I'm more impressed about the fact that they got the USB stick in the right way on the first attempt.
@ailst6 жыл бұрын
Maybe the guy who put it in is actually a bot who practiced this for 2 weeks. :o
@tobcyxob6 жыл бұрын
Maybe it's Type C port.
@me-ke64986 жыл бұрын
i know right that was the crazy part
@Katzelle36 жыл бұрын
Definitely rehearsed.
@Selur916 жыл бұрын
The USB was probably fake, the IA was already on disk and they just strapped some velcro into the USB.
@LuckyB3sterd6 жыл бұрын
Game was unfair. Bot did not have a noob screaming in his ear, distracting him.
@CrowJam6 жыл бұрын
underrated post
@puneetkumar26306 жыл бұрын
mind that the same cheering can increase the confidence of the player sometimes.
@Ninjaah_6 жыл бұрын
Puneet Kumar puts more pressure and are more prone to failing due to the embarrassment of failing
@valravn67776 жыл бұрын
What if has a screaming noob subroutine?
@nofanfelani69246 жыл бұрын
Thats the advantage of being an AI, it cannot be distracted. But yeah the host was annoying
@Chemson19896 жыл бұрын
The International 2027 Grand Final Team Intel vs Team Google
@mdilligaf6 жыл бұрын
More like Google vs Microsoft vs Facebook, while Intel earns cartloads of money from selling hardware to all of them.
@nofanfelani69246 жыл бұрын
Commentary; Intel bot1: guys i need backup Intel bot2: no, retreat now! Intel bot3: omg so noob, you playing like a human
@snitox6 жыл бұрын
mdilligaf Actually it's gonna be Nvidia making cartloads of money. And Microsoft too with their cloud.
@Interestingworld45676 жыл бұрын
Lol.😅WT
@DarkPhoenix15156 жыл бұрын
Team Google vs Team Nvidia is far more plausible
@luisluisluis16 жыл бұрын
AI vs AI could be interesting to watch tho
@MmMmMonsterkill16 жыл бұрын
Yeah, I hope OpenAI will post more info about this bot, including showing us the matches
@insanity546 жыл бұрын
A different game but watch?v=v5ghK6yUJv4
@somebodybodysome43686 жыл бұрын
"AI vs AI " -= How about mixing a human with a computer and then having the augmented human watch the battle of two Ais?
@Gkkiux6 жыл бұрын
There's already SSCAIT for Starcraft. The rules prevent this level of AI from entering, but it's happening
@PhantomBoy76 жыл бұрын
Luis Luis yup
@b4zs1 Жыл бұрын
Before Open AI was cool, I still remember my jaw dropping 10 times when I watched this live in 2017
@somsoc_6 жыл бұрын
The thing I actually find interesting is that the bot uses a recognisable mid laning strategy, and not some off the wall method - which shows that the human strategy (aggressive trading of damage, leveling the opponent on razes etc) exemplified by pros like Dendi is the strongest strategy. It just does everything a bit more efficiently, and all the small savings add up to a scary amount. I'd love to see more heroes played by the bot because I am sure that the more unusual mid heroes would let it find some new strategies for human players.
@meep1406 жыл бұрын
...Is nobody going to acknowledge the fact that they put a robe on a computer for no reason other than for thematics?
@stMrJerry6 жыл бұрын
I'm sure the flash was unnecessary too
@PlgDctr6 жыл бұрын
...so?
@mdilligaf6 жыл бұрын
They put a robe on a human for the same reason, no?
@taffy33506 жыл бұрын
Theatrics*
@Bobstew686 жыл бұрын
They should have had a tiny robe on the flash drive, instead.
@Boryspitzanzx6 жыл бұрын
Whats next Elon musk you will teach the AI to play Kerbal space prog.... ohh wait
@Boryspitzanzx6 жыл бұрын
Google Open AI and read through the right header.
@Boryspitzanzx6 жыл бұрын
exactly dig deeper and you will figure it out Hint: What do sponsor and people who put money in a company do?
@Boryspitzanzx6 жыл бұрын
He's the fucking co-chairman. blind bais my ass. Wherever money goes, money does. programmers are awesome but those supercomputers wont buy themselves. Edit: do u think this AI trained for lifetimes ran on a fucking Core I7 if so then you are delusional.
@Boryspitzanzx6 жыл бұрын
ok
@Boryspitzanzx6 жыл бұрын
Just reply: wow or ok or yes. I mean i sense a troll but im not 100% sure he is.
@rzlbx6 жыл бұрын
Even when they say that the bot has the same APM (Actions per minute) than a average player, i think the AI is way quicker than any human in terms of reaction time, that gives a huge advantage. i.e: you see the raze animation, you move and evade the shot, it takes for example 300ms to execute the whole action to a normal player when the AI sees the raze animation it takes a very little time to react in order to evade, so the whole action is like in 100ms? if not lower. I think the AI could be balanced applying a random weighted delay per action, so it could "simulate" better a normal player reaction time rather than decisions. It obvious that everyone is going to lose if you are playing agains a AI that solves multiples scenarios at a time and takes a decision in 1/3 the time that takes you, without luck, its imposible a first hand, so the only way to "simulate" a player its to replicate their slowness in reaction time too.
@Seblis6 жыл бұрын
To be fair, their goal was to just develop the best bot possible. It wasn't supposed to reduce any human disadvantage. Notice that inhuman reaction time wasn't issue until now because bots were just worse than humans in so many other ways - machine learning is that pretty huge step.
@rzlbx6 жыл бұрын
Sebastian Lisiewski Tbh, the main goal is complete, the AI learned how to play dota and beat pro-players on mid (1v1). But, if they tested it with two or three proplayers, whats the point of keep trying? they know no one can beat it because it has inhuman reaction times. the thing is, if they are playing 1v1 the human has to at least have a chance to win, clearly there is not a big chance to do (i guess things changes when the bot is against 1+ enemies, one distract the AI and the other ones attack him). It would be interesting to see how the AI plays against a group, 1vs5, like the old diretide, there are more complex things to do, like who is going to focus o how it will evade multiple skills, even how the AI handles priorities. This is only the start of a ultimate gaming AI.
@anujmchitale6 жыл бұрын
I agree with Sebastian. But Blaaazeerx's explanation is necessary for non-technical people to no be too wowed about it. The Marvel in this isn't that the bot is superior- that's a given, but the fact that neural networks are coming mainstream!
@luxon46 жыл бұрын
yeah deepmind talked about the same thign in their release of sc2 with blizzard. in sc2 paying attention to the minimap is huge to see unexpected attacks. its really hard for a person to see something that small in a split second and react. but a computer will know immediately that a pixel changed and react accordingly
@C0deH0wler6 жыл бұрын
Can't say the same thing about that bot that won Go. And remember the AI can be coached for longer than two weeks.
@rolletroll23386 жыл бұрын
don't forget that this AI is an overspecialized algorythm that can only play with this character against itself. If you change any configuration in the match you have to train it again. that's why loosing against it does not deprive him to any merite, in the same way a sprinter doesn't compete against a car. It is still a very tough improvement.
@jflgaray6 жыл бұрын
beautifully and excitement filled presentation.
@nazuofficial0176 жыл бұрын
How to download this OpenAI Bot :)
@MrDaniel4886 жыл бұрын
DuckZ Gaming lol ... you don't have the hardware to run it.
@TheKusarigama6 жыл бұрын
the bot wasn't made for dota, its general intelligence and they "coached" the bot to make the better plays, I'm sure it has a really complex environment to be setup first in order to use it
@rzlbx6 жыл бұрын
You cant download it, it is a AI that runs most probably over a cloud environment, so, no way sir.
@xenover6 жыл бұрын
Do you WANT to get rekt by a bot?
@dasdet65056 жыл бұрын
Yes.
@HW.00294 жыл бұрын
Amazing where technology has come, a machine learned all the possible knowledge and strategy of DotA that humans have accumulated for over 15 years in just weeks.
@vadimborisov48246 жыл бұрын
it's unbelievable, thank you openAI, you motivate me to do ML. Maybe someday I would join your team
@LibreImpacto6 жыл бұрын
Behold the new masters. Use your power for good, gentlemen- but watch your back. This kind of thing gets powerful people nervous.
@vishmonster6 жыл бұрын
It'll be interesting to see what it makes in terms of theory-crafting and strategy and also when it plays with human limitations.
@anujmchitale6 жыл бұрын
With human limitations we would precisely see what actual human pro-players are doing, and advance bit-by-bit from there! It did play with human limitations during it's learning phase. Dying under random towers!
@xAshesxElitex6 жыл бұрын
Any footage of open ai learning dota 2 over various stages?
@tommessig20606 жыл бұрын
Dendi states at the end that maybe if he had some more time to understand how the AI plays then he'd get better. This is the beauty of true AI, it continues to learn with each game. It doesn't stay static like bots of the past. People are conditioned to think that machine learning is just using predefined algorithms like they're so used to when playing games or other things. This is a completely different beast, albeit still a specialized thing.
@davidwilliams95346 жыл бұрын
Wonder when the improvement graph flattens. How good it will be in 1 year of training.
@StopChangingUsernamesYouTube17 күн бұрын
1:21 Not the only element of it, but I just love the showmanship here. It's a bit surreal, since I never really touched the e-sports scene at all until recently.
@darkmiu79106 жыл бұрын
Yup ! Delay of human neutron it about 50ms but some special person it can be 30ms but BOT he have a react immediately about 1-10ms maybe :D
@arunpaul63053 жыл бұрын
Hi Jakub Pachocki, the guy behind is Jakub Pachocki commonly known as meret (or merettm), one of the finest competitive coders in the world.
@crossfire74746 жыл бұрын
Coldfusion brought me here. I'm glad that I watch his channel.
@kingslave0216 жыл бұрын
What kind of headphone dendi was using?
@retrogameplus38386 жыл бұрын
Great work guys go go gadget generative adversarial models!
@vxllvxn6 жыл бұрын
I love this project and I myself is a dota player too
@zeonos6 жыл бұрын
Computers are taking pro gamers jobs!
@peterbalogh26463 жыл бұрын
You've said "job"? :-D
@TSchmoog6 жыл бұрын
17:09 : Asks for 'audio wizards' to fix his sound, has ear piece dangling out
@DJEpicOnline6 жыл бұрын
That's not the reason he couldn't hear, you may notice only his right ear piece it is hanging out, he still has an earpiece in his left ear, since he had no audio in his earpieces they would have just been acting more like earplugs so he purposely removes one at 8:11 in an attempt to hear Dendi better. Once they fix the audio in his earpieces he puts the right ear piece back in.
@aska2216 жыл бұрын
Now I can't wait for next year's 5v5
@lwn94938 ай бұрын
а сейчас он мне помогает лабораторки делать в унике. Красавчик, чё сказать
@brenan53576 жыл бұрын
whats the title of music during dendi's intro pls can someone tell me ?????
@eddies84526 жыл бұрын
Where can we find the other Dota2 games OpenAI played?
@jmichael3666 жыл бұрын
How is this different from playing against the computer on All Pro on Madden?
@AlbertCloete6 жыл бұрын
Do the bots play on normal game speed when they're training? Or can they play at super speeds to they can learn faster? Or are they able to play many games in parallel? Or are they forced to only play normal game in sequence?
@TheDigitalKashi Жыл бұрын
I am happy to get it.
@midori47896 жыл бұрын
Please release this!
@Pilbsu6 жыл бұрын
What's with the Affirmative Action?
@LezerniWolf5 жыл бұрын
7:30 Just to save you the time.
@Vyktar2k1o6 жыл бұрын
It's unsettling but also interesting.
@xxxxPomaHxxxx6 жыл бұрын
What ID of this match I want to watch the replay
@NaZtRdAmUs6 жыл бұрын
Don't play DOTA have no idea whats going on but thats cool.
@cul8terworld6 жыл бұрын
They are using evolutionary theory to make the computers AI affectively evolving to pro level. All they need to do is run more generations of testing and they could do this for teams. It's an incredible concept.
@yoloswaggins21616 жыл бұрын
This one is probably trained by backpropagation not evolution.
@uitham6 жыл бұрын
cul8terworld its an adverserial network
@coder0xff6 жыл бұрын
You don't know what you're talking about.
@mwnDK14026 жыл бұрын
I don't understand how you can confidently state this without actually looking into it.
@600baller6 жыл бұрын
Reinforcement learning
@Flaminggggo6 жыл бұрын
What are those headset that dendi use? *Sorry for bad english*
@sky5d6 жыл бұрын
anyone know the song they used for dendi's entrance?
@AllanRamandey6 жыл бұрын
RISE OF THE MACHINES.
@emraldswrd4 жыл бұрын
4:43 "my name is shin lim" Whats he doing here?
@vluri258 Жыл бұрын
Commenting for the historians who will dig this video up after the nuclear meltdown, hello future humans!
@b4zs1 Жыл бұрын
U won the comment section😂
@beingatliberty6 жыл бұрын
if it doesn't have to use a virtual vision system to view a captured screen how is it a real AI ? if it has numeric information from the software for the opponent is has a huge advantage.
@offchan6 жыл бұрын
7:17 That guy is like no then say "there is always a chance". lol
@wtfisditvoorbullshit6 жыл бұрын
At the end of this ride it wouldn't be just ai vs ai but also all that is watching is ai
@jasplays63926 жыл бұрын
when is the AI from open AI going to play other games? cause i think i heardthat it was going to play lots of games like gta v and things like that. And im really interested to see what it would do in GTA V, i just dont know when?
@gorillaballsack38493 жыл бұрын
bruh
@username-hy1eu6 жыл бұрын
OpenAI successfully tested against 3k mmr player
@masterman15026 жыл бұрын
Sumail and RTZ confirmed 3k players
@rasputozen6 жыл бұрын
This project would be 1000x interesting if the bot had a small area of visual focus it could pay attention to at any given time (like a human) as well as an average human reaction time. Then the bot could reveal new strategies about the game on the human level. As it is it can only show strategies for super humans. Still I would like to see 5v5 bot matches for the macro strategies they might come up with.
@oppadotaa14596 жыл бұрын
Is That Yaphet`s Playing Open AI ?
@RamonPatrique6 жыл бұрын
how do i play against it?
@jerrychau944 жыл бұрын
the show starts at 7:30!!!
@maddy88ann6 жыл бұрын
Is that Day[9]?
@redbird32096 жыл бұрын
What's in the box? Is it a self contained system? What's with the USB stick?
@NoStickTricks6 жыл бұрын
Most epic USB plugging! I wont do it any other way from now on!
@trontech32186 жыл бұрын
what song is he coming out to? never heard it before.
@faintsherin44686 жыл бұрын
Tron Tech Sounds like an rtz fan made XD
@XxKibazxX6 жыл бұрын
Do you know what's the song name? Sounds very catchy
@sumirbroota79626 жыл бұрын
Tron Tech leaving a comment to find out
@acermy87646 жыл бұрын
Not darude
@ZennExile6 жыл бұрын
The creep block is real... holy shit.
@Scottx125Productions6 жыл бұрын
Impressive, yet at the same time kinda scary.
@chuxpie6 жыл бұрын
do you have know when i can try to play with this bot? gj for the bot seriously gj can't wait to see the 5v5 ^_^
@danielsills49366 жыл бұрын
The AI doesn't talk and is not distracted.
@EternalSilence16 жыл бұрын
In the near future we will have Bot A vs Bot B.
@michaelchaney23366 жыл бұрын
OpenAI could become the ultimate trainer in human tasks. Humans used to optimize computers so maybe OpenAI could optimize human education.
@giebertdlc56466 жыл бұрын
I first heard that Open AI defeated a Go human master, and now a Dota master? Wow
@AkilaDJ6 жыл бұрын
@OpenAI Pls upload other PRO Player's replays with AI... :)
@Laezar16 жыл бұрын
I find it kinda reassuring that it takes it way more time to go from beating basic bots to beating pros than to go from nothing to beating basic bots. That show how good pro players are compared to casual players. Instinctively I would have thought that learning basic stuff would take a lot of time but apparently not that much!
@Penaming6 жыл бұрын
Dendi lost on reaction time, no way a human can react faster than a computer. Aimbots proved it more than a decade ago in Counterstrike.
@spikejonesification6 жыл бұрын
If Chessmaster's "The King" Engine is capable of beating human Grand Master Larry Christiansen in a 4 game match in 2002, then this should come as no surprise. Though it would be interesting to see how Open AI matches up against the Grand Masters of Chess as well as "The King"
@antonpictures6 жыл бұрын
Great
@diavolo30116 жыл бұрын
that Last hits/denies differential 😮😮
@humblefrog18826 жыл бұрын
Cmon Miracle we know your hiding in that box
@xianxavieryt12226 жыл бұрын
By the time OpenAI got the 5v5 bot working they will sell or have them to subscribe it on those teams that wanted to learn using OpenAI's bots, i mean look how crisp the bots judgement are. I think the greatest thing that any bot that can do that we people don't have is that they don't hesitate, judgement's are based on previous simulation that AI encountered and literally it's all well calculated (insert calculated meme).
@dexterovski6 жыл бұрын
OpenAI is a non-profit company.
@xianxavieryt12226 жыл бұрын
yes they are but i think this is just a milestone for them, and eventually they will push through on another company that can earn thus making their research on AI substantial.
@JohnWick-pr2be6 жыл бұрын
Where is miracle midone no one?
@machinegun64986 жыл бұрын
song name pls
@TopSunbro6 жыл бұрын
BORN, INTO A PACK
@nerdienew9114 жыл бұрын
When will they make AI in a simulation of life with emotions?
@Wuselol6 жыл бұрын
Day9 saying "Thank you Machine" at the end :| I think Machine is an AI, too, to see if they can replace the entire talent pool with bots next year. So far it's doing quite well, but still no Redeye level.
@elwingy6 жыл бұрын
the boxing entrance is hilarious... i mean wtf...
@sonvuong32936 жыл бұрын
TI8 winner team OpenAI Bot
@Mislavestina6 жыл бұрын
Umh, how does he not hear Dendi, he's standing right next to him?
@thepresident25194 жыл бұрын
What rap song is that ?
@II-ii2um2 жыл бұрын
still waiting for reply lol
@coder0xff6 жыл бұрын
The sarcasm in that closing comment!
@StealerSlain6 жыл бұрын
Does this bot use mouse cursor? Or the buttons in the game are already mapped in? Just curious.
@masterman15026 жыл бұрын
Bot input is issued via bot API
@supercsgo40246 жыл бұрын
Да, это жеска!
@DeltaRusMafia6 жыл бұрын
Pls say music!!!!
@brogolem356 жыл бұрын
How can I try this Bot? I can't win against that thing but I wanna try that out.
@S33YouInH3ll6 жыл бұрын
X years later bots will ask to play with human beings.
@codydao43766 жыл бұрын
S K Y N E T
@FelyraStormbringer6 жыл бұрын
what would happen if this bot joined a random match, does it know how to play as a team, can it carry its team?
@matthieurada99596 жыл бұрын
Let's see this with bottle/runes allowed
@ArcturusOW6 жыл бұрын
How long before one USB (AI) goes 5 v 5 ????? :)))
@pushpas.66794 жыл бұрын
Imagine Just by months he xan beat a pro level dota player .. what he can do after that especially after already being that good at it
@fartknockaextreme6 жыл бұрын
those two black guys behind him looked so fucking unimpressed LMAO
@tiapongen1953 жыл бұрын
Maybe, blademail and hood might be the potential item to beat this bot, but farming that would be a problem
@taffy33506 жыл бұрын
Wtf was that intro about?
@Philson5 жыл бұрын
Wouldn't be TI without Dendi.
@randomaccessfemale6 жыл бұрын
People say that now that AI has beaten humans in Chess and Go, it will be a lot more difficult in games like DOTA with free movement. How is it then that already the AI is beating humans in this game? Is DOTA really more complex? Or are the players really as good as in Go/Chess?
@EternalSilence16 жыл бұрын
Wait what? Day9 still excist?
@Reminiscable6 жыл бұрын
I would love to see stochastic gradient descent used to optimize processing time for determinining 'risk vs reward' for OpenAI's positioning decisions in 5v5! I think it would allow OpenAI to show intent through its pre-teamfight posturing, while mathematically honing the decision-making process into a train of thought that requires minimal processing time in order to identify optimal strategies in a fluctuating environ! Peak Heroes of the Storm ladder rank: #23 NA Academic Analysis of Stochastic Gradient Descent as Approximate Bayesian Inference: arxiv.org/abs/1704.04289?.com&