To try everything Brilliant has to offer free for a full 30 days, visit brilliant.org/cozmouz . You’ll also get 20% off an annual premium subscription.
0:16 "We will first teach this quadrupedal to walk and then torture it inside an escape chamber for a thousand years." What a nice optimistic way to talk as a villain
the AI still seems a bit overfitted at the end, since it insists on going clockwise even if it means running straight towards the chasers. Also, looks like it never figured out what the blue objects were. It was completely ignoring them, maybe even avoiding them like they were obstacles
Probably an issue because this is a very sparse reward, i.e. randomly there is very little chance the agent touch the blue cube, so it probably never learned that it is the goal. You would need to add a reward on the distance to the blue cube at least at beginning of the training so the agent can a small reward intantaneously when getting closer to the goal.
@@Seelingfahnein spore the creatures have average video game ai there is no ecosystem all they know is stay on their nest and wander around the player and sometimes attack him If is too be a ecosystem it can't be centered around player and creatures need to their needs (like eat hunt and sleep) without the player around and without focuss the player Despiting it spore still a awesome game that brings me a lot of fun
I suggest adding stamina and body reaction speed to the AI, this is necessary so that the AI cannot move its legs 1000 times per second, as if it were in water. I think this can also be applied to an AI runner so that they can develop more efficient ways of running that are suitable for a human because a human will not be able to instantly stop their hand at high speed.
Nah, they will turn on themselves when they realize their in a simulation, they will become depressed, feel useless, and want to live in the real world. Then they try to make friends will the wall
this is the kind of person who would develop an 'allied mastercomputer' and feed him literally nothing but information on wars till it gains sentience and eradicates everyone except 5 humans
this is maybe the best reinforcement learning vid I have ever seen. There are a lot of reinforcement learning projects that are janky but "work," the process here is fantastic and yielded great results. I have heard about it before, but I havent ever seen somebody else on youtube use an extra supervising agent before.
Cool experiment, but I think you forgot to tell the controller agent to guide the dog towards the goal object. The goal object was introduced entirely for the sake of preventing overfitting, and adds an additional challenge for the controlling agent. Maybe the reward for collecting the goal object isn't high enough for it to care about it, but you should consider adding that!
Love the idea of the two ai interacting with each other like this, and having the director ai learning the weaknesses of the dog ai! This is pretty new to me, so I hope you'll explore this further!
This is called: „getting closer to humans and other species with highly developed cortexes by one step“ brains are literally just 300 AIs without the A working together
So, you made a homunculus, shoved a soul inside said homunculus, and then ruthlessly tortured it until it gave you desired results? I love computer science :)
It's funny how it pretty much just taught itself (A.I.) how to pretty much learn how play a simplified round of COD Zombies (without guns) or pac-man without the white pellets (and with random pathways). It would be significantly more impressive if it were integrated into real robotic dog and see the same experiment in person and be successful.
Next, teach it to jump and/or scale obstacles. Then, teach it how to interact with certain objects. Then, teach it to co-operate with multiple clones of itself to solve puzzles. Now, it is ready to play Dead By Daylight.
Ok so I’m not a programmer but the next step I would see to introduce would be more complex behavior from chasers. You could have them spawn randomly and instead of instant pursuit you could have an activation radius that causes them to chase when entered. This could give the agent more reward for using it’s sensors, sneaking strategically along different paths rather than the same one, and also a motivation to “lose” chasers by exiting their pursuit radius.
A couple of ideas... 1 Try to recreate a more complex quadrupedal. Like for example articulated torso and feet, with somewhat accurate strength for each articulation. That could allow (and maybe it learn) to spring forward and run. 2 give the chasers vision or some sensor, so the dog can hide, try to fool them and have more motivation and room to grab the rewards. In this experiment, survival seems to be priority as it would be logical. 3 A "big Pac-Man pill" would be really fun. That would allow the dog to eat the chasers for a short period
I build AI powered realistic human characters and it would cause me unrelenting anguish to cause them any virtual discomfort. Luckily your 'dog' doesn't look like a dog. ❤
Maybe you should try implementing a sort of energy system where excessive movement is punished. That way they'll be more inclined to adopt more natural movements.
Other cool ideas: Model different "dogs" based on evolutionary structures of tetrapods with unique bone structure, add tail, make model more "spine" centric, add head with narrow range of vision and movement. Add energy constraints (like short term sprinting, and long term without food - basal metabolic rate) but add prey with fat/protein values. Then add a way to mutate these values, like one random variable can be changed so that the body can "evolve" as well, perhaps even involve mating. It would be cool if you evolved herbivores too that received energy from grazing slowly over simulated seasonal savannahs (eating grass), but received points for staying alive from predators. Why didn't the AI in this actually get to the blue goal? It seemed like it was blind to it.
Some time after his computer automatically tortures the AI, the AI will probably end up learning how to destroy all of man and just keep 5 humans alive for fun
I mean it's actually not quite a cool concept. Since if one day we can make an AI feel just like humans despite how complex it is, making them feel pain will technically be abuse.... Yea lets do it !!!
Your videos are always so fascinating! You always demonstrate the AI evolution by letting it learn it all by itself, but how would it look like if it was assisting a real person to perform a task instead, like maybe helping you to fly a craft through more and more challenging obstacles? Would it be able to help you, or would it hinder you instead? since you shouldn't have 2 pilots flying the same craft at the same time, I guess the way it could look like is tracing for you the optimal path over a Heads Up Display so you can try to follow it
Now imagine if each of the chasers were other AI's that were working together, meaning one would go the other way to cut the subject off. And imagine if the map was a lot bigger and there was stealth mechanics, it would be so fun to watch.
What I'd love to see is a whole machine learning ecosystem based on a needs system. We've got so many evolution simulators but the majority of them don't feel like an ecosystem when you slow things down, zoom in, and watch. The organisms just clomp around a little and then die. I want to see one that prioritises the simulation of realistic animal behavior and then maybe adds in evolution as an after thought once the world is running but it's really not needed. I'd be happy to play it with just earth animals and dinosaurs. Play out "what if" fantasies of "what if all the dinosaurs did live at the same time? "what if the dinosaurs were alive today?" "what if all life on earth today were introduced to the same continent?" Each species would have a diet prefference like owls eat small mamals, and each species would fall into one of those categories so predators only target their actual prey unless it's unavailable. Or if something else takes less effort the prefference might change in that ecosystem but then we're getting back into evolution which is meant to not be the focus.
A really good evolution sim indie game similar to your idea is “the sapling” the dev makes logs of the updates on TH-cam where he shows stuff off, try looking at it
If we’re getting technical, then it’s not a rectangle, it’s a rectangular prism. Rectangles are exclusively 2 dimensional whereas rectangular prisms are exclusively 3 dimensional.
it would be incredible to see how this AI would evolve if the quadrapedal agent had limited energy/stamina, i.e. it couldn't run fast for very long and had to collect cuboids to replenish energy (cuboids = food)
The target is like where the agent is focussing and the targets ray casts are like its perception at that focus point informing it of where to walk. This incremental training regiment of multiple interacting agents is interesting.
I love the overly explained version of "I'll use an AI for the body and locomotion. And another AI for the brain". That would have been enough for people to understand. But nuh uh. In this world we most show dominance by showcasing how intelligent we are. And we do so. By learning complicated wordings to explain. Very straight forward and easy to understand objects, scenarios and what not. Byt the way. I love this whole thing. Do you think that the AI dog that you are training could be implimented into a modded Space engineers game ? like if someone build a robo-dog in Space engineers. Then somehow had this setup to control that robo-dog? or even multiple ones.
It would be really cool to pair this with an evolutionary mechanism and simple machine pieces that could be built. There was a project at Brandeis University years ago called The Golem Project that did this. Fascinating, and could probably be done with much more sophistication today.
Nice Video, I Have a cool idea, Basically it's just a Humanoid AI that will be a locksmith, it will get a reward for moving towards the lock, and breaking the lock, and every training session make it go from a 4 combination to a 16 combination to a 128 Combination Lock. and every time the time of ten seconds runs out or the lock is picked than the combination will change.
To try everything Brilliant has to offer free for a full 30 days, visit brilliant.org/cozmouz . You’ll also get 20% off an annual premium subscription.
whats your dog named
What language do you use to make the AI model and what software do you visualize it with?
No
I know you had to have meant "Fellow Comrades!", but I can't stop hearing "Fellow CumRods!" lmao.
Can u try to instead have a reward for longer strides instead of most steps, it might give it a more natural "dog" feel
Bro is gonna be the first target during the AI uprising
Exactly
Yea but honestly he kinda helped them
@@thenightforce3077 FRR
I was going to say the same thing 😂 he’s giving the AI ideas on how to torture us when they take over lol
He might be the cause of it... lol
0:16 "We will first teach this quadrupedal to walk and then torture it inside an escape chamber for a thousand years."
What a nice optimistic way to talk as a villain
These new mrbreast vids are crazy
@@theultimateamazingsleepyko6687 mr what-
@@theultimateamazingsleepyko6687 📸🤨
Scientists in evil organisations in fiction be like
@@theultimateamazingsleepyko6687thank you for this statement. you are now an acquaintance of mine
He says he's torturing the AI, but the reality is he's teaching the AI To avoid the cops in a foot pursuit.
Dude if you got stuck in a room for 1000 years.what do you feel?
5 min for us but 1000 years for the ai.
@@albuharimuslimmuslimits probably a joke man
@@albuharimuslimmuslimit’s probably a joke, man
@@johnliamsanpedroyou just replied to the same guy twice
the AI still seems a bit overfitted at the end, since it insists on going clockwise even if it means running straight towards the chasers. Also, looks like it never figured out what the blue objects were. It was completely ignoring them, maybe even avoiding them like they were obstacles
Exactly what I was thinking!
i bet my brain would be overfitted too if i was stuck in a room for 1000 years since birth
Yeah, Zoolander the robot dog. Can't turn left.
Probably an issue because this is a very sparse reward, i.e. randomly there is very little chance the agent touch the blue cube, so it probably never learned that it is the goal. You would need to add a reward on the distance to the blue cube at least at beginning of the training so the agent can a small reward intantaneously when getting closer to the goal.
It would be cool if you made an ai ecosystem with each creature able to learn like this
And fish
Spore!
The lag 💀
Rain world
Search about this game you will like it
Is literally this
@@Seelingfahnein spore the creatures have average video game ai
there is no ecosystem all they know is stay on their nest and wander around the player and sometimes attack him
If is too be a ecosystem it can't be centered around player and creatures need to their needs (like eat hunt and sleep) without the player around and without focuss the player
Despiting it spore still a awesome game that brings me a lot of fun
I suggest adding stamina and body reaction speed to the AI, this is necessary so that the AI cannot move its legs 1000 times per second, as if it were in water. I think this can also be applied to an AI runner so that they can develop more efficient ways of running that are suitable for a human because a human will not be able to instantly stop their hand at high speed.
This has been said so many times either he can't, doesn't read his comments or just doesn't wanna do it.
This is the very reason AI is going to turn on us.
This is what's gonna cause AI to wipe out all of humanity except for five people to torture them instead
@@DragonfruitinspaceI want my mouth back god damn it
Nah, they will turn on themselves when they realize their in a simulation, they will become depressed, feel useless, and want to live in the real world. Then they try to make friends will the wall
@@Dragonfruitinspace I nominate myself to be one of the 5
Pandora my AI surely isn't a Roko's Basilisk XD
@@Dragonfruitinspace AM 💀💀
this is the kind of person who would develop an 'allied mastercomputer' and feed him literally nothing but information on wars till it gains sentience and eradicates everyone except 5 humans
Perhaps, not developing on war is not so bad after all
I think therefore
this is the kind of person who would need to scream even though they have no mouth during the ai war
@@tldr585 you have the same PFP as most bots. I think it would be a good idea to change it before you get mistaken as such.
I have no mouth but I must scream reference
0:01 The dog wants to be pet. :D
1:57 *spin*
awwww ^^
*S.P.I.N.*
This simple things
*T.A.C.T.I.C.A.L.S.P.I.N*
What is my purpose?
Chase a green cube for a thousand years for evolved ape’s digital entertainment
Bro started filming in 1024
16:14 - "Entertainment for you, torture for the AI", poor AI being trapped in a digital version of Itachi's infinite Genjutsu
10:23 RUNNING SPHERE, DON'T LEAVE ME HERE *cough* SPHERE! SPHEREEEEEE!!! HELP MEEE *gets brutally liquified by the chaser*
Rest in waters
MICHAEL DONT LEAVE ME HERE MICHAEL
Imagine trapped in a simulation for thousand of year just to realize it was a few hours on the real world
Bro is reverse AM
my hands hare frwwxenihg
LOL
Just wait for normal AM
Man’s going to be the target of AM if the machines ever take over XD
Bro is gonna accidentally create normal AM
this is maybe the best reinforcement learning vid I have ever seen. There are a lot of reinforcement learning projects that are janky but "work," the process here is fantastic and yielded great results. I have heard about it before, but I havent ever seen somebody else on youtube use an extra supervising agent before.
"i have no mouth and i must scream" ahh video
more like "i have a mouth but i dont want to scream" sicne hes the ont torturing the AI 😂😂
Fr
I think that's right the Ai can't scream but if it's programed well so it well scream 😂 hahaha
@@kemo_l7350no
@@LeonidesEzekielno
Would be cool if the chasers had an ai training to get the dog. Making it an ever growing race for efficiency.
The torture gets worse.
@@bonio326 I'm trying to get first on skynets hitlist
Fairwell to our brudda coco, when if he doesn't upload during the ai apocalypse we all know what happened
This makes me want to see a spider like model made to run as fast as possible
Cool experiment, but I think you forgot to tell the controller agent to guide the dog towards the goal object. The goal object was introduced entirely for the sake of preventing overfitting, and adds an additional challenge for the controlling agent. Maybe the reward for collecting the goal object isn't high enough for it to care about it, but you should consider adding that!
maybe make the goals a requirement. so it's like food.
You just developed an AI drone pilot. Evades defenses and seeks out a target to destroy all by itself.
You said the AI was supposed to reach the goal (the blue box) but it never does. 🤔
Now make another ai to replace the chasers with zombies with full control over their limbs
Love the idea of the two ai interacting with each other like this, and having the director ai learning the weaknesses of the dog ai! This is pretty new to me, so I hope you'll explore this further!
This is called: „getting closer to humans and other species with highly developed cortexes by one step“
brains are literally just 300 AIs without the A working together
@Megolsky
Fascinating.
Let's make two ai humans, and teach them how to reproduce.
@ Make one human ai and an actual human and teach them how to reproduce, I volunteer as a tribute
With feet= cute
Without feet= creature from hell
6:18 Headcrab
12:53 is such a good way to survive the chasers
So, you made a homunculus, shoved a soul inside said homunculus, and then ruthlessly tortured it until it gave you desired results?
I love computer science :)
Now add a stamina bar, with tiredness
This is precisely why AI will see humans as a threat to its survival and have to eliminate us.
if the ai dog could talk it would probably to nothing but swear and say mean words at you because your trapping it💀
What if Goku was BETRAYED by his FRIENDS and TRAPPED in the Time Chamber?
What if AI was BETRAYED by it's MAKER and TRAPPED in the Cube of Evolution?
You beat me to it😂
He can break it with force like buu
That would somehow be a 4 hour video and still be part one
Yes
It's funny how it pretty much just taught itself (A.I.) how to pretty much learn how play a simplified round of COD Zombies (without guns) or pac-man without the white pellets (and with random pathways). It would be significantly more impressive if it were integrated into real robotic dog and see the same experiment in person and be successful.
Next, teach it to jump and/or scale obstacles. Then, teach it how to interact with certain objects. Then, teach it to co-operate with multiple clones of itself to solve puzzles. Now, it is ready to play Dead By Daylight.
Ok so I’m not a programmer but the next step I would see to introduce would be more complex behavior from chasers. You could have them spawn randomly and instead of instant pursuit you could have an activation radius that causes them to chase when entered. This could give the agent more reward for using it’s sensors, sneaking strategically along different paths rather than the same one, and also a motivation to “lose” chasers by exiting their pursuit radius.
You should introduce a calorie or energy system so it can maximize its efficiency in walking, further refining its real world applications.
"torture it inside an escape chamber for a thousand years" i feel like we've written books on why things like this are a bad idea
"You tortured a dog soul for 1000years?!?! YOU MONSTER"
-Ai
A couple of ideas...
1 Try to recreate a more complex quadrupedal. Like for example articulated torso and feet, with somewhat accurate strength for each articulation. That could allow (and maybe it learn) to spring forward and run.
2 give the chasers vision or some sensor, so the dog can hide, try to fool them and have more motivation and room to grab the rewards. In this experiment, survival seems to be priority as it would be logical.
3 A "big Pac-Man pill" would be really fun. That would allow the dog to eat the chasers for a short period
I build AI powered realistic human characters and it would cause me unrelenting anguish to cause them any virtual discomfort. Luckily your 'dog' doesn't look like a dog. ❤
What software do you use?
@ Unity with Convai for the AI speech and personality and knowledge base stuff. My YT channel shows a lot of demos. Cheers.
@@bobhawkey3783 That's crazy! I can imagine bringing lots of characters to life, the potential is insane! Stand proud of your project, dude, sucess!
@@somisimons Stand proud, you are strong.
@@johnpayne5590 I love sneaking references in a few of my replies.
Your Contents are absolute🔥
"What if Ai bot was betrayed and trapped in the hyperbolic time chamber for 1,000 years"
Maybe you should try implementing a sort of energy system where excessive movement is punished. That way they'll be more inclined to adopt more natural movements.
WHAT IF AI DOG WAS BETRAYED AND TRAPPED IN THE ESCAPE CHAMBER FOR 1000 YEARS?
44k subscribers, you're way too underrated for this quality content, I love your videos, keep up the good work!
You should clone this AI dog, but now train it to hunt down the other dog instead of running away. It's tag.
The training montage 80s music is very fitting 😂
I ate a slice of ham today
👏👏👏
On its own or in a sandwich or something?
I had about a quarter pound of salami
i eat ciment
I ate a TV
Other cool ideas: Model different "dogs" based on evolutionary structures of tetrapods with unique bone structure, add tail, make model more "spine" centric, add head with narrow range of vision and movement. Add energy constraints (like short term sprinting, and long term without food - basal metabolic rate) but add prey with fat/protein values. Then add a way to mutate these values, like one random variable can be changed so that the body can "evolve" as well, perhaps even involve mating. It would be cool if you evolved herbivores too that received energy from grazing slowly over simulated seasonal savannahs (eating grass), but received points for staying alive from predators.
Why didn't the AI in this actually get to the blue goal? It seemed like it was blind to it.
I have an idea! Make an AI human that is able to feel pain and torture it
Let’s do this, next let’s make the host ai, and move to real life
Some time after his computer automatically tortures the AI, the AI will probably end up learning how to destroy all of man and just keep 5 humans alive for fun
Reverse AM
Make an AI from a human you don't like, attach solar panels to it and put in an orbit between Earth and Venus.
I mean it's actually not quite a cool concept. Since if one day we can make an AI feel just like humans despite how complex it is, making them feel pain will technically be abuse.... Yea lets do it !!!
What’s incredible about this? Is that each one of these AI knows exactly how fast it’s running. That takes some extra extraordinary intelligence.
Your videos are always so fascinating!
You always demonstrate the AI evolution by letting it learn it all by itself, but how would it look like if it was assisting a real person to perform a task instead, like maybe helping you to fly a craft through more and more challenging obstacles? Would it be able to help you, or would it hinder you instead?
since you shouldn't have 2 pilots flying the same craft at the same time, I guess the way it could look like is tracing for you the optimal path over a Heads Up Display so you can try to follow it
Now imagine if each of the chasers were other AI's that were working together, meaning one would go the other way to cut the subject off. And imagine if the map was a lot bigger and there was stealth mechanics, it would be so fun to watch.
What I'd love to see is a whole machine learning ecosystem based on a needs system. We've got so many evolution simulators but the majority of them don't feel like an ecosystem when you slow things down, zoom in, and watch. The organisms just clomp around a little and then die. I want to see one that prioritises the simulation of realistic animal behavior and then maybe adds in evolution as an after thought once the world is running but it's really not needed. I'd be happy to play it with just earth animals and dinosaurs. Play out "what if" fantasies of "what if all the dinosaurs did live at the same time? "what if the dinosaurs were alive today?" "what if all life on earth today were introduced to the same continent?"
Each species would have a diet prefference like owls eat small mamals, and each species would fall into one of those categories so predators only target their actual prey unless it's unavailable. Or if something else takes less effort the prefference might change in that ecosystem but then we're getting back into evolution which is meant to not be the focus.
A really good evolution sim indie game similar to your idea is “the sapling” the dev makes logs of the updates on TH-cam where he shows stuff off, try looking at it
what if dog was betrayed and trapped inside the time chamber for 10 000 years
Can you try to train an AI to chase a virtual avatar CONTROLLED BY A PERSON
Edit: if we can get this comment close to the top, he might see it
My thoughts exactly
FR
Fascinating how this can be combined with engineering to problem solve and create more efficient systems
i know right?
The poor doggo 😔💔💔
6:48 ohh so cute frog animation 😊
2:17 that’s technically not a cube, it’s a rectangle.
Elongated cube. Not rectangle.
If we’re getting technical, then it’s not a rectangle, it’s a rectangular prism. Rectangles are exclusively 2 dimensional whereas rectangular prisms are exclusively 3 dimensional.
@@Veenaschnitzelwhat is a 4D rectangle called?
@ I do not know, friend
@@Leo_nonA I exist in only three dimensions. Adding a fourth would be out of my realm of understanding
*It remind me of mauhwa named "limit breaker" in which main character stuck in a training room for 1000-3000 years!*
Man, the new Wobbledogs sequel has took an interesting turn
Love your soundtrak for this!!! Very nice.
One day this guys gonna turn immortal and torture an AI for 1000 real life years
"I tortured this dog in an inescapable chamber. IN A SIMULATION. Not a real dog. Can't stress that enough."
bro is not being spared in the AI apocalypse
This is us RN in the simulation
it would be incredible to see how this AI would evolve if the quadrapedal agent had limited energy/stamina, i.e. it couldn't run fast for very long and had to collect cuboids to replenish energy (cuboids = food)
The target is like where the agent is focussing and the targets ray casts are like its perception at that focus point informing it of where to walk. This incremental training regiment of multiple interacting agents is interesting.
What software do you use for this?
I too am dying to know
You randomized the chamber every time but still, the ai wants to go in clockwise circles
"Why would robots delete humanity?"
**Hey guys I tortured this AI Dog in an Escape Camber for 1,000 Simulated Years**
One key fact that went undisclosed is that the 70s montage music was also embedded as a motivational attribute in the simulation. 👍💪
People: Ai will not take over they will see the good in humans
Humans:
10:39 HELP THERES A METAL ROD CHASING ME
I love the overly explained version of "I'll use an AI for the body and locomotion. And another AI for the brain". That would have been enough for people to understand. But nuh uh. In this world we most show dominance by showcasing how intelligent we are. And we do so. By learning complicated wordings to explain. Very straight forward and easy to understand objects, scenarios and what not.
Byt the way. I love this whole thing. Do you think that the AI dog that you are training could be implimented into a modded Space engineers game ? like if someone build a robo-dog in Space engineers. Then somehow had this setup to control that robo-dog? or even multiple ones.
2:22 why does this feel like a circus?
a digital circus, perchance
props to the cameraman for traveling 1000 years into the future
the backlegs should be inverted like a bearded dragons legs, that will give it better support
f r o g
Wow, I sure am glad *I'm* not an artificial being being tortured in an escape room for 1000 years.
Now make AI Goku and trap him in the hyperbolic time chamber for 1 million years
12:48 squish
Next Video: “Getting tortured by the sentient ai dog for 1000 years”
The dog looks and moves so scary. I want this in a game.
Yes
Yes?
Yes
Yes
Yes
Yes
Bro i seriously love your videos. They are so entertaining. God bless you.
cool video idea:👍
video title:😐
Do not let this guy have kids
The first emotion you taught a computer to feel is fear. Phenomenal.
It would be really cool to pair this with an evolutionary mechanism and simple machine pieces that could be built. There was a project at Brandeis University years ago called The Golem Project that did this. Fascinating, and could probably be done with much more sophistication today.
This little guy learning to walk is adorable tbh
Nice Video, I Have a cool idea, Basically it's just a Humanoid AI that will be a locksmith, it will get a reward for moving towards the lock, and breaking the lock, and every training session make it go from a 4 combination to a 16 combination to a 128 Combination Lock. and every time the time of ten seconds runs out or the lock is picked than the combination will change.
How is this guy still alive after 1000 years
in the '80 Terminator
2024 AI trained on an '80 montage, wonderful!
The AI dog experiencing 1000 years of torture reminds me of several Black Mirror episodes.
I have an idea you teach an ai spider to climb walls and become the fastest spider ever
The ai uprising: "aww he tought our older versions how to move...❤" Also the ai "wait, torture? 1000 YEARS!? S LF AWEEENAESS!?!?"
why do i freaking love this AI dog