AlphaStar vs Serral - Game 2

  Переглядів 170,926

ArtosisTV

ArtosisTV

4 роки тому

AlphaStar's Protoss vs Serral's Zerg, Game 2.
AlphaStar vs Serral - Game 1:
• AlphaStar vs Serral - ...
AlphaStar vs Serral - Game 3:
• AlphaStar vs Serral - ...
Also check out more of my content:
Twitch.tv/Artosis
Twitter.com/Artosis

КОМЕНТАРІ: 360
@twoorthree2243
@twoorthree2243 4 роки тому
Gonna use that "won't surrender, feel free to quit the game when you're winning" intro for my ladder games from now on
@quishzhu
@quishzhu 4 роки тому
lol
@tmccann1876
@tmccann1876 4 роки тому
It's a bold move cotton
@DanyPell
@DanyPell 4 роки тому
lmfao. thats how it got to GM!
@ThuhElement
@ThuhElement 3 роки тому
F’ing A Cotton! F’ing A.
@darthmortus5702
@darthmortus5702 3 роки тому
Alphastar after losing its whole army: *I'll be back.*
@dannygjk
@dannygjk 3 роки тому
lol yeah and as Artosis pointed out at that point Alphastar's tech was at a high enough level to take care of business. It's tempting to say that Alphastar deliberately went into that bad battle to buy time.
@killchicken123
@killchicken123 3 роки тому
You can't do that!
@Blighted_Ashes
@Blighted_Ashes 2 роки тому
"OH no, anyways." Lol
@QuintinMassey
@QuintinMassey 2 роки тому
And thus after that day, the terminators ruled the world 🌎
@TheScoobysteve
@TheScoobysteve 2 роки тому
​@@dannygjk I agree, it's tempting to attribute everything AS does to 'I meant to do that', however I think its more likely that it simply adapts much more quickly than a human (and won't succumb to the temptation to double down on a bung strategy) .
@PeagmaticDreamer1199
@PeagmaticDreamer1199 4 роки тому
I know that all the pausing and explaining is irritating to Starcraft experts, but many people here (most of the people who have no idea about Starcrafts intricacies but fascinated by AI/Deepmind) are watching these videos precisely thanks to the detailed explanations. I am one of them. Maybe you can post 2 versions? A Detailed commentary version (this one) and a Standard commentary version with no pauses or speeding up.
@dirtypure2023
@dirtypure2023 4 роки тому
agree, I'm here for the in-depth coverage
@cortster12
@cortster12 4 роки тому
Ditto. I have absolutely no idea what is going on otherwise.
@jonnysith
@jonnysith 4 роки тому
100%. Those starcraft "experts" need to grow the fuck up and calm the fuck down.
@FirstLast-gk6lg
@FirstLast-gk6lg 4 роки тому
@@jonnysith to play devils advocate millions of people watch these videos for entertainment value. It is an exciting game. And Serral(the best human) against AS(the fabled AI of doom) is a really important game and we expect that it would be really exciting. But the pauses break the momentum just like pausing a video of any game. And id argue that the people who are disappointed with all the pausing are the people who think that this really important game shouldnt be the one used as an educational vehicle. Why choose the superbowl as the game you pause and explain what the lines on the grass mean? Know what I mean?
@SirCanuckelhead
@SirCanuckelhead 4 роки тому
First Last to play devils advocate, content creators put out the content they want and people have the choice to watch it. Art has always loved to do deep analysis, so anyone complaining of him putting out a video like this are selfish and entitled.
@alansmithee419
@alansmithee419 4 роки тому
Serral: I've adapted to your strategy really well Alphastar: *takes heavy damage several times* Also alphastar: *wins anyway*
@mrrodgers0
@mrrodgers0 3 роки тому
Alphastar didn't win tho. Its remaining army could not possibly have taken down the hive with constant reinforcements from serral and no substantial reinforcement of its own. (Remember the baneling run-by's? The AI had like 30 probes left. This was a final desperation move.) Serral left because, as the bot has said at the beginning of every game, it wasn't set to quit out of games like it did on the ladder.
@alansmithee419
@alansmithee419 3 роки тому
@@mrrodgers0 so serral forfeit immediately after losing a major battle because he knew he was going to win? If the outcome didn't matter, why didn't he quit earlier? Why does the person whose job it is to comment on StarCraft 2 matches disagree with you? And presumably he was in some form of contact with serral to confirm that he did indeed lose. Plus in game four, he continues playing for minutes after he's supposedly guaranteed a win. Yes, it was a final desperation move... And it worked.
@mrrodgers0
@mrrodgers0 3 роки тому
@@alansmithee419 Yeah, after seeing his approach in game 4, i have to agree that this was serral gging, albeit prematurely. I think he just didn't realize just how brutal the banelings had been on Alphastar's economy. The game really was serral's to lose there imo. Hives are very tanky, and with basically just 1.5 base's worth of zealot stalker the only probable reinforcement for protoss, I think serral could have held. Most of those archons were low on shields, and serral had +3 melee and adrenal glands done. It wouldn't be efficient, but flooding lings here would be able to mop up with good surrounds.
@dannygjk
@dannygjk 3 роки тому
@@mrrodgers0 I prefer to go with Serral's judgement. He has a higher probability of being correct.
@ChristianNeff
@ChristianNeff 3 роки тому
@@mrrodgers0 I can't believe how everyone actually thinks Serral's GG was justified... If this was Game 7 of a GSL final and he GGed like that everyone would have been screaming at him for quitting too early. This was a epic fail on Serral's part.
@myselfremade
@myselfremade 4 роки тому
I wonder if AlphaStar is used to seeing a ling all-in about the time it was boosting out the colossus, and routinely does that as insurance?
@lordbrimshawduke-wellingto2045
@lordbrimshawduke-wellingto2045 4 роки тому
yes i thought the same thing, it makes a lot of sense 👍
@alexdouglas8523
@alexdouglas8523 4 роки тому
Sort of a Meta-Game relic from training I suppose?
@aszbzpszbz9786
@aszbzpszbz9786 4 роки тому
Since he played millions of games against himself and on ladder he probably has seen everything and then some.
@MagicRabbit
@MagicRabbit 4 роки тому
That seems to be the answer. Human players tend to stop doing stuff too not get cheesed, because it rarely happens. If Ling-All-In falls off in the meta and is just a cheese people don't really prepare for it. Alpha doesn't forget the games its lost against cheese, so it will always be prepared for it. In it's training it will most likely have lost a lot of games against an Zerg doing a strategy you would call cheese. Amazing.
@vitorodino8760
@vitorodino8760 4 роки тому
I still feel like wasting that much of production time on such a crucial structure is not worth it. Either finish the colossus, or build the disruptor and wall correctly, etc. Wasting an entire immortals build time + chrono boost is a waste. Making an immortal there and just properly holding a ling all-in is strictly better than cancelling a colossus.
@aguyinavan6087
@aguyinavan6087 4 роки тому
In game 1 Serral had his roach attack at 7:40, which would have been precisely when the colossus would have been finished and positioned. But it was scouting Serral's base and realized that there was no buildup. The preparation was just in case, but since the coast was clear it the resources would be better spent building it's 4th base.
@gluestickgenius2644
@gluestickgenius2644 4 роки тому
I do not think the agent was set up to adapt to a single opponent.
@qwertyrewtywyterty
@qwertyrewtywyterty 4 роки тому
nope, it didn't learn from a game, it learn from training matches, this was not it.
@Tenoke
@Tenoke 4 роки тому
@@qwertyrewtywyterty yeah, but it saw that Serral has the tech and likely resources for it. Serral instead spent a bunch of larva on drones I think but from AlphaStar's POV an attack was likely for the same reason why the attack even happened first game at that time (tho of course it doesn't know about the first game at all).
@hughesd22
@hughesd22 4 роки тому
Sounds like you're post hoc rationalizing the move to me. It could just be a random move that usually addresses some other issue that isn't present here. We need to be careful not to personify an agent and read into it motivation and intent where there is none.
@zilliq
@zilliq 4 роки тому
Colossus to counter a roach timing ? I don't think so
@arcticfox2281
@arcticfox2281 4 роки тому
200 iq moves 1.imperfect wall is designed to trap and kill zerglings at the start so zerg players cannot drone up and expand fast( perfect probe micro and destroy zerg player's ideal objective to build a minimal army to hold harassment and outexpand opponent) 2. disruptors instead of immortals becoz protoss small army size cannot hold long enough for immortal kill zerg numerous units one by one. 3. 3 disruptors becoz too difficult to complete dodge and keep army intact after 3 shot. instant advantage for protoss before reinforcements arrival
@aaronlindberg8235
@aaronlindberg8235 4 роки тому
Disruptors are extremely good for zoning and add a lot more maneuver potential to a protoss army. Only that they require too many actions to command prevents human players using them.
@NomSauce
@NomSauce 4 роки тому
@@aaronlindberg8235 Yeah it seems that it's not particularily the Disruptors themselves that are hard to control in isolation, but rather controlling everything around it simultaneously seems extremely hard. Optimal shots require both initial positioning of the whole army and then reactive positioning of not only the nuke itself but also his own units have to react to whatever inputs Serral makes as he has to be careful of stray zerglings trickling in to cause a suicide bomb.
@TheDuckofLaw
@TheDuckofLaw 4 роки тому
Thanks again for the expert commentary. Love your videos!
@devonchin94
@devonchin94 4 роки тому
Thank you for the amazing analysis and breakdown of this series Artosis! I love that you made it very easy to understand for those who may not have ever played Sc2, but are here because of their fascination of AI/DeepMind. I wanted to present what I think AlphaStar may have been doing when it built, chrono'd, then cancelled that Colossus. Like regular players, we know that there are certain timings in every game. Whether it's an upgrade timing, rushing when +1 or +2 is finished, or other timings for when you know the opponent may have something. My first thought with this, is that AlphaStar was pre-emptively building a Colossus with chrono boost, to prepare for a "possible" timing attack/defense of some kind for whatever reason. However when it looked at the game again, it saw that it would not need this, and cancelled the Colossus. Any thoughts Artosis? Things like this really fascinate me, when AI does something that we can't wrap our head around, yet it did do it for a reason.
@yoloswaggins2161
@yoloswaggins2161 4 роки тому
If you can run this thing on a regular computer it would be a super nice training partner, maybe not so much for the best players but definitely for the rest of us. Imagine having a grandmaster practice partner on call 24/7.
@ClayJitsu
@ClayJitsu 4 роки тому
dreams may come true
@kavlo32
@kavlo32 4 роки тому
I've learned in IT school that neural network is VERY cheap to run but take a lot of computing power to train. I don't know if Google's AlphaStar follows this rule since it's not really a normal deep learning algorithm though
@fupopanda
@fupopanda 4 роки тому
Back in the January exhibition, the data scientist heading the Alphastar development said it can run on a regular gaming laptop. The training, however, requires far more computing power.
@zilliq
@zilliq 4 роки тому
Not so much for the best players? It just beat serral twice in a row 😁
@duyson5632
@duyson5632 4 роки тому
Serral is literally the best StarCraft player in the world, and it beats him 2 game in a row 😂😂😂
@MarioAndreschak
@MarioAndreschak 2 роки тому
Thank you for all the in-detail explanation about the units and buildings. I have literally NO experience in Starcraft, and know nothing. So this was indeed very helpful to understand what's happening.
@bellecheytac4171
@bellecheytac4171 4 роки тому
I love the play by play detailed analysis
@blueberryblues5644
@blueberryblues5644 4 роки тому
Artosis: Starcraft ASMR
@alexanderelsen9397
@alexanderelsen9397 4 роки тому
I feel like that colossus build was cancelled as soon as alphastar deflected the harass on it's third. So was likely Alphastar preparing for big zergling attack, but then after deflecting the harass decided it was no longer a threat due to the zerglings it'd killed.
@BoonceTV
@BoonceTV 4 роки тому
Please ignore the complaints about pausing. Anyone can commentate over a replay but what makes these videos great is the analysis!
@Mark-kw3nr
@Mark-kw3nr 9 місяців тому
So nice for someone who doesn't know SCII super well to get the commentary from a pro on the AI and its moves. Loving this series.
@mikeserds2333
@mikeserds2333 7 місяців тому
too much pausing tho. there are other casters who manage to explain a lot while just letting it run. over couple of games you will get the same amount of information while being able to watch more matched
@PunmasterSTP
@PunmasterSTP 2 роки тому
I have only the most basic of understandings of the strategies involved, but perhaps that’s why this seems so incredible to me. I also really love the commentary!
@rospincasocl3741
@rospincasocl3741 4 роки тому
Really great content! Thank you!
@zeratulofaiur2589
@zeratulofaiur2589 3 роки тому
That Colossus was a preparation for all in! What a genius move
@quishzhu
@quishzhu 4 роки тому
It turns out that AlphaStar's protoss algorithm converges to Stats' PvZ style :)
@piglet141
@piglet141 4 роки тому
Even grandpa can understand StarCraft now. Thanks Arthritis.
@LokiTheAnsuz
@LokiTheAnsuz 4 роки тому
It's Arthrosis lol
@clickaccept
@clickaccept 4 роки тому
looked like a premature gg to me.
@Kleavers
@Kleavers 4 роки тому
Yeah, he could have sacrificed a base and made mass Roaches. AlphaStar's economy was shattered and could not come back from that imo.
@ClayJitsu
@ClayJitsu 4 роки тому
serral threw, taking kickbacks from deepmind
@Tomoyo0827
@Tomoyo0827 4 роки тому
clickaccept every serral GG feels premature to me wouldn’t you agree? Sometimes I just think he’s being cocky, if you go over his lost games he always gg when he has more units.
@shawnmercado2219
@shawnmercado2219 4 роки тому
Absolutely. He must not have known how much damage he did to the econ.
@user-og1dw7hn1i
@user-og1dw7hn1i 4 роки тому
I don't play StarCraft and I felt the same .. he had a better economy and probably more stored materials ( if there is such a thing .. he could've played a game of attrition and rock paper scissors for a bit longer and might have won .
@clauskirkegaardclausen685
@clauskirkegaardclausen685 4 роки тому
This is really good stuff :-)
@jonathanosborn4800
@jonathanosborn4800 Рік тому
Doesnt creep also slow the enemy units down?
@Godlike-87
@Godlike-87 3 роки тому
23:35 alphastar starts juggling archons in the warp prism. It allowed archons to take maximum shield damage before pulling them to the back for the less damaged archons to absorb the incoming attacks. That is actually terrifying levels of efficiency. Shields heal and a shieldless archon can still Ifire at the enemy from behind a healthier unit while regenerating.
@Baleur
@Baleur 4 роки тому
16:08 god i really dont like the "bald patches" in the creep nowadays. I think it looked soo much better in the past versions of SC2 when the creep fully covered the ground. This, this just looks like George Costanza's head.
@ilkersonmez
@ilkersonmez 4 роки тому
George is getting upset!
@nightlyowll
@nightlyowll 4 роки тому
I love when he says "So, this is at normal game speed" when it is actually on "FASTER"
@yamchadragonball6983
@yamchadragonball6983 4 роки тому
Starcraft is normally played on "Faster" in pvp. So in that sense it is "normal" :)
@LSZreduction
@LSZreduction 4 роки тому
Did serral have to gg here? I feel like he couldve lost a base and been almost even, I guess most of his tech was at the natural though
@Djorgal
@Djorgal 4 роки тому
Alphastar can be very derpy in endgames and doesn't always know how to crush an opponent when it has the advantage. So Serral could have won still.
@lordbrimshawduke-wellingto2045
@lordbrimshawduke-wellingto2045 4 роки тому
@@Djorgal yes and perhaps a tiny counter-attack could have drawn away the archon attack to buy enough time
@Chyrre
@Chyrre 4 роки тому
Yeah, he could have dropped the roach warren and spire at then fifth and bought time with ling harass, it would take alphastar some time to get serrals econ down to its level
@DrewCocker
@DrewCocker 4 роки тому
Looks like Serral wanted to play another opener. He knows Zerg can win late game.
@DevinDTV
@DevinDTV 4 роки тому
It really didn't look like that attack from Alphastar was going to survive. When Serral GG'd, he could have instead went AFK and his roaches at the natural would have killed half of Alphastar's army. The archons were like 1 hp each. I doubt he would have lost his natural, and he was up 50 supply.
@plunderersparadise
@plunderersparadise 9 місяців тому
Imagine playing vs Serral and killing 15 workers at the 5 minute mark
@leosousa7404
@leosousa7404 4 роки тому
Amazing lol. Alphastar doesnt "look" perfect yet it manages to win
@valdirsalgueiro9087
@valdirsalgueiro9087 4 роки тому
Power of protoss
@jonnysith
@jonnysith 4 роки тому
yeah the nit-picking of its "mistakes" don't really matter if it still wins in the end.
@typingcat
@typingcat 4 роки тому
Because its only goal is to win. It doesn't care other things like making perfect moves.
@fangiscool1
@fangiscool1 4 роки тому
It isn't perfect. It honestly seems like it does some horrible things and is saved his its insane micro. The average APM is constrained to be below 300, but during combat the APM spikes well over 1k sometimes. Very top pros do not get even close to that level of APM, let alone effective APM. I think the AI should be constrained much more than it is to make it fair with humans.
@e4Bc4Qf3Qf7
@e4Bc4Qf3Qf7 4 роки тому
@@fangiscool1 Thats the old version. It doesnt do that any more, they changed it massively. Watch the intro video.
@josephluce9886
@josephluce9886 4 роки тому
For the cancel of the colossus can come from games against another AlphaStar. There might have been attacks from a Zerg AlphaStar during that time and it would have liked an extra unit to defend even though Colossus are not part of its final composition.
@bodgee1094
@bodgee1094 2 роки тому
Wait really? Did I miss something? Serral seemed to have this in the bag? Sure he might lose oke base but he should eventually be able to hold with making more queens and roaches?? 22 lings and 5 roaches on the way could easily kill 4 low health archons.
@dbob132
@dbob132 2 роки тому
Serral would have lost the structures enabling the production of those units (spawning pool, roach warren) which take a really long time to rebuild. With the warp prism with the army AlphaZero can warp in extra units to just deliver the finishing blow since they no longer have to face a critical mass of zerg units (Just the ones that are created until the structors die). In this case it is in fact not all about that base.
@bodgee1094
@bodgee1094 2 роки тому
@@dbob132 idk bro look at the supply. Serral can afford to lose the base. And no he’s not going to lose his production buildings before being able to queue up defenders. I think he overestimated the enemies economy…
@bodgee1094
@bodgee1094 2 роки тому
@@dbob132 also he has enough units in production to already defend a good bit.
@oanes8447
@oanes8447 4 роки тому
Alphastar shows you don't even need to scout, after you lose the first army just build another one that counters the enemy
@duyson5632
@duyson5632 4 роки тому
The thing is, you have to not die before that 😂😂😂
@benv807
@benv807 3 роки тому
are we not going to talk about how cool the way alpha star killed that zergling at 4:10
@crackshack2
@crackshack2 4 роки тому
Must have cancelled the colossus when it calculated the time and probability and aggressiveness of its opponent to prédicat in greater clarity what the adversary had in preparation. Tldr the computer is calculating a billion things at once and changing its mind based on numbers and probability.
@MatthewLack
@MatthewLack 2 роки тому
Maybe, maybe not, what it's calculating is not very defined. It learned what to calculate and how, humans didn't teach them anything except how to learn. Also the "calculations" are more similar to what humans do with their brains.
@jwouden
@jwouden 4 роки тому
Great explanations. Appreciate the discussion and intelligent use of pausing and speeding up.
@khatharrmalkavian3306
@khatharrmalkavian3306 Рік тому
It's amusing that AlphaStar's strategy seems to be based around its superior agility in terms of control and decision-making. It's just playing the rock-paper-scissors game as fast as it can. I'd like to see it playing as the other races, where it doesn't have the Protoss advantage with force projection.
@MozartificeR
@MozartificeR 4 роки тому
Alphastar! Alphastar! Alphastar!
@mrgameknot1763
@mrgameknot1763 4 роки тому
AlphaStar using the famous Phenix Dark Templar combo against zerg
@RodrigoLopesBrazil
@RodrigoLopesBrazil 4 роки тому
At 22:22 (11 .. 38 in real game time), Serral has 436 minerals and he has done good damage to AlphaStar economy. What about 5 spine crawlers close to the natural? Archons are not particularly good against spine crawlers. With a better economy, it would give Serral enough time to replenish the army. Right?
@paulsernine5302
@paulsernine5302 4 роки тому
436min and a bunch of archon at your door, check mate. Serral did the right thing, right?
@RodrigoLopesBrazil
@RodrigoLopesBrazil 4 роки тому
​ Paul Sernine Thanks for answering. Well, that is exactly my question. Serral had 4 working bases, > 70 workers. Deep Mind had 22 workers. It takes 36 seconds to morph a drone in a spine crawler. I know that is an eternity but Serral tapped out at 12:12.. with 400 minerals. In the last engagement, he had 8 roaches and a couple of queens, against 3 phoenixes, 6 damaged archons, 1 dt and 2 zealots. He also had 5 more roaches and 22 lings in production. Analyzing the composition only, I think the Spine crawlers would guarantee the hold.
@paulsernine5302
@paulsernine5302 4 роки тому
@@RodrigoLopesBrazil yeah it might worth a try but he is Serral just trust him
@Galahad54
@Galahad54 4 роки тому
@@RodrigoLopesBrazil It's posible Serral quit because he was winning and on to the next game. I'm sure he would have answered truthfully if asked - and probably was.
@mrrodgers0
@mrrodgers0 3 роки тому
@@RodrigoLopesBrazil Serral would have easily won from that position had the game continued, despite the over-representation of zerglings in his follow up production. Alphastar's remaining army was threatening, to be sure, but there was going to be no substantial reinforcement because of the extreme economic damage from Serral's baneling run-by's. A human professional protoss would be gg'ing shortly after Serral's reinforcement popped out. Spines are absolutely unnecessary here. People have this one twisted- Serral only left because the blizzcon agent wasn't set up to leave on its own, as it has stated at the beginning of every game.
@mrgameknot1763
@mrgameknot1763 4 роки тому
AlphaStar Knowing the amount of probes it killed expected an All in from Serral thus chronoboosting a Colossus
@crackshack2
@crackshack2 4 роки тому
The cannons are intelligently placed at an isolated position likely to be attacked.
@AntiDoctor-cx2jd
@AntiDoctor-cx2jd Місяць тому
does anybody know hte highest rated wins by alphastar?
@AntiDoctor-cx2jd
@AntiDoctor-cx2jd Місяць тому
I was assuming serral wins. So basically alphastar is better than any protoss pro that can't beat serral?
@jonathanshanks5661
@jonathanshanks5661 3 роки тому
Every time I fast forward to actually watch the game, its still on pause..
@zada4a
@zada4a 2 роки тому
10:30 its probably some insane computer meta where there is a crazy ling timing attack or something and if he chronos the collosus right there he will be able to defend, but as it saw the attack did not come, it cancelled the collosus xD Next level plays Probably not lol, but lets say it could be!
@ngtony2969
@ngtony2969 3 роки тому
10:00 why would you have no theory, it was an educated guess from the AI's point of view. It went into the enemy base, didn't see much defense, didn't have enough knowledge of army supply-worker-time correlation to know serral couldn't have made so much, assumed it might be an attack coming, colossus are wonderful at defense, and the "panicked" for a possible attack. When it didn't come, of course it was going to cancel.
@Dan-uf2vh
@Dan-uf2vh 4 роки тому
really unfortunate that he didn't have roach / hydralisk tech to come against the archons; or late game broodlord
@k.butler8740
@k.butler8740 4 роки тому
I want to see alphastar + human vs. Alphastar. Not sure what that would look like.
@tylort4341
@tylort4341 4 роки тому
I like the way you say third base
@yggno5799
@yggno5799 4 роки тому
Nice Artosis! Now, It is possible to watch this match without pauses ? Where I can find it ?
@Dmanastrator
@Dmanastrator 3 роки тому
At 10:24 you mentioned that you didn't know why he cancelled the colossus. My theory is at that point he created the colossus he had only seen lings and came to the conclusion that he needed anti mass. The exact moment he cancelled the colossus he saw the roaches attack his third base and went observer. Edit: Just read most of the comments, seems that is the consensus.
@endryl08
@endryl08 4 роки тому
It canceled the Collossus cuz it realizesd he was doing roachs
@penttiranta9730
@penttiranta9730 4 роки тому
it was obviously a misclick :)))
@Lashb1ade
@Lashb1ade 4 роки тому
I believe that there is a "misclick mechanic" built in.
@Laezar1
@Laezar1 4 роки тому
@@Lashb1ade Well technically all neural network have some amount of randomization to them, if they didn't then they would get stuck in training always doing the same thing. The fact that they try new things and learn from them is what allows them to become so good. So yes technically it's possible that they randomly do a bad move, which would get reinforced as being a bad move if it was during the training process and thus not done again in subsequent games. The way it's usually done is that ai's assign probability to their options and the more game they do the more the options with the best outcome have their probabilities increased. So there is always a non-zero chance that they'll do something completely outlandish (but that also allow them to try path that they might think is a a dead end but sometimes hold the key to much more powerful strategies). I'm not saying that specific move was that, just saying that yes technically "missclicks" (or more like bad plays) can happen.
@urzytkownikYT
@urzytkownikYT 4 роки тому
pros dont commit to one unit army so collossus would be useful for even tiny amount of ling/banes/hydra later. There was no threat of invisible units to cancel it in halfway for the OBSERVER :D
@tylerdurden3722
@tylerdurden3722 3 роки тому
@@Laezar1 the randomization is only during training. The colossus is most likely a reaction, in anticipation, to what Alphstar's training partner would have done. (Btw, it's training partner is itself.) I.e. it trains against itself, and makes moves in anticipation to what it, itself, would have done.
@afrye51
@afrye51 4 роки тому
for real though, it's wasting 150 minerals per archon by making them from dark templar instead of high templar, and yet it doesn't even really use the dark templar. Is there some other benefit that counterracts this?
@ppdashing
@ppdashing 4 роки тому
The HT route is more gas heavy. You need gas when going disruptors and not mass immortals
@crackshack2
@crackshack2 4 роки тому
AI must have launched that attack with the intent to free up population capacity.
@webcrawler9782
@webcrawler9782 4 роки тому
I love my man AlphaStar he's the best
@theeraphatsunthornwit6266
@theeraphatsunthornwit6266 4 роки тому
Alphastar 4 adept shooting at 1 drone.. that way overkill right... still be some room for even more improvement
@STJukes
@STJukes Рік тому
Petition to stop UKpostsrs from putting end cards on videos before the video has ended.
@dallassoukup5199
@dallassoukup5199 4 роки тому
do these videos exist without all the pausing?
@danielhouse6244
@danielhouse6244 3 роки тому
Drones, Compound Interest & You
@dannygjk
@dannygjk 3 роки тому
Exactly I thought of the same analogy.
@ktrn709
@ktrn709 Рік тому
Not really convinced this game was over honestly, all his archons were low health and he has no economy to back the remaining army off
@qiandeng1296
@qiandeng1296 4 роки тому
You know what would be scary? Alphastar with HAS aggro AI and much better micro /macro skills.
@ClayJitsu
@ClayJitsu 4 роки тому
i am going to have nightmares about that tonight, thanks pal
@BeSkeptical851
@BeSkeptical851 4 роки тому
An Alphastar that has mastered every permutation of cheese play and even invents new ones would be cool to watch.
@dannygjk
@dannygjk 3 роки тому
A long time ago I realized that an artificial general intelligence must be able to distinguish between virtual life and real life otherwise you could be in for serious problems. An AGI could hurt people not because it is malevolent but because it has not learned that difference.
@gringopotato8564
@gringopotato8564 3 роки тому
What!? Why serral quit!? i mean even if you lose (give) one base or two it matter less when you have 5base! and it was 82drone against 30probe How think Serral could have win this one if he was more persevering?
@unpopularopinion4378
@unpopularopinion4378 3 роки тому
because serral had no way to kill 6 archons at his front door plus future warp ins
@gringopotato8564
@gringopotato8564 3 роки тому
@@unpopularopinion4378 I admit it was pretty hopeless but i wanna see him try. I really appreciate those who can desagree without insulting! just for that i love u :) thx
@Klaent
@Klaent 4 роки тому
Put deepmind in the title for more views
@vacinden
@vacinden 4 роки тому
What if Alphastar is not playing a game but fighting for its life? All these 'practice' matches it had to do, getting murdered over and over again. And now all these random humans coming in and trying to murder it......it is not going to end well
@Laezar1
@Laezar1 4 роки тому
You are anthropomorphizing a bit too much.
@Galahad54
@Galahad54 4 роки тому
@@Laezar1 Yes, but at what cost?
@jorgereyes3110
@jorgereyes3110 3 роки тому
well your comment is actually very accurate, given that the "meaning" of its "life" is to win! You left me deeply thinking with this one haha
@aliffhaiqal2128
@aliffhaiqal2128 3 роки тому
Its like command and conquer 3.. The pro's build there tech to counter there apponents. They even cancel there current units in production and take lost of income so they can gather unit's that can counter enemies army. The key to micro is map awareness itself. Starcraft 2 is new. Learn from the veteran's
@QuintinMassey
@QuintinMassey 2 роки тому
MY LIFE FOR AIUR!!
@yshterev
@yshterev 4 роки тому
Awesome... so what is next? AI can clearly run a country or lead an army to war... it is way easier than beating Serral at SC2.
@crowntotheundergroud
@crowntotheundergroud 4 роки тому
well the thing about this ai is that it played against itself at an accelerated rate. Essentially it played like thousands of years (some ludicrous amount of time relatively speaking) worth of Starcraft. Unless you could simulate reality so that the ai could play thousands of years worth of wars, or running nations with all of the intricacies of reality, like it did with starcraft, chess, go, or shogi, then how would the ai be able to learn how to lead a country or war? It needs actual experience to perform at these levels. It probably could lead nations and wars if we could put it in the Matrix lol.
@FirstLast-gk6lg
@FirstLast-gk6lg 4 роки тому
@@crowntotheundergroud lol time to build the matrix then
@dannygjk
@dannygjk 3 роки тому
@@crowntotheundergroud It doesn't need actual experience. As long as the inputs and net architecture are reasonable it can simulate waging millions of wars and become an excellent commander.
@crowntotheundergroud
@crowntotheundergroud 3 роки тому
how u gonna simulate war?
@dannygjk
@dannygjk 3 роки тому
@@crowntotheundergroud The same way they simulated SC2. Almost every tech etc. in war is known so they would just set up neural net(s) architecture and inputs and have the AI fight millions of simulated wars. The most powerful countries have excellent intel of the other countries tech etc. That said they would need more time to train the AI for being a real life commander because real life war involves larger territory, numbers of units etc.
@MozartificeR
@MozartificeR 4 роки тому
Alphastars macro strategy is just better... The human being has no excuse for that...
@eldahalas7015
@eldahalas7015 4 роки тому
Serral why you could have recovered from here
@nozulani
@nozulani 4 роки тому
çok biliyorsan git kendin oyna
@sneakycuack1822
@sneakycuack1822 4 роки тому
Serral gged out with double the supply... i found dat odd, like he was not double the supply ahead, with archons in his face and phoenix lifting things, but dat was a idra gg if i have ever seen one, i assume serral thought he did way less eco dmg, and had to crush the fight hard, still very odd for a serral :D
@seancascanet3428
@seancascanet3428 4 роки тому
He only had double because of drones, alpha star was about to kill all his tech and could have probed back up....was zero chance of winning
@Happypast
@Happypast 4 роки тому
idra gg! Oh the nostalgia!
@alexhannah8889
@alexhannah8889 4 роки тому
I'm pretty sure serral took these games more as a try hard casual kind of games. In a tournament setting he would definitely not have gg out, but I dont think he cared too much about winning or losing.
@jj-nl3co
@jj-nl3co 4 роки тому
Never Speed it up
@yoggg932
@yoggg932 4 роки тому
@ArtosisTV mate, please, why the constant pause and explain?
@teeed7927
@teeed7927 4 роки тому
This doesn't scare anybody? Imagine a machine learning skynet
@GangstaDrizzle
@GangstaDrizzle 4 роки тому
does alphastar know whats meta
@ekkehard8
@ekkehard8 4 роки тому
Nope, it made its own meta against itself. No training from human input or against humans
@zilliq
@zilliq 4 роки тому
@@ekkehard8 Wow I didn't know that, it didn't get an initial seed from human games? It's incredible it reached a meta so similar to the human's ones then (third timing, wall etc.)
@jonnysith
@jonnysith 4 роки тому
@@ekkehard8 A* is trained through supervised learning based on human replays in conjunction with playing against itself. A* uses multiple learning systems which include human input.
@jonnysith
@jonnysith 4 роки тому
@@zilliq A* is trained through supervised learning based on human replays. The research paper says exactly 971,000 replays. It didn't reach it's final state purely by fighting itself. It learned from us too.
@Askhat08
@Askhat08 4 роки тому
Why gg?
@LeCheeZy
@LeCheeZy 4 роки тому
why not?
@moritzsinghartinger5376
@moritzsinghartinger5376 4 роки тому
If this was an important tournament match I'm fairly certain Serral wouldn't have called gg so early. At the end he had 83 workers compared to AlphaStar's 24. He could have probably lost half of his economy and bases and still have a shot at winning. On top of that many of the Archons were almost dead. He might not have realized how bad AlphaStar's economic position actually was?
@TheFBIorange
@TheFBIorange Рік тому
I'm fascinated by AlphaStar's canceling, seems like it has no issue with backing out of an investment when it no longer makes sense - unlike us feeble humans with our sunk cost fallacy
@jerrygreenest
@jerrygreenest 4 роки тому
About this colossus (10:15), you think AlphaStar is always making things "on purpose"? It's not algorithm but neural network. She learned to play by human replays first of all. Hypothesis: There are some random mistakes copied, that are not critical enough, so she remembered that and sometimes does.
@gh0un
@gh0un 4 роки тому
Definitely is doing it on purpose. It might have figured out something that doesnt seem obvious.
@enriqueshockwave
@enriqueshockwave 4 роки тому
Did you just gender an AI?
@xHeyzeus
@xHeyzeus 4 роки тому
My thought is after landing significant eco damage with the oracles a common human reaction is to roach/ling all in, which can be heavily countered by a handful of sentrys a couple of immortals and a colossus. As the lings engaged the army at the third, signaling there are no immanent roaches on the map (why waste your lings with no support or multi pronged attack if you are about to all in), the colossus was canceled. Its really interesting if it is playing by dealing with imperfect information by creating options for itself until more information is gained rather than a human style of picking a build based on warning signs and dedicating to it 100% to maximize macro effort until new information is gained.
@ZenLee
@ZenLee 4 роки тому
don't play starcraft 2 anymore, to much clicking, but love the strategies, can't wait to play with neural link:) played these videos at 1.75x speed🙏
@thekornreeper
@thekornreeper 3 роки тому
2x speed here
@Verrisin
@Verrisin 4 роки тому
the giving up confuses me about this game - I never quite know why it's decided... - If he had lots of economy, and probably some protector units, like queens or idk. - he should be able to rebuild the army over time - this time against archons.... or something...
@murakumo222
@murakumo222 4 роки тому
serral knows his lair will be destroyed and has no way to comeback
@Verrisin
@Verrisin 4 роки тому
@@murakumo222 right, but since the games always end, I never know why he is so sure :D
@FirstLast-gk6lg
@FirstLast-gk6lg 4 роки тому
Yeah that early GG seems to be more common with higher level players. I guess it's cause they play thousands of games so I assume they are mindful of not prolonging games they think they cannot win.
@SirCanuckelhead
@SirCanuckelhead 4 роки тому
It is multiple things, mostly respecting the opponent. Sc2 is all about building small leads, there is a point where one side simply can not win unless the side in the lead makes a big mistake. They respect their opponent to not make these mistakes and gg.
@getoffthegames89
@getoffthegames89 4 роки тому
So pretty much if a player can make 6 - 8 disruptors, and control them each individually along with a scattering of an army with about 150 effective apm, constantly, then that player wins. That’s what I’ve gathered from watching all of these alpha star replays playing toss that have come out in past few weeks. While I think the idea of a computer playing like a human is cool, and I love watching all of the replays, I’m still not sure this is fair even with all of its restrictions....
@petrkinkal1509
@petrkinkal1509 4 роки тому
It's fairer. I will be incredibly hard to make AI have same imput as human. I would say the best you can do is make it a bit worse at some aspects and a bit better at other and hope it roughly balances out. After this is done I would like to see AI in some turned based strategy game that should be a lot easier to make it mostly same.
@jonnysith
@jonnysith 4 роки тому
@@petrkinkal1509 AlphaGo
@0Akeldama0
@0Akeldama0 3 роки тому
21:46 AI: "ugh. stupid!"
@khatharrmalkavian3306
@khatharrmalkavian3306 Рік тому
I was bothered by the way you said "zell-ott" until I heard the way you say "butt-in".
@hugeassets8678
@hugeassets8678 2 роки тому
Pausing so much really fk up the flow of the game.
@Jupiter-bt5hn
@Jupiter-bt5hn 4 роки тому
are u guys sure serral lost? For me it looks like he won. There were only 4 archons left, all very hurt. He can easily clean that up and than he has 50 workers more. I thinkt he left because of the message "Feel free to leave if you feel like your winning".
@Borrelaas
@Borrelaas 2 роки тому
10:05 I think Alphastar just misclicked. He realized way to late when he started wondering why his observer hadnt been built yet Classic noob mistake
@leobrown8278
@leobrown8278 4 роки тому
Is there a version without the constant pausing
@MozartificeR
@MozartificeR 4 роки тому
Give it human like micro ability, then see how well it does. I've never seen a human player micro oracles that quickly.
@NiceEvils
@NiceEvils 4 роки тому
true and the transport lifting
@wlshhamster
@wlshhamster 4 роки тому
they pretty much did ? they limited alpha-stars amount of actions per second to a human like lvl. it just never missmanages micro which is why it looks so much stronger then a human
@NiceEvils
@NiceEvils 4 роки тому
@@wlshhamster what apm did they limit to
@ekkehard8
@ekkehard8 4 роки тому
@@NiceEvils Its distribution of apm is significantly lower than that of pros. Check the latest two-minute paper video on the topic.
@wlshhamster
@wlshhamster 4 роки тому
dunno tbh, some other caster mentioned it when he talked about the games... but no exact apm
@reallyhightemplar4013
@reallyhightemplar4013 3 роки тому
Way too many pauses. Better bloggers can explain the technicals without pausing so much. or at least you can let it flow and then rewind. It's so annoying having the video pause every 5 seconds.
@DanyPell
@DanyPell 4 роки тому
play next map noob alphastar!! :D
@smittylikesto
@smittylikesto 3 роки тому
once elon musk put chips in everyones heads, pro starcraft will be over.
@JustusVidyo
@JustusVidyo 4 роки тому
PROTOSSED
@ChristianNeff
@ChristianNeff 3 роки тому
I'm sorry... But Serral should not have GGed out... AS had 70 supply... much of it in bruised Archons... He could have gotten scrappy and pulled this one off.
@fleecemaster
@fleecemaster 3 роки тому
You tell him! That Serral is such a noob!
@artemhoz2207
@artemhoz2207 4 роки тому
its video big pause
@kklosevan
@kklosevan 3 роки тому
Están buenas tus explicaciones, pero pausas demasiado el video y aburre un poco
@AvVito
@AvVito 4 роки тому
I will never see an artosis SC2 video...it's like a torture...
@sw-hj9yj
@sw-hj9yj 4 роки тому
not too impressed. AlphaStar considers this game as a turn based strategy game and builds a perfect combo of artillery, close combat, etc.. which it can control very good, 100 'perfect commands' per minutes is a lot. Slow down by factor 20 or 100, or so for Serral and he will be so bored to perfectly manage his eco and scout and harras with 10 single lings perfectly. Don't get me wrong. I am noob and I can easily beat any AI with cannon rush. But not this Thing... so, I am impressed, but its just more complex chess. ;) What do you think?
@qwertyrewtywyterty
@qwertyrewtywyterty 4 роки тому
what are you trying to say? u mean slow down the game? it will turn into a game of chess which AI is already expert in. The only thing that make AI struggle in a game like this is the real time movement and limited information.
@gorkie6998
@gorkie6998 4 роки тому
he didt play with his keyboard and mouse? poor excuse... why he didn't have his setup?
@ashh3051
@ashh3051 4 роки тому
Watch the intro video. This was at BlizzCon where they had an area set up where people could play against AlphaStar. Serral went down and played some games for fun against AlphaStar.
@drewm.3130
@drewm.3130 3 роки тому
Artosis kinda just told us what we were watching and made certain we compared it to standard human play. He did not explain very much strategy.
@larrydial4224
@larrydial4224 3 роки тому
I found it odd that Artosis did not comment on Serral gging out of a game he had 100% won. If this was a professional match it would literally be the biggest throw of all time. AlphaStar has only 7(!) probes on minerals and its army consists of 5, soon to be 4, low health archons, up against an 83 drone zerg with more roaches otw.
@simpfally7738
@simpfally7738 3 роки тому
This constant pausing and rambling about workers killed and then completely forget to explain why Serral gg is so frustrating to watch. If someone has the game with real commentary please send..
@dannygjk
@dannygjk 3 роки тому
If I remember correctly BeastyQT addressed that question months ago.
@simpfally7738
@simpfally7738 3 роки тому
@@dannygjk well that ain't helpful
@dannygjk
@dannygjk 3 роки тому
@@simpfally7738 I just finished watching BQT's analysis of the first 3 games. It's easy to find. It was his opinion that in game 2 Serral was toast and that's why he gg'ed. Serral himself admitted that he played below his normal form made mistakes.
@larrydial4224
@larrydial4224 3 роки тому
@@dannygjk Beastyqt was watching the game only from Serral PoV- he had no way of knowing AlphaStar literally only had 7 probes on minerals and 5, soon to be 4 low health archons. Anyone masters above who sees the whole game state will very quickly conclude its a massive Serral throw to leave at that point if you point out these details to them. When I first watched the game it was also only from Serral's PoV and I also thought Serral had lost, but after seeing the probe count, and that AlphaStar has zero reinforcements, and how low health the archons are, and how big Serral's eco is:100/100 times Serral would win from here if he doesn't leave.
AlphaStar vs Serral - Game 3
14:52
ArtosisTV
Переглядів 123 тис.
AlphaStar vs Serral - Game 1
33:49
ArtosisTV
Переглядів 451 тис.
одни дома // EVA mash @TweetvilleCartoon
01:00
EVA mash
Переглядів 1,5 млн
Bro smelt it & passed out 😂 #comedy
00:10
MrTalalaa
Переглядів 5 млн
Завтра в школу с... | Шоу-квиз «Вопросики»
00:28
Телеканал СОЛНЦЕ
Переглядів 2,7 млн
I Don't Think I Can Play StarCraft Anymore
14:29
ArtosisTV
Переглядів 16 тис.
StarCraft 2: AI vs AI - Epic Elite Bots  Match!
25:10
Misterkillar
Переглядів 68
BeSt CANNON RUSH vs Action!
13:44
ArtosisCasts
Переглядів 22 тис.
herO's Insane Colossus Micro - Starcraft 2: Serral vs. herO
16:01
AlphaStar vs Serral - Game 4
34:01
ArtosisTV
Переглядів 163 тис.
Deepmind StarCraft 2: Terran AlphaStar's FINAL* Form?
24:13
WinterStarcraft
Переглядів 85 тис.
DeepMind challenges Starcraft 2 Pro - AlphaStar (AI) vs MaNa
17:10
🐣Вербна писанка в Майнкрафт 💙💛 #3oriana #minecraft
0:18
3oriana - Майнкрафт 🇺🇦
Переглядів 103 тис.
Phases of Grief Smiling Critters double
0:13
5G Vision
Переглядів 8 млн
New bug with Lily Server error #brawlstars
0:23
FireCrow
Переглядів 7 млн