning to walk (as an example, being a doctor), if you want a General Intelligence AI, it has to be able to learn things itself and getting these early processes down in a way that they learn quickly is a big thing since we don't have infinite computing power.

[@awesomepsume]

they should program a stage where points are awarded for maintaining an straight back, since all forms of human movement benefit from that besides yoga type, or acrobatic, or dance type movements.

[@ObeyCamp]

It's actually a really good point you make about how kids skip because of efficiency of movement related to their bodymass or proportions thereof, and how Albert here also figured that out. There might actually be something to that. Notice that Albert's body has a large head and short limbs. Those are exactly the proportions human children have (though the extent of Albert's childlike proportions are more akin to a human baby than a human child, but still.) Human children-especially babies-have different proportions from adults, most notably the proportionally larger heads and shorter limbs. So an AI with childlike proportions learns to move first in a manor that human children learn very early on because of their proportions? Maybe it's just a coincidence. Probably. But what if this AI is smarter (or at least more realistic) than we think? Maybe learning to skip is a much more natural development than we think? Very interesting.

[@DragoonPaladin]

Albert just started existing. Getting blasted by flying cubes. It'll remember that one.

[@bj0urne]

Now just do "AI Learns to solve world hunger" and we're golden.

[@WelshJesus]

I didn’t know that about skipping. Makes a lot of sense though. I used to skip everywhere as a kid, it was way faster. Speed run technique

[@LeAFyromster]

this is how AI and robot integration would work. the ai simulates movement from sensors and the robot moves based on AI

[@brettbutler4013]

The trackmania training Ai on pipes video is crazy

[@ohmygodbecky6829]

It looks like a googly eyed cube of cheddar But this thing never needs sleep, gets stressed, traumatized, discouraged, distracted, or depressed. It will learn, keep learning, and if it ever gets far enough, it will be that little spark of eternal learning tenacity uploaded into all robots wiping us out lol

[@williamthatsmyname]

What people don't know, is that Albert can learn to walk, within a few seconds. The reason that the video is long, is that the trainer wanted to show us the progress.

[@CuppaGi]

It's almost as if the AI is learning the same way a human would.

[@Minimax04]

‘What do you call it…my dad’s friend’s house’ That’s such an unusual ‘Whaddya call it?’ moment.

[@Clint_Yeastwood]

When you compare it to how long it takes humans tp learn to walk it really puts into perspective how terrifying this seemingly innocuous thing actually is

[@aero4856]

The problem with ai isn't that it isn't/won't be good at doing tasks. It's that it necessitates stealing from a source.

[@AmonetSVK]

This A.I. is making better progress in learning to walk than 99% of people in raid progression.

[@wyatttilley7849]

I remember side skipping in my house from ages 6-13. It was just the better alternative to running in the house. I also wouldn't stub my toe at all.

[@Designsecrets]

Is it actually ai learning, or just brute forcing all options till it gets the result

[@arizona_anime_fan]

the reason why they teach the AI to do these things is because the goal is to build a general purpose AI without teaching it a set of skills, so it has to learn to think, reasons and do things for itself. An AI which can teach itself to walk can teach itself a lot of things...

[@anyoneanyone28]

It would have learned much faster if the gravity wasn’t set to moon

[@stagnation4626]

I think the reason why people like asmon think AI is smarter than it currently is is due to the fact that us computer scientists use humanizing terms to make it easier to explain algorithms and concepts that people just think its doing what humans are doing when you can literally apply the same logic to every single code ever made.

[@Max-ej4oh]

The AI in a couple of years, capturing humans and calling us Monkeys after they take Frieza's shape. *C'mon MONKEYS, walk! careful the rotating saws on the floor*

[@DanielTGC]

It took twice as long to learn (at least) because the limbs are too short to stand up. Trusting in ai to do everything by its on is not always the answer

[@Sympanet]

Crazy how it went through its “doing the worm” phase before learning to walk just like a real baby

[@quickfingers5979]

Asmon: It’s a more efficient way of moving. Michael McIntyre: Trust me, I’ve got this.

[@miinyoo]

Think about it. The astronauts on the moon skipped and it worked. Mass ratios. Asmon's an observant guy.

[@Vazlist]

Albert is a trooper. I suppose he has no choice but to troop on, but he's still a trooper o7

[@Spacecoke]

Albert learned to walk 100000x faster than bethesda can add an FoV slider. Let that sink in.

[@bobbycrosby9765]

Machine learning has been used for controlling AI in games for a couple decades. It's the only real option where the controls and systems are complex enough such that you can't hard code it. Such as a racing game. Or a physics simulator like this. That's why you teach it. It would be more difficult to hard code the proper responses to all the physics than use machine learning to create a model to handle it for you. Note that this is cool, but all of this would have been possible when I was in college 20 years ago. I had some projects where we had to do very similar things. Evolutionary algorithms were definitely some of the coolest things I learned back then.

[@enny7377]

Rare misfire from asmon to say that people who disprove of AI art are saying that it doesn't "work". Like they're denying reality. It's just ethically fraught with peril is all.

[@sin_fc3822]

A child actually learns far faster than an AI. I remember a study where a toddler can learn and adapt a simple task seeing it a few times where as an AI requires hundreds of thousands of different inputs to get a satisfactory result

[@Darko807]

What Albert really shows is that pain is and will be always the strongest factor in learning. Which -in an abstract way- is something beautiful because AI and Machines will never be able to feel pain or feel emotions as a fundamental given. That is also the reason why this "Terminator will be true" conspiracy isn't bothering me. People actually think the world is that interesting as if it is a blockbuster😂

[@plfaproductions]

wait until you can't undersatand its code and it can do everything that you do better than you, we alredy have friendly fire due to ai understanding things wrong and we being unablle to stop it, imagine when ai is everywhere, 1 mistake and they all turn against us, like ask for world peace and then the robot goes "understood eliminate all lifeforms"