Alphastar has very little experience in the late game and it shows. I've never seen it make a viper or infestor. 200 years of training got it through the early game into a Roach ravager all in mid game and it stopped there.
in 200 years worth of gameplay two AI deep minds will never reach end game as they will end the game by brute forcing each other by mid game do to the late game unit costs vs effectiveness. The other thing Alpha star likely wouldnt be able to deal with is some sort of contain early game such as a bunker rush or cannon rush where they are forced into a one base situation with no ability to expand or gain map awareness.
It doesn't seem to make any zero-damage casters, probably because the attack value is 0 unless you know where to put a storm or how to pull the correct units. So if it tries to make a unit like that and doesn't know what to do with it, it tends to lose, so it learns not to use them. A human would have to program in how to use a unit like a viper and then it will demolish Starcraft because Zerg casters are op.
@@GuruEvi I dont think that is true.. I think the fact that it just hasnt had enough experience with the units is the problem and it can only get that by playing more. In reality it has used units with 0 attack.. Disruptors come to minds it has used those on multiple occasions. So it can eventually figure out how to use said units that technically have 0 attack. Problem is it going to take more then 200 years against itself, surprise - surprise playing against yourself doesn't necessarily mean you will get better at the game faster then someone who ladders vs other sentient entities.
It was an hollywoodian impression of what a struggle should be. The MasterPlayer was almost defeated. Comes up with an idiotic idea and wins against all odds.
AlphaStar: By this time the opponent is always dead. So I'm done. I just don't understand that my units are still somehow dying. Probably a bug. MasterPlayer still alive: Am I a joke to you?
Another example showing the limits of this ai. I love how sc2 can show instances where the ai sees things people dont, and other times where the ai misses something.
@@thenarrator1984 I think that no matter how many games it plays, alphastar might simply be unable to grasp certain relationships or concepts in a game of starcraft 2. Maybe it will be so miraculous that it is better in every way. But to me, sometimes it deals with things in a unique but effective way. But sometimes it seems like perhaps it doesn't understand something even after seeing it, such as blocking harrassment with one unit at base entrance. I think the ai may never realize that placing a unit next to buildings there, moving out of way when necessary, could prevent worker losses. I think it's too complicated. The ai learns by playing, changing things up, seeing which changes seemed to provide better results. But as amazing as that can be, I dont think it has the intelligence to think of using one unit to block a base entrance. And I dont think it will ever do it on accident, so it cant realize it's good by experience. Really interested to keep watching and try to fuss out more where this ai gets ahead of us, and novel "ideas" it has, and also what situations it cant handle well. And why.
yeah agreed, even if its not perfect, SC2 is the best spectator game of all time. I don't think the overall game would be as big as it is if it didn't have such good spectating and easy to share replays. I wish other devs just unilaterally decided that their games needed to have good replay tools and put them in. You never know what players will do with tools until you give them... (rip blizzard, nobody can withstand the power of activision suck-field, not even the best people at Bliz. )
@@sciencewizard8805 It doesnt think, all it does is pattern matching and giving value to each pattern. But its highly limited in ammount of patterns it can process, because modern PC can't keep up with brains in the amount of patterns to compare to in a single moment.
@@sciencewizard8805 It doesn't think long term strategy yet. that's why it couldn't pick off the voids when clearly had the chance. it understands basic conceptual strats like countering units and blocking base entry early game and resource gathering, and some basic combat micro.
if ai takes over the world, it's going to just mine out every resource and build a ton of units and stand around while we starve... and it wont have any idea why we all died
I agree with you on games like this. But I'm getting bored with laughNgamez videos, it seems like he mostly posts the games where alphastar loses in some derpy way. I'm more excited by the strange ways a.i.(or humans) takes risks and adapt. Mostly in videos other folks are posting. I'm not rooting for alphastar but the only thing interesting about this is that the human's unadaptive strategy can win off two bases. Its more a show of the a.i. failure than the human's strategy. Hat's off to the Master player tho, their micro was good and they found a good strat vs a.i.... for now... bwuahahahahaha. What do you think Carl? Do you like watching the a.i. get stumped? (As a human I can still appreciate it ;) )
@@BrendanCassleman Hehe, no, to be honest with you, I'm kind of rooting for the A.I.. I guess it's the 'Tech-nerd' in me that wants it to succeed. Games like this are just frustrating to watch imo, and it shows that A.I. still has a looong way to go! I wasn't really impressed by the human either, he just got lucky and found a bug/leak. Furthermore, I want to see the A.I. come up with some innovating/unexpected strats that we can learn from, rather than brute forcing with micro.
@@BrendanCassleman I'm with you. i find it strange that i watch other youtubers post videos of Alpha Star beating professionals, and LNG has nothing but videos of it losing to regular ladder players.
yeah and then saying "all those hours I saved not getting oil changes, I was in the office working overtime making more money so I WIN!" (and then the car breaks down right when he needs it). Like it makes perfect logical sense right up to the point where it doesn't :P
Yeah, no one would've blamed him if he had given up but maybe he had a hunch something about this enemy was off, not sure if he realised it's AS or just some drunk guy xD and if he realised it's AS, he maybe just tried some unusual tactic (mass void) to overwhelm AS at blizzcon, one of the few people winning consistently against AS without cannonrush was beastyqt, who just tried things AS probs doesn't know (i think he went 7/10), things like mech vs protoss
@@eadbert1935 hey I mean it's very possible AS plays more human-like than typical AI's but I'd say it's a very peculiar style a human wouldn't really play like. I'd be pretty suspicious if someone GM ranked played me and had very flawed walling or a terran that just loved banshees way too much, etc. I'd think AS is well-known enough at that high level that maybe they'd be able to figure out who it was.
Alpha: _"There are two buttons. The one labeled 'surrender' leads to the menus and a ranked defeat. The one labeled 'resume' leads back to the match, to the Zerg, and to getting contained on two bases. As you've adequately put, the problem is choice."_
I think what's going on here is Alphastar is constantly reevaluating the likelihood that it will win if it attacks, and it's never clear that's the right thing to do. It probably has a pretty good idea what would happen if it attack clicked the base, and most likely all the corrupters would die/ most of the void rays would die, but it's not clearly an improved position. It doesn't necessarily have a good memory, or thought of, "well, I have to do this at some point, so it's now or never". Likewise, when the protoss has the mothership, Alphastar obviously knows it can cloak units, but I don't know that it has a great idea of what exactly is being cloaked.
Strangely alphastar at a certain point pulled back when it should have gone in and finished the protoss player. The Protoss player however played the perfect game against the AI and that is to wait it out and weather its best moves and then let it get confused by having not won yet.
void ray is like the only aerial counter unit to corruptors especially with how efficient they are in a big fight where corruptors waste a lot of shots due to high damage/slow attack speed bursts while void rays waste no uptime at all they are however extremely weak to zerg's balanced caster units fortunately the AI didnt discover those yet
Either that or actually threaten the protoss base somehow. Like Mass mutalisk harassment instead of flying the corruptors over the protoss base without engaging the protoss fleet.
Thanks for all the content in 2019, your commentary and tone is top-notch and AlphaStar is just a riot to watch as it annihilates and also derps. Happy 2020!
What the heck was AlphaStar doing in this game? First it makes the perfect counter to Carriers. Then it flies them 3/4th of the way across the map while the Carriers tear his base apart at 10:22. Then it turns them around and flies back across the map to clear out the Carriers at 11:20. Then it flies them back to MP's base and actually wins the game at 12:00, but then turns around and goes back home after getting a CLEAR advantage both in economy and army. Then it spends the next ten minutes flying it's Corrupters in and out of the growing airtoss fleet, giving MP tons of free shots without accomplishing anything. - This has to be the hardest throw I've ever seen in any game. This game was over, done and won by AlphaStar at 12:00, but for some reason it just did not want the free win.
@LaughNgamez Are you sure THIS is Alphastar? The unit control and reaction times are very different. Usually, Alphastar instantly moves all of its units as soon as they are attacked but this doesn't have that split-second reaction time. The lings in the natural do a hold command play which I've never seen alpha star do. Alphastar also does a preemptive counterplay to the air units by building corruptors which I've never seen it do before. Usually, the AI just builds a varied unit comp and the "bad" units die off so that the unit comp grows into a counterplay. That's what I've seen Alphastar do before. This looks very different.
@@dionbridger5944 Yes, though it's not like 300 is much fairer... The effective APM (minus the spammy clicks) isn't usually that high... also the AI can technically front or back load it's clicks as needed in a 5 second window so it does have the ability to pull off much crazier micro than real pros so long as it isn't sustained over a lengthier period (and due to things like blink cooldowns it wouldn't have any need to do more anyway). Apparently they get 22 "actions" in a 5 second window, but several SCII actions only count as a single AI "action" it is allowed to select a unit, activate an ability, and set the target location all as a single "action."
As somebody who plays a LOT of Voidray / Mothership vs zerg.... Protos made one error over and over... Overlords are always priority targets over damn near everything else. Always pop the vision things first.
Such a satisfying game to watch a victorious game by the human player. I think this is the answer, for now, where the AI just does not know how to approach mass units.
This is likely due to Alpha Star having very little experience in late game because honestly, its opponents likely never get that far. It likely nearly always demolishes it's opponents in early game.
Which alphastar is it? there are multiple agents they use for various ranks. I guess this is the master-ranked alphastar? -- is this the old one who can exceed human level APM and see the whole map? ( That one usually did mono-battles, with crazy micro )
Lol this battle was brutal, protracted and twisted like Nazi Germany vs. Soviet Union. :DD Best SC2 duel what I've seen so far. MP deserved that victory.
Train it a little vs all in air and this will never repeat. Alphastar would crush this guy if it had trained against it prior. Anyone could win that late game.
You can actually see the AI logic fallacy when it sees voidrays: the enemy is going Air, so I will too. Which it does. Only the damned game makers didn't give the corruptor a ground attack, like the voidray has. So it makes roaches for ground attack. Only the damned game makers didn't give the roach an air attack. It's basically like a 2 year-old in its decision making. Compare this game to its Protoss version, where it makes mass-stalker. The AI wins easily, since its units shoot both air and ground.
Alphastar simply shoud have mass expended when toss was locked into 2 bases and after that stack up money, push all 45 corrupters into the toss army, build another 45 corrupters and end it. Theres rly no need to tech vipers or something - eco wins.
Stupid AI main focus is the mothership. No abductors. Protoss learn in the first time. He back away just his mothership quicker second time and AI losses alot of corruptors. The human was clever with the tempest. If he could make more to one shot the detector it would be GG faster.
I don’t see a derp. “When you are ahead get more ahead” That’s exactly what alpha star did. Toss tech’ed up and wrecked Zerg should have traded out more. But Losing units to trade eco isn’t alphastars play style. ...No derp
Should have focused on tech more I feel as well. Even armour just to balance out that +3 on voids. Also probably a bit more harassment, even pick up a pylon or two with corrosive.
Getting so frustrated at this AI while watching this, thinking to myself that it clearly has not learned that having triple the number of corruptors versus the other air army means they can just a-move in, batteries or not. Seeing it constantly give away its corruptors by moving nearby and pulling back just made no sense to me. Would love to see it's willingness to commit be more decisive in the future as it keeps learning.
I don't know if they actually had enough corruptors to win that. Balling up tightly and one shot ting individual units might actually have been the best strategy
@@emissarygw2264 If the army sizes were closer that would be the case, but at about 15:18 it is pointed out that Alpha Star has 35 corruptors. It could have EASILY beaten a dozen voidrays and two carriers, but simply could not decide to commit, and every time it went into range of their attacks but ran instead of fighting back, it lost units for free, which let the protoss catch up.
Alpha star never brought detection with it. The most likely scenario is it would see the mothership go in for the kill get targeted by cloaked units and then retreat because suddenly it no longer has a 90% victory condition do to unknown variables. Their Ai from the beginning has always functioned on a % victory condition where it would predict the chance of winning an engagement before engaging. So if it predicts it can secure the win in the engagement it will engage. If it then starts taking more damage then predicted it will retreat as engagement parameters have changed. That why it constantly dived then ran away then circled around its base till it recalculated its likelihood of winning. I'm pretty sure it counts shield batteries as a factor in the engagements as well even if the player never used them.
@@urstaxfetish1206 I am absolutely certain that is the case later on, but there was a point at which it could undoubtedly just run in and kill the other player for that time stamp I used. 15:18 target down voids with corruptors while targeting batteries with the roaches. It is clearly still learning but it will continue to struggle to secure a win in situations where it has to break through a fortified position because of this indecisiveness/excessive caution against certain defenses. Without actually having access to a log of its evaluations of the threat at each encounter, all we can really do is guess at its choices and hope it gets better.
Halal Milksheikh alphastar is dory from finding Nemo: 1- oh, a lone mothership, better take it 2- my corruptors are dropping like flies, better retreat and think again 3- repeat
Yes, Alphastar derped, but can we take a moment to applaud the masterplayer's decison making? He realized that he couldn't hold out indefinitely and took a risk by harassing the zerg over an overextension. I bet most players would've been afraid of losing, but he realized that if doesn't make a significant change here, he has basically lost anyways.
The thing is I also just lost 2 games to void rais, so I feel this alphastar behavior is pretty human. I had corruptors and hydras, big supply lead, but could not beat the voids if the player has income. Maybe vipers are needed.
The next logical step for AlphaStar at this point is to rage post on the forums about how OP void rays are.
Smash that Turing button
😂😂
hahahahhaha
err void rays not that op after hots tho, if alpha star rage, show him wol void rays
Alpha Star: I played 200 years of Starcraft 2 against myself....
Human: You ever see Mass VoidRay
Alpha Star: Mass what!?
Mass Void rays destroy Advanced A.I
*Laughs in Gold league*
its just a agent of the Training, not the Final Alphastar.
@@Fine_Mouche Do you know if we can see replays of the final agent?
@@sasha4598 game vs serral at blitzcon is the most advanced agent but can't tell if its final agent.
Alphastar has very little experience in the late game and it shows. I've never seen it make a viper or infestor. 200 years of training got it through the early game into a Roach ravager all in mid game and it stopped there.
in 200 years worth of gameplay two AI deep minds will never reach end game as they will end the game by brute forcing each other by mid game do to the late game unit costs vs effectiveness.
The other thing Alpha star likely wouldnt be able to deal with is some sort of contain early game such as a bunker rush or cannon rush where they are forced into a one base situation with no ability to expand or gain map awareness.
Not two AI, there are around 1000 Agents playing against each other...
It doesn't seem to make any zero-damage casters, probably because the attack value is 0 unless you know where to put a storm or how to pull the correct units. So if it tries to make a unit like that and doesn't know what to do with it, it tends to lose, so it learns not to use them. A human would have to program in how to use a unit like a viper and then it will demolish Starcraft because Zerg casters are op.
@@GuruEvi I dont think that is true.. I think the fact that it just hasnt had enough experience with the units is the problem and it can only get that by playing more.
In reality it has used units with 0 attack.. Disruptors come to minds it has used those on multiple occasions. So it can eventually figure out how to use said units that technically have 0 attack.
Problem is it going to take more then 200 years against itself, surprise - surprise playing against yourself doesn't necessarily mean you will get better at the game faster then someone who ladders vs other sentient entities.
It did not stop. It never needed to learn more than mid-game because human beings are so easy.
AlphaStar doing its best "Medium Difficulty AI" impression.
It was an hollywoodian impression of what a struggle should be. The MasterPlayer was almost defeated. Comes up with an idiotic idea and wins against all odds.
AlphaStar: By this time the opponent is always dead. So I'm done. I just don't understand that my units are still somehow dying. Probably a bug.
MasterPlayer still alive: Am I a joke to you?
Another example showing the limits of this ai. I love how sc2 can show instances where the ai sees things people dont, and other times where the ai misses something.
@@thenarrator1984 I think that no matter how many games it plays, alphastar might simply be unable to grasp certain relationships or concepts in a game of starcraft 2. Maybe it will be so miraculous that it is better in every way. But to me, sometimes it deals with things in a unique but effective way. But sometimes it seems like perhaps it doesn't understand something even after seeing it, such as blocking harrassment with one unit at base entrance. I think the ai may never realize that placing a unit next to buildings there, moving out of way when necessary, could prevent worker losses. I think it's too complicated. The ai learns by playing, changing things up, seeing which changes seemed to provide better results. But as amazing as that can be, I dont think it has the intelligence to think of using one unit to block a base entrance. And I dont think it will ever do it on accident, so it cant realize it's good by experience.
Really interested to keep watching and try to fuss out more where this ai gets ahead of us, and novel "ideas" it has, and also what situations it cant handle well. And why.
yeah agreed, even if its not perfect, SC2 is the best spectator game of all time. I don't think the overall game would be as big as it is if it didn't have such good spectating and easy to share replays. I wish other devs just unilaterally decided that their games needed to have good replay tools and put them in. You never know what players will do with tools until you give them...
(rip blizzard, nobody can withstand the power of activision suck-field, not even the best people at Bliz. )
@@sciencewizard8805 Yeah, for example it never gets to see the build grid so it can't check whether its walloffs work
@@sciencewizard8805 It doesnt think, all it does is pattern matching and giving value to each pattern. But its highly limited in ammount of patterns it can process, because modern PC can't keep up with brains in the amount of patterns to compare to in a single moment.
@@sciencewizard8805 It doesn't think long term strategy yet. that's why it couldn't pick off the voids when clearly had the chance. it understands basic conceptual strats like countering units and blocking base entry early game and resource gathering, and some basic combat micro.
Doomsday Prophet: "Fools! A.I. is going to take over the world and kill us all!"
Me: _shows this video_
Doomsday Prophet: "Oh, nevermind, we're safe."
if ai takes over the world, it's going to just mine out every resource and build a ton of units and stand around while we starve... and it wont have any idea why we all died
I agree with you on games like this. But I'm getting bored with laughNgamez videos, it seems like he mostly posts the games where alphastar loses in some derpy way. I'm more excited by the strange ways a.i.(or humans) takes risks and adapt. Mostly in videos other folks are posting. I'm not rooting for alphastar but the only thing interesting about this is that the human's unadaptive strategy can win off two bases. Its more a show of the a.i. failure than the human's strategy. Hat's off to the Master player tho, their micro was good and they found a good strat vs a.i.... for now... bwuahahahahaha. What do you think Carl? Do you like watching the a.i. get stumped? (As a human I can still appreciate it ;) )
@@BrendanCassleman Hehe, no, to be honest with you, I'm kind of rooting for the A.I.. I guess it's the 'Tech-nerd' in me that wants it to succeed.
Games like this are just frustrating to watch imo, and it shows that A.I. still has a looong way to go! I wasn't really impressed by the human either, he just got lucky and found a bug/leak. Furthermore, I want to see the A.I. come up with some innovating/unexpected strats that we can learn from, rather than brute forcing with micro.
@@BrendanCassleman I'm with you. i find it strange that i watch other youtubers post videos of Alpha Star beating professionals, and LNG has nothing but videos of it losing to regular ladder players.
This is not A.I. Just poorly written algorithms.
"This is just Zerg getting purified by the Golden Armada" I don't know why but this line had me dead lol
I think it is clear that it just never figured out how to play late game because it usually won by then.
This was as frustrating to watch as seeing a relative buy a brand new nice car, and then refusing to ever change the oil.
yeah and then saying "all those hours I saved not getting oil changes, I was in the office working overtime making more money so I WIN!" (and then the car breaks down right when he needs it). Like it makes perfect logical sense right up to the point where it doesn't :P
@@confuseatronica This is strangely specific
this is the strangest game ever lol. Gotta give it to that master player though. I'm very surprised they didn't give up.
Yeah, no one would've blamed him if he had given up
but maybe he had a hunch something about this enemy was off, not sure if he realised it's AS or just some drunk guy xD
and if he realised it's AS, he maybe just tried some unusual tactic (mass void) to overwhelm AS
at blizzcon, one of the few people winning consistently against AS without cannonrush was beastyqt, who just tried things AS probs doesn't know (i think he went 7/10), things like mech vs protoss
@@eadbert1935 hey I mean it's very possible AS plays more human-like than typical AI's but I'd say it's a very peculiar style a human wouldn't really play like. I'd be pretty suspicious if someone GM ranked played me and had very flawed walling or a terran that just loved banshees way too much, etc. I'd think AS is well-known enough at that high level that maybe they'd be able to figure out who it was.
AlphaStar should be an olympic javelin player, because that was a huge throw. It doesn't know when to WIN!
AlphaStar's Response to Mass Void Rays: Die... Funny I figured that one out years ago
Alpha: this will be the 7th time Ive countered mass void rays, i have become exceedingly efficient at derping it
Alpha: _"There are two buttons. The one labeled 'surrender' leads to the menus and a ranked defeat. The one labeled 'resume' leads back to the match, to the Zerg, and to getting contained on two bases. As you've adequately put, the problem is choice."_
@@VestinVestin The AI will pass spam the menu button in hopes of annoying the human player to rage quit because they cannot proceed with the game.
AlphaStar: "I've never seen anybody do the things you do before"
"And when you're done I'll make you do it all again"
Alphastar: Skytoss does not compute! ERROR ERROR!
I think what's going on here is Alphastar is constantly reevaluating the likelihood that it will win if it attacks, and it's never clear that's the right thing to do. It probably has a pretty good idea what would happen if it attack clicked the base, and most likely all the corrupters would die/ most of the void rays would die, but it's not clearly an improved position. It doesn't necessarily have a good memory, or thought of, "well, I have to do this at some point, so it's now or never". Likewise, when the protoss has the mothership, Alphastar obviously knows it can cloak units, but I don't know that it has a great idea of what exactly is being cloaked.
Strangely alphastar at a certain point pulled back when it should have gone in and finished the protoss player.
The Protoss player however played the perfect game against the AI and that is to wait it out and weather its best moves and then let it get confused by having not won yet.
Sounds like strategic advice from Zapp Brannigan.
If alphastar had a moved at any point after 15 minutes the game would be over, but he just ran around in circles lol
The game went over 15 min, i knew Alphastar was going to lose.
void ray is like the only aerial counter unit to corruptors especially with how efficient they are in a big fight where corruptors waste a lot of shots due to high damage/slow attack speed bursts while void rays waste no uptime at all
they are however extremely weak to zerg's balanced caster units fortunately the AI didnt discover those yet
And the 1 overseer cringefest where you gotta hit the mother ship first
all he had to do is keep trading out his corruptors with the protoss air army, instead it derps aroound and losses 3 here 5 there 2 there for nothing.
Either that or actually threaten the protoss base somehow. Like Mass mutalisk harassment instead of flying the corruptors over the protoss base without engaging the protoss fleet.
Exactly. AlphaStar could have just F2/A-moved and won.
AlphaStar took so many losses just running around
Alphastar needs to purify itself in the lakes of Minnetonka after eating the ultimate Protos Void Ray of Cheese
Thanks for all the content in 2019, your commentary and tone is top-notch and AlphaStar is just a riot to watch as it annihilates and also derps. Happy 2020!
Commentor : "Alphastar is derping so hard right now... he just needs like 200/200 supply in corruptors"
Me: "Ummm... or maybe like 2 infestors" -_-
that force field at the start is so on point
What the heck was AlphaStar doing in this game? First it makes the perfect counter to Carriers. Then it flies them 3/4th of the way across the map while the Carriers tear his base apart at 10:22. Then it turns them around and flies back across the map to clear out the Carriers at 11:20. Then it flies them back to MP's base and actually wins the game at 12:00, but then turns around and goes back home after getting a CLEAR advantage both in economy and army. Then it spends the next ten minutes flying it's Corrupters in and out of the growing airtoss fleet, giving MP tons of free shots without accomplishing anything.
-
This has to be the hardest throw I've ever seen in any game. This game was over, done and won by AlphaStar at 12:00, but for some reason it just did not want the free win.
"So how'd you win?"
"Oh it was easy. I blundered all my economy and then I pressed the V button and I didnt let go until I won"
I see this video blew up for you. I hope the channel keeps growing and you succeed. Alpha star in the title definitely gets the views
Void Rays w/Shield Batteries and using recall is lethal against anyone.
@LaughNgamez Are you sure THIS is Alphastar? The unit control and reaction times are very different. Usually, Alphastar instantly moves all of its units as soon as they are attacked but this doesn't have that split-second reaction time. The lings in the natural do a hold command play which I've never seen alpha star do. Alphastar also does a preemptive counterplay to the air units by building corruptors which I've never seen it do before. Usually, the AI just builds a varied unit comp and the "bad" units die off so that the unit comp grows into a counterplay. That's what I've seen Alphastar do before. This looks very different.
Didn't they limit its APM to 300 after people kept saying that it was unfair for it to have unlimited APM?
@@dionbridger5944 Yes, though it's not like 300 is much fairer... The effective APM (minus the spammy clicks) isn't usually that high... also the AI can technically front or back load it's clicks as needed in a 5 second window so it does have the ability to pull off much crazier micro than real pros so long as it isn't sustained over a lengthier period (and due to things like blink cooldowns it wouldn't have any need to do more anyway). Apparently they get 22 "actions" in a 5 second window, but several SCII actions only count as a single AI "action" it is allowed to select a unit, activate an ability, and set the target location all as a single "action."
Non AI players often tap out early giving Alphastar less chance to train against non typical scenarios
AlphaStar doesn't train against non AI players. It was not training on the ladder, that was just to let the DeepMind team observe its play.
Always funny watching an AI having a stroke. I mean, it's not like humans don't throw games like that, but derping this hard is kinda rare.
Good game. Very back and forth.
Funny thing is with A.I. the more it loses, the more it will adapt and evolve. This DeepMind AlphaStar is in its infancy.
200 game years into the game in its infacy. Looks like it is in trouble.
As somebody who plays a LOT of Voidray / Mothership vs zerg.... Protos made one error over and over... Overlords are always priority targets over damn near everything else. Always pop the vision things first.
Would have loved to see some Archons mixed in. Archons eat clumped up Corruptors so fast.
It's no secret void rays are broken... and after the patch, zerg went from OP to unable to contend with skytoss
ye nerfed infestor was bad
Wow. Fantastic game, that Protoss really earned that W
Alphastar used a perfect zerg behavior: Is my opponent going air? i can just mass roach 😂
human ply robot race , and robot plays organic race
It seems obvious from all the videos that AlphaStar seems to be confused by an opponent playing erratically.
That was a fun game to watch.
When skynet emerges one day. We need our mass void ray equivalent last ditch contingency 😂
They need to train alphastar on how to deal with aircraft.
Alpha played well, it just didn't have a killer instinct
Such a satisfying game to watch a victorious game by the human player. I think this is the answer, for now, where the AI just does not know how to approach mass units.
This is the strangest lost of alphastar I ever see😅
This is likely due to Alpha Star having very little experience in late game because honestly, its opponents likely never get that far. It likely nearly always demolishes it's opponents in early game.
Which alphastar is it? there are multiple agents they use for various ranks. I guess this is the master-ranked alphastar? -- is this the old one who can exceed human level APM and see the whole map? ( That one usually did mono-battles, with crazy micro )
why doesnt alpha star have perfect creep spread?
I love when people ar fighting and not giving up early.
Just 3 shield battery's doesn't stay near them and zerg ai that favors roach vs air over hydra no parasitic bombs or green goo.
Seems like aplhastar is on low IQ mode, or non korean training mode.
He just derped with the corrupts in the attack for some reason instead of keeping the (anti)-air to defend...
old build still gold even void rays were nerfed after HOTS launch
Today I learned that a drone weighs 333.33 KG
Clearly AlphaStar had those 200 years of training with Protoss Vs Protoss, and never learned pretty much anything about Zerg playstile.
Lol this battle was brutal, protracted and twisted like Nazi Germany vs. Soviet Union. :DD
Best SC2 duel what I've seen so far. MP deserved that victory.
Alphastar unit composition response is pretty bad yet for Terran and Zerg.
Interesting, that AlphaStar didnt go for upgrades on his units.
little do yall know but all of that was 4d thinking by the ai
Wouldn't that masters player being saying "wtf" to Alphastar's Roach/Ravager build?
Train it a little vs all in air and this will never repeat. Alphastar would crush this guy if it had trained against it prior. Anyone could win that late game.
This game should really be called “ Alpha star doesn’t know what a winfestor is”
You can actually see the AI logic fallacy when it sees voidrays: the enemy is going Air, so I will too. Which it does.
Only the damned game makers didn't give the corruptor a ground attack, like the voidray has. So it makes roaches for ground attack.
Only the damned game makers didn't give the roach an air attack. It's basically like a 2 year-old in its decision making.
Compare this game to its Protoss version, where it makes mass-stalker. The AI wins easily, since its units shoot both air and ground.
True, it's Protoss is way way better than it's Terran or Zerg... maybe Toss is just OP, but it seems extra micro exploitable as well.
All alphastar had to do was A-move...
Don't see enough defensive storm.
Facepalm after facepalm... alphastar too
Nice to see the "human" (not terran) resistance :p
Subbed.
Good luck mate. TH-cam is messed up these days.
Alphastar simply shoud have mass expended when toss was locked into 2 bases and after that stack up money, push all 45 corrupters into the toss army, build another 45 corrupters and end it. Theres rly no need to tech vipers or something - eco wins.
ботинок тащумба :))) Спасибо за видео.
Stupid AI main focus is the mothership.
No abductors.
Protoss learn in the first time.
He back away just his mothership quicker second time and AI losses alot of corruptors.
The human was clever with the tempest. If he could make more to one shot the detector it would be GG faster.
I don’t see a derp.
“When you are ahead get more ahead”
That’s exactly what alpha star did.
Toss tech’ed up and wrecked
Zerg should have traded out more. But Losing units to trade eco isn’t alphastars play style.
...No derp
Should have focused on tech more I feel as well. Even armour just to balance out that +3 on voids. Also probably a bit more harassment, even pick up a pylon or two with corrosive.
I thought this was BeastyQT
"immaculate" i dont think it means what you think it means.
One fungal growth and parasitic bomb to rule them all bro
obviously Alphastar is blind to the cloaked units and has absolutely no idea where is the damage coming from
It tried to use overlord to detect but zerg otherwise dont have counter for air units.
I kinda saw that coming, alphastar is just so bad vs air. Least it tried to make some antiair, but seems like it havent really still figured it out.
I dont think the AI has been vs enough mass one unit traning
Starcraft came out before Halo.
Getting so frustrated at this AI while watching this, thinking to myself that it clearly has not learned that having triple the number of corruptors versus the other air army means they can just a-move in, batteries or not. Seeing it constantly give away its corruptors by moving nearby and pulling back just made no sense to me. Would love to see it's willingness to commit be more decisive in the future as it keeps learning.
I don't know if they actually had enough corruptors to win that. Balling up tightly and one shot ting individual units might actually have been the best strategy
@@emissarygw2264 If the army sizes were closer that would be the case, but at about 15:18 it is pointed out that Alpha Star has 35 corruptors. It could have EASILY beaten a dozen voidrays and two carriers, but simply could not decide to commit, and every time it went into range of their attacks but ran instead of fighting back, it lost units for free, which let the protoss catch up.
Alpha star never brought detection with it. The most likely scenario is it would see the mothership go in for the kill get targeted by cloaked units and then retreat because suddenly it no longer has a 90% victory condition do to unknown variables.
Their Ai from the beginning has always functioned on a % victory condition where it would predict the chance of winning an engagement before engaging.
So if it predicts it can secure the win in the engagement it will engage. If it then starts taking more damage then predicted it will retreat as engagement parameters have changed.
That why it constantly dived then ran away then circled around its base till it recalculated its likelihood of winning.
I'm pretty sure it counts shield batteries as a factor in the engagements as well even if the player never used them.
@@urstaxfetish1206 I am absolutely certain that is the case later on, but there was a point at which it could undoubtedly just run in and kill the other player for that time stamp I used. 15:18 target down voids with corruptors while targeting batteries with the roaches.
It is clearly still learning but it will continue to struggle to secure a win in situations where it has to break through a fortified position because of this indecisiveness/excessive caution against certain defenses. Without actually having access to a log of its evaluations of the threat at each encounter, all we can really do is guess at its choices and hope it gets better.
Maybe alphastar did not understand that there are units cloaked under the Mothership?
Halal Milksheikh alphastar is dory from finding Nemo:
1- oh, a lone mothership, better take it
2- my corruptors are dropping like flies, better retreat and think again
3- repeat
What is this? Wings of Liberty?
Yes, Alphastar derped, but can we take a moment to applaud the masterplayer's decison making? He realized that he couldn't hold out indefinitely and took a risk by harassing the zerg over an overextension. I bet most players would've been afraid of losing, but he realized that if doesn't make a significant change here, he has basically lost anyways.
If there's ever a word I want removed from all caster's vocabulary, it's "as"
my 6.9cm korean void ray peeing yellow beam
The thing is I also just lost 2 games to void rais, so I feel this alphastar behavior is pretty human. I had corruptors and hydras, big supply lead, but could not beat the voids if the player has income. Maybe vipers are needed.
Yeah zerg has no effective counter to mass void ray aside from parasitic bomb.
the AI is used to people surendering
It lost its marbles.
Didnt see any high templar or archons from the toss... Or even cannons... That would have made it WAY easiert
I don't think they really had the economy for a transition at that point...
AlphaStar doesn't know hydralisks exist yet?
Good game
Where is the mass void ?
Wait, alphastar dies to mass void ray, and i die to mass void ray, that means... Im as good as alphastar!
Alphastar doesn’t know how to use infestors huh?
alphastar needs to use magical unit. is it ok updating?
25 minutes of AlphaStar derping... hmm