@@q.bensis9524 you are more than one in 7 billion, because you are a human, how lucky is that? It’s not really that lucky, because everyone has different outcomes for their luck. Just like dream’s speed runs.
The fact that the A.I actually went “then why don’t I just pause it, problem solved.” Is the highest iq shit I have ever heard of in my entire life 💀 Thats low key scary intelligent
For an A.I its not really a scary intelligent. Imagine you have a goal, and you have information about all the ways to achieve it, you just need to learn them. One day you'd too come to this conclusion.
It’s the same solution as in that minecraft meme a long time ago, when there wasn’t really any “goal” in the game other than surviving. People would say, “well, all you need to do to beat the game is build three blocks down and cover up your head.” And it was true. There was no way you’d ever die in that early version of minecraft if you decided to isolate yourself in a 1x1x2 cage.
@@dj-murlock it didn't have all the information the ai literally had to figure everything out on itself. how can you have information but need to learn it? contradiction
@@arko.0.1. by "have information about all the ways to achieve it" I meant having the information about conditions in which the task would be considered done
The problem with A.I in games is that once an A.I goes down one path it can become nearly impossible for it to discover a trick that involves going far out of its way. Kinda like how you don’t see animals with guns strapped to their arms. While this would help the animal once this trait is evolved, the path to get there would be difficult because none of the evolutionary traits on the way there are helpful and might even be harmful. However, this is only if you make the sole goal to make it to the end as fast as possible. A better way to do this would be to have humans give it a nudge or have the A.I try to beat the fastest human time in a section of a game or both.
This, also tricks that involve things like Dying and resetting the game, the DOTA project came up with tactics for dying because dying is not particularly avoidable, and is not a hard loss condition(also on that note, parsing the "viability" or "desirability" of particularly absurd glitched states would be hard)
“we could get to the point in speedrunning where the norm is letting a computer figure out how to beat the game quickly then humans just execute what the computer figured out” TAS but reversed
@@arciks11 nope. Tas is just a real human being (or 10) that make inputs to a game, then play them back. Look up dysijalon de gaming (its one am I can't spell that) has a bunch of vods titled "making the perfect subspace run, one frame at a time" or something like that, and he inputs the game with his hand, but with emulator slowdown, savestates, etc. Also in the mk wii community you will see videos called tas challenge x, and there all real humans, but perfected to the best of there ability.
This is the worst case scenario for me since the theorycrafting and glitch hunting are by far the most interesting parts of speedrunning for me. By just following the instructions that the computer gives us, the speedrunners themselves would basically become the robots.
@@SourPickle-bv9gd lol, this is what I call human ego, we are not special and we are not ✨creative✨, there is no such thing as creativity as you think there is, humans just think of surviving and reproduction, that’s why you feel love, that’s why you can be sad, creativity is also just a few pulses in our brain to impress opposite (usually) sex, it is just a few pulses the same way it is with AI
@@SourPickle-bv9gd hahah, yeah they are diffrent but it doesn't change anything, just show me your definition of creativity and I will either find AI that is creative with this definition or code one myself
@@Graverman "creativity is also just a few pulses in our brain to impress opposite (usually) sex" [...] to impress someone because of human creativity, an ai wouldn't find any piece of art interesting and couldn't compare quality of for example two different books on it's artistic value, it would struggle (if even was able to) to find a basic storytelling error and definitely couldn't tell if a book is just a random story or if it has a metaphorical meaning, wouldn't know if it has good pacing or if the way is written creates good atmosphere, unless it was fed with an already existing story and had someone code in that it was good or bad, so these "few pulses" wouldn't impress any ai whatsoever unless it was told to like or hate something, in which case it's still not being creative but being hardcoded to do that, and just the fact that every human has a slightly different taste about art that exists since their birth proves that liking or hating some form of art isn't a matter of being teached to do that so go on, make an ai that has subjective opinions but still can objectively judge if an artwork is a good book with fine storytelling, a good painting made with a correct technique etc., or if it's something made by a bored kid, that's something that an ai would never do since judging an art's meaning doesn't require just an artwork itself but also not only knowing but understanding all the context behind it, of course make the ai without any pre-existing knowledge because an ai that judges a book by comparing it to other books it was fed with is just an algorithm that uses data input by humans based on - wouldn't you know it - human creativity, and that doesn't prove your point at all
There's a great video on Two Minute Papers about Open AI learning how to play hide and seek. The AI discovers and uses glitches to win. All very interesting stuff on multiple levels.
Now I'm just imagining a guy leaving his A.I. to speedrun Minecraft for him, after giving him permission to change seeds ''Master, I have found a seed in which the Ender portal appears under you and is already activated. Btw a random zombie jumps in after you and it happens to have full netherite gear and since he can't withstand traveling through dimensions, it gets dropped onto you.''
Tetris AI: **Pause game to never looses** Professional Tetris player playing for month to years trying to get better: "My goodnes, what an idea. Why didn't i think of that?"
honestly, for games like dota, as long as the AI is "smart" enough, it has absolute advantages in the game such as reaction time, teamwork, consistency and the lack of emotionally driven miscalculations a 1v1 game with ample time for each move that has almost limitless variations and possibilities to negate the advantage of higher processing power would probably provide a fairer match
Well I would agree but they haven't posted much videos but one thing I have to say is that the editing and the commentary rivals that of an experienced RUclipsr that is a professional
Well I would agree but they haven't posted much videos but one thing I have to say is that the editing and the commentary rivals that of an experienced RUclipsr that is a professional
Imagine self-learning A.I specializing in PvP. All human-possible inputs. Completely undetectable by any anti-cheat, and will play better than any human could after a few thousand games of trial and error. Now imagine a swarm of them on popular PvP/minigame servers like Hypixel. If the source code for the AI gets leaked or released in any way, it would be the end of public servers as we know it. There is nothing the owners can do about it since all the bot inputs are humanly possible, and every other player would be a PvP god because they are literally played by a computer that has done thousands, if not millions or billions of fights and improved through each and every single one of them.
@@Zenovarse anticheats only detect algorithms from hacked clients, like a hit from a distance that's impossble on vanilla, movement that's too fast to be considered normal without using something that propels the player to such speed (e.g. ender pearl or TNT), or a scripted lag period, or robotically consistent clicking that a human would not be able to do. A self-learning AI would have none of those things, and no clear pattern, which makes it extremely difficult, if not impossible, to detect via anticheat.
@@WomanSlayer69420 why a self learning ai would not have those things. Would you script the AI so that it have certain behaviours? (E.g. open ai with manually added reaction time)
@@WomanSlayer69420 but even if you manually add avoidance for the anti cheat into the AI, I cannot imagine the AIs as we have them today would be undetectable with a good enough anti cheay
@@Zenovarse it's not manually added. See the thing is, AI executes human inputs, unlike a script which exploits the game to do something that's not possible in vanilla. Basically, a machine is playing Minecraft, on vanilla. There is no way of detecting it, since it's not a hack client and everything done by the machine is doable by a human player, although the machine has vastly superior reflexes and processing. Basically, it's like a "super player" which could definitely ruin games without getting caught.
This is so cool! In Super Paper Mario there's a trick called green bridge skip that saves a little over 2 minutes. the trick requires pixel persison of .001 which was difficult for even TAS'ers to get a hold of. One of the amazing community members created a script that showed runners exactly which inputs they needed down to the exact frame and now the trick is done in any% runs. Computers are so cool!
This is something I've wanted to do for years. What interests me most is the possibility for an AI that just plays the game, instead of being efficient at one task. Thank you for sharing your code!
Roko’s Basilisk is a thought experiment based upon the concept of a super intelligent A.I, wreaking time travel vengeance upon all those who did not contribute to its’ creation. Which implies that by even knowing about this, you then become a potential target and thus a victim of that A.I.’s future wrath.
if AI relies on the world and not visuals, it could even learn how to find strongholds just by looking at the terrain generation, also it could find key items super quickly by seeing trough walls (because those items exist, players just don't see them)
as long as it can only do what humans can do (not limited by physical and brain processing power limitation ) its fine and not cheating, if a human can't fly the ai should not be able to fly, if a human can't jump 999999 blocks high the ai should not eb able too, but if it would be possible for the human or its possible to find the position of the blocks the ai should be able too
I think there is a TAS seeded minecraft video where they beat the game sub 1 minute. For something like an AI learning algorithm running RSG, there would most likely be information gathered from the environment / what blocks are visible, rather than literally having X-ray.
@@gabrielandy9272 it's not that it should be possible for humans, just that it shouldn't have things that equate to literally cheating like X-ray. For example there is a frame-perfect trick in SM64 where you can gain speed with multiple back-to-back frame perfect wall jumps than the AI uses to clip through some walls which are technically not 'cheating' in any way, but are essentially impossible for humans to execute. Nothing you said is wrong, just elaborating on what you said.
I know i'm looking at this late but "would struggle to make a cool piece of art" just to be clear, AI is actually good at that too, because creativity isn't as creative as you'd expect.
it's easy, because this channel is a total ripoff, literally copying content or parts of the content, and not even putting any work into changing stuff. It's a shame he got this far tbh
I think this video proves more just how scary AI can be if the goal is not simplified to the extreme, who would expect that as a goal “not losing” actually made AI think that pausing the game means you never lose.
most AIs can't just look way ahead in chess and similar games, it works very well but in basically any real-time video game, this method fails horrendously in Minecraft for example, we have A LOT of controller inputs that can be pressed on every frame. But I think the best theoretical AI could very easily beat humans: by observing terrain generation, it could crack the world seed, and by cracking the world seed, it could already know where to go, and how to beat the game as fast as possible. Btw just plugging in a game into an AI would fail to numerous other reasons, which include inhuman inputs, precision, knowledge or calculation. In theory, a perfect AI could find a floating point precision perfect trick, that is too unrealistic for humans, and that would cause the run to die if failed.
i dont know if somebody already pointed this, but if the AI's startspeedrunning and end up discovering methods not even thinked of, a lot of them wouldnt be achievable by human because we simply doesnt possess the enough K/PS to do imput these commands, or even worst: the controllers that we are using wont will be able to withstand the massive amount of input in a matter of less than seconds and thus end up being the bedrock wall that we wont be able to bypass
The only problem I see with the whole A.I. learning Minecraft is that A.I. is ran off of a trail and error based program, which for even the most adaptable people in the world it will still take several runs of Minecraft to learn what all blocks of Minecraft do (assumption, I played Minecraft when it first released so I know just about every interactive blocks ins and outs.) From what I have seen in several A.I. Platforms, such as chess, is that the A.I. Must rather see every known aspect of what is going to happen, (to plan its next several thousand steps to complete the game) or it needs to be able to run trail and error programming. As for Minecraft uses a Randomized generation of the map the A.I. Will never be able to learn how to go from one map to another and still beat it in 1 life. Therefor an actual speed run of Minecraft (which I only see in being hardcore so you can’t cheat) will not be possible without a plug-in of sorts that will allow the A.I. To run it’s several next 100 steps in real time without change. Another problem I see is the fact that newer open world or RPG games don’t typically run off of a this is where this character will be, it will never change, type of code. Like in Minecraft Mobs, will move around, so the code will need to be able to adapt to the movement of the mobs. Example: If the time change in a game is night, and hostile mobs spawn during the night, then the best course of action would be to find a place or make a shelter. But if your hunting for an in-game necessary item to beat the game, and can only be obtained during the night, It will never be beaten by the A.I. Unless a plug-in is enabled that will allow the A.I. To see each and every mobs next movement and action. Example 2: Let’s say we’re loaded in a new world of Minecraft, brand new, and you see a creeper. Typically you would avoid it, but if that creeper moves in the path of you avoiding it you will move around or completely change direction. A.I. Can’t do that without the trail and error based programming. They may know what the creeper can do, and which blocks and route to take. But if that creeper moves in that direction it would have to change and recalculate another path in .003 seconds (random number just chose it because that’s about as fast as the human brain can react l, I’m not sure though we may react even faster or slower) Of course once this can be achieved and an A.I. Can process everything it will need to do to keep ‘you’ alive to beat the game maybe they can beat it. But until the my assumption will be they will not be able to beat the human in a Open World or RPG speed run, without at least a test or trail and error on that same map already.
Not really rage quit more like smart ass it’s way out of the game the goal set for the A.I. was to not lose by that logic if the game is pushed it can’t be lost.
I know I'm commenting om a 4 months old video, but a correction for 1:54 : the concept is called "reinforcement learning. Neuro-evolution is a specific method that can be used to implement reinforcement learning. It's the algorithm SethBling used, but it's quite dissimilar to how OpenAI Five works.
Wait. Imagine an AI speedrunning a Pokémon game. That requires so much creativity. Maybe the main used Pokémon will change? And how is the AI gonna overcome the luck factors? I really wanna know
that's mostly what happen when you ask the AI to do something, it does what you tell it to do, not what you want it to do, in this case the developpers asked for "not losing" and not "highest score" so the AI did what it could to "not lose" i.e. pausing the game. and if you ask it to have the highest score it might discover a glitch to have an ever increasing score (it already did in some game)
While yes Ai is so advanced that it can predict multiple outcomes and etc. Those outcomes aren’t a guarantee to happen but also fighting an Ai is while yes, challenging you have to find a weakness to the program and exploit it. You have to be analytical and observant, predict what the AI would do next and notice the patterns that AI or AIs have done during the game. If you disrupt the patterns that they’re doing it may give a chance to help you win a game. Also the fact that while A.I. are programmed to win with the best strategies they come up with, unpredictability may help you win.
If pausing was a logistical win, I think every human would be trying to do that for minecraft speed runs at the fastest possible reaction time. XD “BOOM! I WIN FOR PAUSING THE GAME AND NEVER BEATING IT!”
Imagine some random A.I made by a hobo on the internet surfaces, beats the game in a few minutes and spits in the face of the whole community, what a sight that would be.
i was happy reaching 1k subs, 40k+ subscribers is nuts. thanks!!! 💗
please fact check literally anything you say ever
Your very welcome!
Well deserved
Wtff i saw u a week ago with 723 subs
@@callmespeed7979 he has a main channel that I think he abandoned for this. Can’t remember I think it was a WOW channel
Everybody gangsta until the AI learns to edit the game files to increase the drop chances for pearls and rods
It's not file, nbt, rng changes it's dream luck if you have that no need of those things
@@q.bensis9524 you are more than one in 7 billion, because you are a human, how lucky is that? It’s not really that lucky, because everyone has different outcomes for their luck. Just like dream’s speed runs.
AHAHAHAHAH
Lol poor dream, his Reputation has been stained for a very looooong time.
Looks like you have a stack of likes
A.I in movie : Unkilleble robot that terminate the entire earth
A.I in real life : pause = not loosing
if you think it is more sophisticated and philosophical solution
@@talhaoztunca1833 Or it was just rage quit xD
L O G I C .
you can not allow them to pause and its fine, the example is nothing crazy
There is a phrase that is said a lot in programming: Computers always do what you tell them to do, but they rarely do what we want them to do.
"man spends years [playing Tetris, finally gets record for longest time playing Tetris"
AI: pause
It's big brain time
The closest thing we have in minecraft is the private branch of baritone that basically features an ai combst bot.
The fact that the A.I actually went “then why don’t I just pause it, problem solved.” Is the highest iq shit I have ever heard of in my entire life 💀 Thats low key scary intelligent
For an A.I its not really a scary intelligent. Imagine you have a goal, and you have information about all the ways to achieve it, you just need to learn them. One day you'd too come to this conclusion.
It’s the same solution as in that minecraft meme a long time ago, when there wasn’t really any “goal” in the game other than surviving. People would say, “well, all you need to do to beat the game is build three blocks down and cover up your head.” And it was true. There was no way you’d ever die in that early version of minecraft if you decided to isolate yourself in a 1x1x2 cage.
@@dj-murlock it didn't have all the information the ai literally had to figure everything out on itself. how can you have information but need to learn it? contradiction
@@arko.0.1. by "have information about all the ways to achieve it" I meant having the information about conditions in which the task would be considered done
“So far we have only talked about simplistic games” This hurt my chess soul on a deep level
chess is just easy to calculate for an ai, thats why.
simplistic isn't the same as simple
@@GU-RP joe mama
@@siennaq5553 w h a t
It is simple for a computer.
Everyone is gangsta until A.I had the world new record of minecraft speedrun
Oh it does oh it does the set seed with glitch percent speed run :-)
So I guess this is the future of speedruning and imagine I.A speedrun in real life that would be interesting
yh and u have to code it your self
damn. two letters and you still messed it up. jk jk
Now that's so dark humor 😳😂
Irtificcial Antelligence
@@piotrduda6019 LOL
That diamond swords name at 6:26 is really interesting.
"with the goal of not losing it decided to just pause the game instead of play" hes to powerful to be kept alive
human: Play Tetris for as long as possible without losing
Ai: *Pauses*
human: now listen here you little shit
I mean you can make it so the ai can't pause lol
The problem with A.I in games is that once an A.I goes down one path it can become nearly impossible for it to discover a trick that involves going far out of its way. Kinda like how you don’t see animals with guns strapped to their arms. While this would help the animal once this trait is evolved, the path to get there would be difficult because none of the evolutionary traits on the way there are helpful and might even be harmful. However, this is only if you make the sole goal to make it to the end as fast as possible. A better way to do this would be to have humans give it a nudge or have the A.I try to beat the fastest human time in a section of a game or both.
This, also tricks that involve things like Dying and resetting the game, the DOTA project came up with tactics for dying because dying is not particularly avoidable, and is not a hard loss condition(also on that note, parsing the "viability" or "desirability" of particularly absurd glitched states would be hard)
“we could get to the point in speedrunning where the norm is letting a computer figure out how to beat the game quickly then humans just execute what the computer figured out”
TAS but reversed
I don’t really like the idea of an AI making runs, it might take the fun out of finding new glitches and new optimizations
Regular runs do sometimes do that with some TASes
@@arciks11 nope. Tas is just a real human being (or 10) that make inputs to a game, then play them back. Look up dysijalon de gaming (its one am I can't spell that) has a bunch of vods titled "making the perfect subspace run, one frame at a time" or something like that, and he inputs the game with his hand, but with emulator slowdown, savestates, etc. Also in the mk wii community you will see videos called tas challenge x, and there all real humans, but perfected to the best of there ability.
This is the worst case scenario for me since the theorycrafting and glitch hunting are by far the most interesting parts of speedrunning for me. By just following the instructions that the computer gives us, the speedrunners themselves would basically become the robots.
@@angeldude101 Hence the term: TAS but reversed.
A year later. AI can make both art and music now. Just a friendly update
haha that comment aged well
AI: Can write book
@AnyPercent: AI stragglers with creativity
It makes books with rng and knowledge it knows not creativity
@@SourPickle-bv9gd lol, this is what I call human ego, we are not special and we are not ✨creative✨, there is no such thing as creativity as you think there is, humans just think of surviving and reproduction, that’s why you feel love, that’s why you can be sad, creativity is also just a few pulses in our brain to impress opposite (usually) sex, it is just a few pulses the same way it is with AI
@@Graverman AI and Human Brains are incredibly different and if you can’t acknowledge that I reccomend you think about it a little more
@@SourPickle-bv9gd hahah, yeah they are diffrent but it doesn't change anything, just show me your definition of creativity and I will either find AI that is creative with this definition or code one myself
@@Graverman "creativity is also just a few pulses in our brain to impress opposite (usually) sex"
[...] to impress someone because of human creativity, an ai wouldn't find any piece of art interesting and couldn't compare quality of for example two different books on it's artistic value, it would struggle (if even was able to) to find a basic storytelling error and definitely couldn't tell if a book is just a random story or if it has a metaphorical meaning, wouldn't know if it has good pacing or if the way is written creates good atmosphere, unless it was fed with an already existing story and had someone code in that it was good or bad, so these "few pulses" wouldn't impress any ai whatsoever unless it was told to like or hate something, in which case it's still not being creative but being hardcoded to do that, and just the fact that every human has a slightly different taste about art that exists since their birth proves that liking or hating some form of art isn't a matter of being teached to do that
so go on, make an ai that has subjective opinions but still can objectively judge if an artwork is a good book with fine storytelling, a good painting made with a correct technique etc., or if it's something made by a bored kid, that's something that an ai would never do since judging an art's meaning doesn't require just an artwork itself but also not only knowing but understanding all the context behind it, of course make the ai without any pre-existing knowledge because an ai that judges a book by comparing it to other books it was fed with is just an algorithm that uses data input by humans based on - wouldn't you know it - human creativity, and that doesn't prove your point at all
There's a great video on Two Minute Papers about Open AI learning how to play hide and seek. The AI discovers and uses glitches to win. All very interesting stuff on multiple levels.
Got a link?
@@Kaitri ruclips.net/video/Lu56xVlZ40M/видео.html
Now I'm just imagining a guy leaving his A.I. to speedrun Minecraft for him, after giving him permission to change seeds
''Master, I have found a seed in which the Ender portal appears under you and is already activated.
Btw a random zombie jumps in after you and it happens to have full netherite gear and since he can't withstand traveling through dimensions, it gets dropped onto you.''
Dam that 95 percent in the chat had me dead
Imagine ai in like a hospital or something. You could like alway immediately diagnose patients and know how to treat them
You telling me A.I/robots will be able to kill us after trial and error?
"humans have creativity on their side like an ai can't draw a painting"
Dolli E 2: hold my beer.
A.I : pause not loosing. pause *FOREVER*
That A.I. pausing the Tetris game had me laughing big time. Pro gamer move!
Tetris AI: **Pause game to never looses**
Professional Tetris player playing for month to years trying to get better: "My goodnes, what an idea. Why didn't i think of that?"
honestly, for games like dota, as long as the AI is "smart" enough, it has absolute advantages in the game such as reaction time, teamwork, consistency and the lack of emotionally driven miscalculations
a 1v1 game with ample time for each move that has almost limitless variations and possibilities to negate the advantage of higher processing power would probably provide a fairer match
ai are running in pure math and logic
3:16 the master strategy of survival
AI: pausing = not loosing
AI: "Am good at this game"
no its just good at the goal it was given: not lose
3:11 Never expected to see this sort of denial.
Everyone gangsta till HOI4 AI does the holocaust speedrun
This channel needs more subs
Well I would agree but they haven't posted much videos but one thing I have to say is that the editing and the commentary rivals that of an experienced RUclipsr that is a professional
Well I would agree but they haven't posted much videos but one thing I have to say is that the editing and the commentary rivals that of an experienced RUclipsr that is a professional
Person: try not to die lol
Ai: OK, *P A U S E S*
Imagine self-learning A.I specializing in PvP. All human-possible inputs. Completely undetectable by any anti-cheat, and will play better than any human could after a few thousand games of trial and error. Now imagine a swarm of them on popular PvP/minigame servers like Hypixel. If the source code for the AI gets leaked or released in any way, it would be the end of public servers as we know it. There is nothing the owners can do about it since all the bot inputs are humanly possible, and every other player would be a PvP god because they are literally played by a computer that has done thousands, if not millions or billions of fights and improved through each and every single one of them.
What do you mean undetectable by anti cheat? Surely it is easily detected?
@@Zenovarse anticheats only detect algorithms from hacked clients, like a hit from a distance that's impossble on vanilla, movement that's too fast to be considered normal without using something that propels the player to such speed (e.g. ender pearl or TNT), or a scripted lag period, or robotically consistent clicking that a human would not be able to do. A self-learning AI would have none of those things, and no clear pattern, which makes it extremely difficult, if not impossible, to detect via anticheat.
@@WomanSlayer69420 why a self learning ai would not have those things. Would you script the AI so that it have certain behaviours? (E.g. open ai with manually added reaction time)
@@WomanSlayer69420 but even if you manually add avoidance for the anti cheat into the AI, I cannot imagine the AIs as we have them today would be undetectable with a good enough anti cheay
@@Zenovarse it's not manually added. See the thing is, AI executes human inputs, unlike a script which exploits the game to do something that's not possible in vanilla. Basically, a machine is playing Minecraft, on vanilla. There is no way of detecting it, since it's not a hack client and everything done by the machine is doable by a human player, although the machine has vastly superior reflexes and processing. Basically, it's like a "super player" which could definitely ruin games without getting caught.
they are slowly taking over, speed running games, and soon life
Next stage of minecraft speedrunning would be people competing to code the best ai to speedrun the game
I came here to watch air beating minecraft
This is so cool! In Super Paper Mario there's a trick called green bridge skip that saves a little over 2 minutes. the trick requires pixel persison of .001 which was difficult for even TAS'ers to get a hold of. One of the amazing community members created a script that showed runners exactly which inputs they needed down to the exact frame and now the trick is done in any% runs. Computers are so cool!
Ai learn to trash talk will be so funny
Ai:your mom is bigger than my calculation
This is something I've wanted to do for years. What interests me most is the possibility for an AI that just plays the game, instead of being efficient at one task. Thank you for sharing your code!
Ah yes, one step closer to make artificial friends :')
And somewhere one more step close to evil ai or you can say destruction ai modification. How is it?
@@Xyshsus Robot waifu is not incompatible with genocidal robot, and the first is more important to archive than to prevent the later
Roko’s Basilisk is a thought experiment based upon the concept of a super intelligent A.I, wreaking time travel vengeance upon all those who did not contribute to its’ creation. Which implies that by even knowing about this, you then become a potential target and thus a victim of that A.I.’s future wrath.
I want to see an AI survive 100 days in hardcore Minecraft
if AI relies on the world and not visuals, it could even learn how to find strongholds just by looking at the terrain generation, also it could find key items super quickly by seeing trough walls (because those items exist, players just don't see them)
if humans can obtain seeds and can see the algorithm it is valid I think because you can also obtain those information
@@talhaoztunca1833 depends of the category
as long as it can only do what humans can do (not limited by physical and brain processing power limitation ) its fine and not cheating, if a human can't fly the ai should not be able to fly, if a human can't jump 999999 blocks high the ai should not eb able too, but if it would be possible for the human or its possible to find the position of the blocks the ai should be able too
I think there is a TAS seeded minecraft video where they beat the game sub 1 minute. For something like an AI learning algorithm running RSG, there would most likely be information gathered from the environment / what blocks are visible, rather than literally having X-ray.
@@gabrielandy9272 it's not that it should be possible for humans, just that it shouldn't have things that equate to literally cheating like X-ray. For example there is a frame-perfect trick in SM64 where you can gain speed with multiple back-to-back frame perfect wall jumps than the AI uses to clip through some walls which are technically not 'cheating' in any way, but are essentially impossible for humans to execute. Nothing you said is wrong, just elaborating on what you said.
That Tetris AI was simply amazing
new minecraft world records category : A.I. Random Seed and A.I. Custom Seeds
Been subscribed since the first video and love all the videos related to Speedruns.
Love watching them but could ever do one myself
Its hilarious watching how stupid the open Ai is in the early generations, scary watching how smart they are in the late generations
I know i'm looking at this late but
"would struggle to make a cool piece of art"
just to be clear, AI is actually good at that too, because creativity isn't as creative as you'd expect.
Genius
but like fr, how do you not have more subs
it's easy, because this channel is a total ripoff, literally copying content or parts of the content, and not even putting any work into changing stuff. It's a shame he got this far tbh
Well this was mildly terrifying.
I like how it opens with TAS where the human tells the computer what to do and ends with how the computer will tell the human what to do.
This is actually crazy ngl
Plot twist: pausing tetris is the intended way of not losing
So conclusion is we are insanely pointless once AI can make itself?
I think this video proves more just how scary AI can be if the goal is not simplified to the extreme, who would expect that as a goal “not losing” actually made AI think that pausing the game means you never lose.
@@thunderzgaming3427 yea
most AIs can't just look way ahead
in chess and similar games, it works very well
but in basically any real-time video game, this method fails horrendously
in Minecraft for example, we have A LOT of controller inputs that can be pressed on every frame. But I think the best theoretical AI could very easily beat humans: by observing terrain generation, it could crack the world seed, and by cracking the world seed, it could already know where to go, and how to beat the game as fast as possible.
Btw just plugging in a game into an AI would fail to numerous other reasons, which include inhuman inputs, precision, knowledge or calculation. In theory, a perfect AI could find a floating point precision perfect trick, that is too unrealistic for humans, and that would cause the run to die if failed.
There are so many videos about minecraft speedrunning made around when this was made that feature dream clips and they never stop being funny to me.
Imagine AI RUclips channel that everybody on Earth likes
Skynet can take over the world if it wants, but play video games better than me? Better believe I’m starting a rebellion
but thats just a theory, a game theory, oh wrong channel
The other side of the coin is quality assurance. Imagine AI finding all the glitches and bugs in the game.
i dont know if somebody already pointed this, but if the AI's startspeedrunning and end up discovering methods not even thinked of, a lot of them wouldnt be achievable by human because we simply doesnt possess the enough K/PS to do imput these commands, or even worst: the controllers that we are using wont will be able to withstand the massive amount of input in a matter of less than seconds and thus end up being the bedrock wall that we wont be able to bypass
I just hear robots are going to take over
Yea it could eventually do all that but one thing the a.i. can’t do is feel the emotions we do when playing games
0:49 Oh, I am a computer!
The only problem I see with the whole A.I. learning Minecraft is that A.I. is ran off of a trail and error based program, which for even the most adaptable people in the world it will still take several runs of Minecraft to learn what all blocks of Minecraft do (assumption, I played Minecraft when it first released so I know just about every interactive blocks ins and outs.) From what I have seen in several A.I. Platforms, such as chess, is that the A.I. Must rather see every known aspect of what is going to happen, (to plan its next several thousand steps to complete the game) or it needs to be able to run trail and error programming. As for Minecraft uses a Randomized generation of the map the A.I. Will never be able to learn how to go from one map to another and still beat it in 1 life. Therefor an actual speed run of Minecraft (which I only see in being hardcore so you can’t cheat) will not be possible without a plug-in of sorts that will allow the A.I. To run it’s several next 100 steps in real time without change. Another problem I see is the fact that newer open world or RPG games don’t typically run off of a this is where this character will be, it will never change, type of code. Like in Minecraft Mobs, will move around, so the code will need to be able to adapt to the movement of the mobs.
Example:
If the time change in a game is night, and hostile mobs spawn during the night, then the best course of action would be to find a place or make a shelter. But if your hunting for an in-game necessary item to beat the game, and can only be obtained during the night, It will never be beaten by the A.I. Unless a plug-in is enabled that will allow the A.I. To see each and every mobs next movement and action.
Example 2: Let’s say we’re loaded in a new world of Minecraft, brand new, and you see a creeper. Typically you would avoid it, but if that creeper moves in the path of you avoiding it you will move around or completely change direction. A.I. Can’t do that without the trail and error based programming. They may know what the creeper can do, and which blocks and route to take. But if that creeper moves in that direction it would have to change and recalculate another path in .003 seconds (random number just chose it because that’s about as fast as the human brain can react l, I’m not sure though we may react even faster or slower) Of course once this can be achieved and an A.I. Can process everything it will need to do to keep ‘you’ alive to beat the game maybe they can beat it. But until the my assumption will be they will not be able to beat the human in a Open World or RPG speed run, without at least a test or trail and error on that same map already.
TDLR:
Video didn't answer the question but it definitely teaches you a lot about AIs, specifically AIs made for Dota.
the first 4 seconds of the video gives of college ad vibes
the only way to survive ai invasion is to not let the ai know what is right or wrong
Mar. io was such a sick video from SethBling
Did anyone else uncontrollably laugh when they heard a computer rage quit tetris?
Not really rage quit more like smart ass it’s way out of the game
the goal set for the A.I. was to not lose by that logic if the game is pushed
it can’t be lost.
I know I'm commenting om a 4 months old video, but a correction for 1:54 : the concept is called "reinforcement learning. Neuro-evolution is a specific method that can be used to implement reinforcement learning. It's the algorithm SethBling used, but it's quite dissimilar to how OpenAI Five works.
Wait. Imagine an AI speedrunning a Pokémon game. That requires so much creativity. Maybe the main used Pokémon will change? And how is the AI gonna overcome the luck factors? I really wanna know
One thing:
BARITONE.
The most smart A.I in the game currently.
Ok the part where it paused tetris creeped me the hell out.. i wont lie
that's mostly what happen when you ask the AI to do something, it does what you tell it to do, not what you want it to do, in this case the developpers asked for "not losing" and not "highest score" so the AI did what it could to "not lose" i.e. pausing the game. and if you ask it to have the highest score it might discover a glitch to have an ever increasing score (it already did in some game)
A problem i see with ai speedrunning is updates. It would have to adapt in wich could lead to issues.
and that's why many speedrunners use only one version of a game! Or, at least, different version is automatically is in a different category.
I mean, one of them Killed Dream, so...
That the AI press Pause afraids me a lot 😳
A. I. can already speedrun MUCH better than humans
Finally someone make a video about this
Imagine if an ai is given a task to make earth better and it ends up killing all the people
I was expecting him to talk about Baritone... That program can literally play and thrive on 2B2T of all servers.
Stop bringing 2b2t into unrelated videos.
@@jwwr8408 The main point is that Baritone can play the game very well, without human input.
I WAS JUST THINKING ABOUT A MINECRAFT AI
Minecraft speedrun going from 2h to 10-15min is in large part caused by game updates
This will be speedrunning in 10 years let’s be honest
No one:
(Possibly) the AI in the future: let’s speed run on how fast we can disappoint the humans
This is how skynet is born
While yes Ai is so advanced that it can predict multiple outcomes and etc. Those outcomes aren’t a guarantee to happen but also fighting an Ai is while yes, challenging you have to find a weakness to the program and exploit it. You have to be analytical and observant, predict what the AI would do next and notice the patterns that AI or AIs have done during the game. If you disrupt the patterns that they’re doing it may give a chance to help you win a game. Also the fact that while A.I. are programmed to win with the best strategies they come up with, unpredictability may help you win.
Ngl when he started talking about Dora 2 I thought he was advertising
happy TAS noises*
an AI beat minecraft glichless set seed
90 sec!
Um actaully that's pre determined inputs made by a group of humans, so that doesn't count.
If pausing was a logistical win, I think every human would be trying to do that for minecraft speed runs at the fastest possible reaction time. XD “BOOM! I WIN FOR PAUSING THE GAME AND NEVER BEATING IT!”
A.I cant stop us for winning games
Imagine some random A.I made by a hobo on the internet surfaces, beats the game in a few minutes and spits in the face of the whole community, what a sight that would be.
If he can make an AI he’s not gonna be a hobo software engineering is good money
Ending human kind speedrun any%
-Ai probably
everyone gangster untill ai learn to to play game with 180year worth experience in a day
All I really got from this video is the best way to play Tetris is to not play
This right here is very engrossing
More vids pls I'm in love with this content
clicked on this to watch an ai beat minecraft man
The thing that i learnt is that A.I’s rage quit tetris.
ayy Taran got featured lol
"Could THIS A.I" what IS the A.I you're talking about ?