The AI: Does over 9,000APM and have some crazy moves that no human can mimic. Also the AI: Have a single brain cell leased from Hearthstone balance team, leaving them none.
The AI lost, but it was so cool to see it play. The way it played really looked like a hive of insects. The creep was over the entire map by 10 minutes, units were engaging and disengaging individually instead of as a group. And each unit either burrowed or tried to run away when damaged, just like ants or the like.
Feels like SC1 Cerebrates, each of them specialised in certain aspect. This one is super aggressive, fast expand, creep creep creep, relative low-tech composition, paving the way for other high-tech force from other Cerebrates.
@@Warclam on YT mobile you can see the top comment without opening the comments section, it spoiled it for me too. Shame there's not a spoiler tag like on reddit.
Preface - I am not familiar with any of the StarCraft AI works. As soon as the game starts in 1:40, the AI maximises the mineral gathering rate by manually right-clicking each worker right before they arrive to their destination to minimise speed lost from decelerating upon arrival. I don't know how much that'll affect early game, but I think that's a very interesting kind of optimisation.
It was a concept introduced by Scarlett. It increases the amount of money you get by about 30% assuming perfect micro of the workers, but generally you should only do it if you’re confident enough in your APM to do that AND micro AND produce units.
@@dresdenmini8586 for an AI, yeah, for a human, you can barely do it with a few workers at the start of the game, but the second you have to do anything other than it you literally cannot even attempt it, even if you've got the APM of serral whilst playing skytoss.
@@Santisima_Trinidad bots have a limit to their APM. At some point the SC2 engine cannot handle the number of inputs and starts dropping some. IIRC it's around 80000 APM. So yes, at some point they have to prioritize where to dedicate their APM
@@gorkemaykut5230 Yes. For example, it's possible to individually micro Zerglings as they get targeted by Colossi so that Zerglings actually counter Colossi by dodging the shots.
This was awesome to watch. Not only does this illustrate how far having excellent macro can take you, and also being able to micro with it, but it also illustrates that APM isn’t everything. Also, you’re amazing Grant! Your commentary as you played the bot allowed for us noobs to get a look inside the thinking that goes into the highest levels of starcraft play.
Thing about APM (from my experience) is it's value doesn't increase linearly forever. The difference between 20APM and 100APM is massive and gamewinning, but the difference between 100APM and 500APM isn't nearly as huge in terms of power.
These bots also don't seem to be good at decision-making. The AI could have killed him within the first three minutes when he didn't build a wall at the start. Likewise, it kept splitting off little chunks of its army to do ineffective harrassment with units that aren't great at it (a couple roaches, for example), so that its army was always too small to fight Grant's straight up despite having an enormous economy. Likewise, even when it did keep its units together, it attacked into really tight chokes that let them get slaughtered by colossus. And then it kept building roaches long after they were any good against late-game Toss. All in all, the APM tricks are impressive in their own right, but these AI need a lot of work still.
@@Draeckon True, it could have, and it all but highlights how important decision making and game sense really is. I don't think the AI would have defeated him, though, Grant would have found a way, whether by capitalizing on the AI's consistent over microing, or by getting a good angle, and getting a hold that doesn't result in game ending economic damage. While the wall isn't as effective against early zerg roach pressure, any ling pressure the AI could mount would be held fairly easily with the wall, which means roaches would be needed to bust that wall, and by that time, Grant would have had a robo and more units, thus a hold would be achieved yet again. Not saying it wouldn't be difficult, but it's not impossible. Also, it's really difficult to code an AI to do this sort of thing, so they are, regardless of any flaws, works of art.
@@laffinggas8068 Yeah, and I was addressing the fact that a lot of people seem to harp on the perception that somehow, a higher APM matters, when it's effective APM that matters. If there was someone with 20 APM that did something useful with every bit, as opposed to someone who had 100 APM, but it was a good bit of spam, I would bet on the 20 APM player any day of the week. So, while what you say is correct, I was more addressing a common concern of players that the game requires huge amounts of APM, when in fact, all that's needed is to know what you want to do and to execute it well, as shown in this video. Grant knew what he wanted to do, executed it well, and it got him a sound victory.
Well, even after that failed wall at the start and constantly being on the back foot for half the game, Grant sure showed his strength here and crushed the AI. All his Nightmare Edition play throughs and other challenges having him be on his toes really paid off.
Eh, the AI never made a serious attack aside from that one with 2 lurkers, but even that wasn't particularly threatening. It didn't seem to have much in the way of reactionary play, given that it went mass roach vs a maxed carrier + colossus army...
I'm pretty sure this would be very difficult to set up, but I'd love to see you try to take on a starcraft AI while having the full power of the campaign Protoss. Where you could choose the upgraded versions of the units you get in the campaign and have access to the spear of adun, and have to beat an AI so hard no human has a chance.
The only AIs hard enough that most master league players (who Grant had been back in the day, by the way) would lose to them are MicroMachine and Google's alphastar. The latter is unavailable to the public, and the first can be cheesed out of the game by a diamond league player. The only way to make it fair is cheater AI.
I think your lack of recent ladder experience is a strength as the AI probably expects it's opponent to play like a ladder player. Also I'd love to see you try out the other AI to see how they play.
I think the bot expects its opponent to play like another bot. The bots practice against each other so they can get in more practice games in less time. This does leave some of them with some exploitable weaknesses that don't come up in bot practice games because the other bots haven't yet realized how to exploit them, but it also means they come up with some crazy strategies you'd never expect to encounter in a ladder game.
I'm surprised at how quickly the bot falls. One minute it expanded and took the whole map over, the next, it can barely manage to max out on roaches and builds no anti-air even if it knows the composition of Grant's army
It did have anti air. The original strat of the AI was roach/hydra. Little to no focus on lings. If I didn't know any better I would have said Winter was playing this game.
By 10 minutes 90% of the map was covered in creep, the ai even went so far as to bring an overlord to poop creep in your expansion. Wouldn't have it any other way.
The bots might be considered a nightmare difficulty for most players, but Grant? Grant already beat nightmare difficulty. In spite of the janky start, well played! ^v^
Any chance we could see you vs other bots Grant? There are some really fun bots like Ketroc that know how to planetary fortress rush or spam manner mules if you let them get their eco up
This was new to me as someone who never plays ladder (still only vaguely aware of what an oracle does lol) but it was so fun to watch. I kept expecting a mondo counter attack to end it all
Something really weird happened when the AI was defending grants army, there were times he would stutter step backwards, or even just back up for better positioning, and the AI would recognize this as retreating and immediately start moving back units, so grant would be backing up while attacking and the AI would be stuck issuing move commands
An ultimate zerg AI would be completely unstoppable First, as we see, the creep is at Grant's base in 10 minutes. Holy. Second, their units no longer take most AOE damage, because any AOE they would take is instead micro'ed away. Colossi strike in a line, tanks and archons in a circle, but ultralisks will still deal damage. Third, perfect injects. In most cases, there will be no lack of larvae, ever. Four, the tiny micro tricks, like the mineral trick, or in case of Terran(I know I said zerg but) the bunker/vespene trick, would give them a small amount of income over a human player. Five, there is no such thing as multi-pronging an AI, because it can micro three fights at once. (This is a number I pulled out of a cloud,.) Six, perfect infestor/viper micro. Enough said. Seven, timing attacks accurate down to the second. No such thing as a little misclick or oversight, if the AI says it can hit you with a wave of 1/1 roaches at 7:20, it's not hitting you at 7:21. This will be how terrifying a perfect AI is. But for now, let's just watch roaches run into carriers.
I think the AI didn't know what tech harass is. It kept building bases, but never rebuilt any tech other than the spawning pool. XD unorthodox strategy does it again!
The hit-and-run micro with individual units was interesting to watch, but if instead running away it used this micro to get closer and focus fired it could've easily won the engagement. I assume it doesn't focus fire because it trained against similar bots (or even itself), and against this much pullback micro focus fire isn't very useful (although Roaches running down the Colossi on creep would've still been possible, even if the Colossi were retreating between attacks). I think one video about this was enough for now, but I could certainly see this type of content used to fill gaps between two longer campaigns.
so...will there be an AI screen as well since you recorded it? Also, very good work, i think the problem with the AI was that it didn'tknow how to transition armies and rebuild/protect all their bases
The AI focused so much on economy it didn't have room for enough combat units. Insane APM and execution but the strategy isn't there. Which is fortunate, because that level of micro with an actual game plan? Ouch.
I'm surprised by the burrow tactics. Since these bots play against other bots, i wouldn't expect burrow to be a common thing, as i assume that other bots would be able to perfectly see the little shimmer of cloaked units? Although I'm not sure exactly how it works from their end
Personally, I'm more than fine with your build orders not beeing super clean. But, what I think might be interesting is, you looking at the replay (if there is one) and whatching, what exactly the Bot did.
These bots are just as good as Automaton2000 in the micro department, if not better (see the showmatch between the bot Grant played, Eris, and a Terran bot called MicroMachine casted by Harstem earlier this year), but their build orders are tailored for fighting other bots with comparable micro and so won't (usually) be able to take advantage of a human's lesser micro.
I liked this content. It was a deviation from the normal, and really interesting to me. But can someone who plays Protoss explain why Grant didn't get Observers immediately to control the creep spread? Just limited APM?
He decided that the oracles were good enough, and that his robo facilities would be better used to make colossi rather than observers. I do agree, though, that observers earlier would have been good
While the bot's APM is really impressive, its decision-making seems to be kind of bad. The fact it didn't scout Grant not building a wall for that long in the early game and send a flood of lings to just kill or horribly cripple him is really funny. That's the fastest and most guaranteed way to lose in PvZ on ladder, at almost any level Silver or above. EDIT: The way it kept splitting its army up into smaller bite-sized chunks also doesn't help win. It looks really cool and it's neat, but that's a great way to lose against a proper deathball army that's just barreling straight into you. In the occasions where it kept its army together, it decided to attack into really tight chokes against colossus. In short, it's cool, but these AI need a lot of work if they're intended to play against humans.
Grant, does the way you launched the game stop you from using rapid fire? I noticed at 13:34 that you weren't using it. Edit: Watched the clip again, I don't think I can tell. Nevermind ...
You're correct I didn't use rapid fire warp-ins. I had it set up so that I could. Absolutely worked properly. No idea why I didn't use it. My brain must have been creeped over by that point!
The main differences are speed (the newer AlphaStar builds are capped to a ~300 apm average vs the 40k+ apm of these bots. AlphaStar appears like a human is controlling it most of the time due to the apm limit, though it can have superhuman micro during small engagements, while the other bots are able to micro every unit individually across the entire map), vision (AlphaStar is limited to a normal screen's worth of the map for issuing commands and must use an action to change its "view"; the other bots can issue commands anywhere on the map, they are effectively playing with the screen zoomed out so that the entire map is on screen, with fog of war still over locations that aren't currently in vision range of units as usual), and "style" (see below.) AlphaStar is a machine learning algorithm that was trained on thousands of pro replays and then played millions of matches against itself to learn the "best" strategies ("best" is in quotes because the strategies it uses are those that won when piloted by AlphaStar against itself, they may not be the objectively best play in general); it plays more like a human with weird build orders than a bot but has more trouble anticipating changes and adapting to new situations than a human would. In contrast, these bots are following a script created by an individual programmer to beat other bots with every interaction having to be hard coded in. Hence, they typically have at most two or three build orders with branching reactions for every enemy action the programmer can think of. In general, the build order foundations are either super aggressive with strong timing attacks or cheeses that take advantage of bot-only weaknesses/strategies or, like the bot Grant played, very defensive and reactionary with the goal of surviving any early aggression from enemy bots (for instance, Grant's opponent Eris plays the same safe opener on every map, including a long scouting trip to search for proxies on its side of the map) and trading efficiently in a long grind of a game. One way the difference between bot types manifests is in decisiveness; AlphaStar will press the advantage after favorable engagements, presumably based on its database of pro games (the games vs Serral from a few years ago show this well as AlphaStar holds off timing attacks from Serral and immediately counters to win.) For an example of the more indecisive style, see the Eris vs Terran bot MicroMachine showmatch Harstem casted earlier this year where Eris had a very sizable advantage throughout the midgame but never presses its advantage with a large scale attack and was eventually bled dry by MicroMachine (who got a huge lead over Eris in the lategame but took over 5 minutes to finish off an opponent with 100 less supply.)
Grant: I don't do ladder.
Also Grant: I train with Nightmare WoL & HotS
AI: Oh... Oh, no...
i see the nightmare mode training has paid off
The AI: Does over 9,000APM and have some crazy moves that no human can mimic.
Also the AI: Have a single brain cell leased from Hearthstone balance team, leaving them none.
the way the individual units bounce in and out based on their attack cooldown reminds me a lot of BWAPI - really cool stuff
The AI lost, but it was so cool to see it play. The way it played really looked like a hive of insects. The creep was over the entire map by 10 minutes, units were engaging and disengaging individually instead of as a group. And each unit either burrowed or tried to run away when damaged, just like ants or the like.
Feels like SC1 Cerebrates, each of them specialised in certain aspect. This one is super aggressive, fast expand, creep creep creep, relative low-tech composition, paving the way for other high-tech force from other Cerebrates.
Why is this the top comment it spoiled the video before it even started
@@fungusgnat8559 Because people who care about that don't read the comments before the video.
@@Warclam on YT mobile you can see the top comment without opening the comments section, it spoiled it for me too. Shame there's not a spoiler tag like on reddit.
@@jacobbireiner9395 just because something is on screen doesn't mean you have to read it. have some self control.
Preface - I am not familiar with any of the StarCraft AI works.
As soon as the game starts in 1:40, the AI maximises the mineral gathering rate by manually right-clicking each worker right before they arrive to their destination to minimise speed lost from decelerating upon arrival. I don't know how much that'll affect early game, but I think that's a very interesting kind of optimisation.
It was a concept introduced by Scarlett. It increases the amount of money you get by about 30% assuming perfect micro of the workers, but generally you should only do it if you’re confident enough in your APM to do that AND micro AND produce units.
@@dresdenmini8586 for an AI, yeah, for a human, you can barely do it with a few workers at the start of the game, but the second you have to do anything other than it you literally cannot even attempt it, even if you've got the APM of serral whilst playing skytoss.
@@Santisima_Trinidad bots have a limit to their APM. At some point the SC2 engine cannot handle the number of inputs and starts dropping some. IIRC it's around 80000 APM. So yes, at some point they have to prioritize where to dedicate their APM
@@Ghi102 is there even enough thing in the game for you to click 80k times in a minute?
@@gorkemaykut5230 Yes. For example, it's possible to individually micro Zerglings as they get targeted by Colossi so that Zerglings actually counter Colossi by dodging the shots.
Wow, with the amount of creep spread during this match, it makes the mini-map look like a game of Creeperworld.
This was awesome to watch. Not only does this illustrate how far having excellent macro can take you, and also being able to micro with it, but it also illustrates that APM isn’t everything.
Also, you’re amazing Grant! Your commentary as you played the bot allowed for us noobs to get a look inside the thinking that goes into the highest levels of starcraft play.
Thing about APM (from my experience) is it's value doesn't increase linearly forever. The difference between 20APM and 100APM is massive and gamewinning, but the difference between 100APM and 500APM isn't nearly as huge in terms of power.
These bots also don't seem to be good at decision-making. The AI could have killed him within the first three minutes when he didn't build a wall at the start. Likewise, it kept splitting off little chunks of its army to do ineffective harrassment with units that aren't great at it (a couple roaches, for example), so that its army was always too small to fight Grant's straight up despite having an enormous economy. Likewise, even when it did keep its units together, it attacked into really tight chokes that let them get slaughtered by colossus. And then it kept building roaches long after they were any good against late-game Toss.
All in all, the APM tricks are impressive in their own right, but these AI need a lot of work still.
@@Draeckon True, it could have, and it all but highlights how important decision making and game sense really is. I don't think the AI would have defeated him, though, Grant would have found a way, whether by capitalizing on the AI's consistent over microing, or by getting a good angle, and getting a hold that doesn't result in game ending economic damage. While the wall isn't as effective against early zerg roach pressure, any ling pressure the AI could mount would be held fairly easily with the wall, which means roaches would be needed to bust that wall, and by that time, Grant would have had a robo and more units, thus a hold would be achieved yet again. Not saying it wouldn't be difficult, but it's not impossible. Also, it's really difficult to code an AI to do this sort of thing, so they are, regardless of any flaws, works of art.
@@laffinggas8068 Yeah, and I was addressing the fact that a lot of people seem to harp on the perception that somehow, a higher APM matters, when it's effective APM that matters. If there was someone with 20 APM that did something useful with every bit, as opposed to someone who had 100 APM, but it was a good bit of spam, I would bet on the 20 APM player any day of the week. So, while what you say is correct, I was more addressing a common concern of players that the game requires huge amounts of APM, when in fact, all that's needed is to know what you want to do and to execute it well, as shown in this video. Grant knew what he wanted to do, executed it well, and it got him a sound victory.
I love watching bot plays, but there are few interesting commentators! It's really enjoyable when someone like you or Harstem tries their hand at it!
Well, even after that failed wall at the start and constantly being on the back foot for half the game, Grant sure showed his strength here and crushed the AI. All his Nightmare Edition play throughs and other challenges having him be on his toes really paid off.
Eh, the AI never made a serious attack aside from that one with 2 lurkers, but even that wasn't particularly threatening. It didn't seem to have much in the way of reactionary play, given that it went mass roach vs a maxed carrier + colossus army...
I'm pretty sure this would be very difficult to set up, but I'd love to see you try to take on a starcraft AI while having the full power of the campaign Protoss. Where you could choose the upgraded versions of the units you get in the campaign and have access to the spear of adun, and have to beat an AI so hard no human has a chance.
Imagine a versus mode where the factions are the campaign factions, with Kerrigan and the spear of adun.
There’s definitely a mod for that on custom. Not sure how viable it is to use it with these AI, but it exists.
At least make you cannot use the final call down slot. Time stop to snipe tech buildings would make it trivial.
The only AIs hard enough that most master league players (who Grant had been back in the day, by the way) would lose to them are MicroMachine and Google's alphastar. The latter is unavailable to the public, and the first can be cheesed out of the game by a diamond league player. The only way to make it fair is cheater AI.
I think your lack of recent ladder experience is a strength as the AI probably expects it's opponent to play like a ladder player.
Also I'd love to see you try out the other AI to see how they play.
I think the bot expects its opponent to play like another bot. The bots practice against each other so they can get in more practice games in less time. This does leave some of them with some exploitable weaknesses that don't come up in bot practice games because the other bots haven't yet realized how to exploit them, but it also means they come up with some crazy strategies you'd never expect to encounter in a ladder game.
Welcome to the AI's, Grant! I'd love to see more of these Grant vs. AI matches!
Would love to see the other bots. Very cool that this is a thing. (Congrats on the win vs the murder AI)
I'm surprised at how quickly the bot falls. One minute it expanded and took the whole map over, the next, it can barely manage to max out on roaches and builds no anti-air even if it knows the composition of Grant's army
It did have anti air. The original strat of the AI was roach/hydra. Little to no focus on lings. If I didn't know any better I would have said Winter was playing this game.
By 10 minutes 90% of the map was covered in creep, the ai even went so far as to bring an overlord to poop creep in your expansion. Wouldn't have it any other way.
I would so be down for more AI versus videos! this was a cool and fun mix up Grant!
Every unit and building is basically a separate player but they all coordinate with each other, that's amazing.
>that thumbnail
“What’s the AI’s APM?”
It’s OVER 9000!!!
The bots might be considered a nightmare difficulty for most players, but Grant? Grant already beat nightmare difficulty. In spite of the janky start, well played! ^v^
Any chance we could see you vs other bots Grant? There are some really fun bots like Ketroc that know how to planetary fortress rush or spam manner mules if you let them get their eco up
I didn't know bots have preferences depending on which you match up against. Love when games do that
This was really fun to watch. I wouldn't mind seeing more 1v1 stuff, be it other bots or ladder
This was new to me as someone who never plays ladder (still only vaguely aware of what an oracle does lol) but it was so fun to watch. I kept expecting a mondo counter attack to end it all
Grant: 'I hope I don't die to a 6 pool or something'
Me: 'Well golly oh gosh Mr Grant it really has been a long time since you laddered last.'
Grant wouldn't have lost Aiur to the Zerg.
Vote Grant for Protoss Day llama
Now we need an A.I. that knows how to play as Grant so we can have a Grant vs Grant match.
I’d have no issue watching this again. It was really interesting and different but not in a bad way.
yo this was insane man and really entertaining i would love to see more of this
I think it was cool to see the zerg act semi zerg ish. It would be cool to see some kind of implementation of this into a the missions.
id love to see you try this either as or agaist the other races
GGG: Oh im so nervouse becouse i dont ladder
Also GGG: Oh no ai has written gg, becouse of humilation he suffered and my finnes...
Play all 3 races against all 3 races. I would love to see more of this!
Something really weird happened when the AI was defending grants army, there were times he would stutter step backwards, or even just back up for better positioning, and the AI would recognize this as retreating and immediately start moving back units, so grant would be backing up while attacking and the AI would be stuck issuing move commands
I found this fun and wouldn't mind more! Good job on the win! :)
You’re amazing GGG
An ultimate zerg AI would be completely unstoppable
First, as we see, the creep is at Grant's base in 10 minutes. Holy.
Second, their units no longer take most AOE damage, because any AOE they would take is instead micro'ed away. Colossi strike in a line, tanks and archons in a circle, but ultralisks will still deal damage.
Third, perfect injects. In most cases, there will be no lack of larvae, ever.
Four, the tiny micro tricks, like the mineral trick, or in case of Terran(I know I said zerg but) the bunker/vespene trick, would give them a small amount of income over a human player.
Five, there is no such thing as multi-pronging an AI, because it can micro three fights at once. (This is a number I pulled out of a cloud,.)
Six, perfect infestor/viper micro. Enough said.
Seven, timing attacks accurate down to the second. No such thing as a little misclick or oversight, if the AI says it can hit you with a wave of 1/1 roaches at 7:20, it's not hitting you at 7:21.
This will be how terrifying a perfect AI is. But for now, let's just watch roaches run into carriers.
All I could think about during the first part of the video is...make...a...observer!
This was brilliant. Would love to see more. Looked like a proper zerg conflict. X
Now we need Grant to fight with mr. talk-a-lot Micromachine
Please more of this!
Looks fun, but please show the score screen! Or maybe a run through the replay. Would love to see the resources lost comparison.
I think the AI didn't know what tech harass is. It kept building bases, but never rebuilt any tech other than the spawning pool. XD
unorthodox strategy does it again!
This was surprisingly fun to watch. I have watched other VS AI and they were not really good. This was better
The hit-and-run micro with individual units was interesting to watch, but if instead running away it used this micro to get closer and focus fired it could've easily won the engagement. I assume it doesn't focus fire because it trained against similar bots (or even itself), and against this much pullback micro focus fire isn't very useful (although Roaches running down the Colossi on creep would've still been possible, even if the Colossi were retreating between attacks).
I think one video about this was enough for now, but I could certainly see this type of content used to fill gaps between two longer campaigns.
so...will there be an AI screen as well since you recorded it?
Also, very good work, i think the problem with the AI was that it didn'tknow how to transition armies and rebuild/protect all their bases
When Aiur Protoss want to eradicates IMBA APM Zerg on Char.
imagine a human having that much APM. THE POWER THE POWERRR
Somehow this reminded me of my thought that Purifier Beam, Nuke, and Flamethrower units should be able to clear creep directly.
you need to do more of these with MORE AI's 2 vs 1 and see what happens!
I would gladly watch more of this.
lol bot just let you kill half its workers AND stop its mining by getting burrow. 9000 iq play there
you ever see those thermal videos where he solos like, all the ai? i challenge you to do that without turtling lol.
I read the Player Name as Serral But Butter and my brain is confused for a whole minute.
SerralButBetter was hilarious, good one
Thanks for the fun! ^^
Love to see man beat AI, they have such an advantage.
Darn :D
Fighting terran with max api sounds like a nightmare... xD
His campaign ai knowledge is gonna go on overdrive now
Archon mode vs super AI sounds like a !fun! time
The AI focused so much on economy it didn't have room for enough combat units. Insane APM and execution but the strategy isn't there. Which is fortunate, because that level of micro with an actual game plan? Ouch.
I wonder if it's possible to put these Ais into a campaign to make it ultra nightmare campaigns
Didn't expect that ending O.O
I'm surprised by the burrow tactics. Since these bots play against other bots, i wouldn't expect burrow to be a common thing, as i assume that other bots would be able to perfectly see the little shimmer of cloaked units? Although I'm not sure exactly how it works from their end
After playing touhou for so long grant is above infinity apm
I absolutely loved this.
This was really great, play the other A.I.s!
Yes, please, more!
Personally, I'm more than fine with your build orders not beeing super clean.
But, what I think might be interesting is, you looking at the replay (if there is one) and whatching, what exactly the Bot did.
great map choice :D
That was cool, maybe try against MicroMschien next time. It plays Terran and has some incredeble micro
The mining micro
You should do this at least 9 times so you can do all the match ups. It could be a new years stream or something.
I have absolutely no idea what’s going on (I only really play co-op and campaign) but you sure did warp those zerg good
I so wish someone would bring back the Automaton2000 from 2011. Go up against THAT AI.
These bots are just as good as Automaton2000 in the micro department, if not better (see the showmatch between the bot Grant played, Eris, and a Terran bot called MicroMachine casted by Harstem earlier this year), but their build orders are tailored for fighting other bots with comparable micro and so won't (usually) be able to take advantage of a human's lesser micro.
Hi @Grant I just wanted to let you know, YT is doing its SC1 shenanigans on your SC2 vids again.
we want more! i want... at least...
More of that stuff
yes! more of this
I liked this content. It was a deviation from the normal, and really interesting to me.
But can someone who plays Protoss explain why Grant didn't get Observers immediately to control the creep spread? Just limited APM?
He decided that the oracles were good enough, and that his robo facilities would be better used to make colossi rather than observers. I do agree, though, that observers earlier would have been good
You should check out the BroodWar bots as well! There's a platform called SCHNAIL
While the bot's APM is really impressive, its decision-making seems to be kind of bad. The fact it didn't scout Grant not building a wall for that long in the early game and send a flood of lings to just kill or horribly cripple him is really funny. That's the fastest and most guaranteed way to lose in PvZ on ladder, at almost any level Silver or above.
EDIT: The way it kept splitting its army up into smaller bite-sized chunks also doesn't help win. It looks really cool and it's neat, but that's a great way to lose against a proper deathball army that's just barreling straight into you. In the occasions where it kept its army together, it decided to attack into really tight chokes against colossus.
In short, it's cool, but these AI need a lot of work if they're intended to play against humans.
How did Grant know I was watching his 3v3 series?
For those in the know, whatever happened to Giant Grant Grandmaster?
How can one play against alpha-star?
Are there better AIs than alpha-star?
Is this bot better than Scarlett? How high level is its Zerg play?
Sure do its fun !
That's scary..
yeah uhuh more of this shit please
Did the ai start mining before the countdown timer even ended??
It did lol
LaughNGames posts a good few of the bot v bot matches if yall wanna see some crazy stuff (including Eris)
Micromachine go
Over 9k APM AI vs over 9K IQ GGG, the outcome was predictable. After all one of the participants isn't a giant.
Grant, does the way you launched the game stop you from using rapid fire? I noticed at 13:34 that you weren't using it.
Edit: Watched the clip again, I don't think I can tell. Nevermind ...
You're correct I didn't use rapid fire warp-ins.
I had it set up so that I could. Absolutely worked properly. No idea why I didn't use it. My brain must have been creeped over by that point!
@@GiantGrantGamesArchives Well, the A.I.'s creep spread totally would have left the screen if you'd played for any longer lolol
So, the site is giving a 404. Is there any way to still play against this AI or is dead forever like any good AI that ever comes out in this game?
Yes
That was some solid macro, do you often play ladder?
Grant... You need to make more observers not just oracles. You could have cut off/cleared the creep so much easier that way.
Wanna see match against Terrain, MicroMachine or something who is dominant as terrain in AI ladder)
What’s the difference between this type of bot and AlphaStar?
The main differences are speed (the newer AlphaStar builds are capped to a ~300 apm average vs the 40k+ apm of these bots. AlphaStar appears like a human is controlling it most of the time due to the apm limit, though it can have superhuman micro during small engagements, while the other bots are able to micro every unit individually across the entire map), vision (AlphaStar is limited to a normal screen's worth of the map for issuing commands and must use an action to change its "view"; the other bots can issue commands anywhere on the map, they are effectively playing with the screen zoomed out so that the entire map is on screen, with fog of war still over locations that aren't currently in vision range of units as usual), and "style" (see below.)
AlphaStar is a machine learning algorithm that was trained on thousands of pro replays and then played millions of matches against itself to learn the "best" strategies ("best" is in quotes because the strategies it uses are those that won when piloted by AlphaStar against itself, they may not be the objectively best play in general); it plays more like a human with weird build orders than a bot but has more trouble anticipating changes and adapting to new situations than a human would.
In contrast, these bots are following a script created by an individual programmer to beat other bots with every interaction having to be hard coded in. Hence, they typically have at most two or three build orders with branching reactions for every enemy action the programmer can think of. In general, the build order foundations are either super aggressive with strong timing attacks or cheeses that take advantage of bot-only weaknesses/strategies or, like the bot Grant played, very defensive and reactionary with the goal of surviving any early aggression from enemy bots (for instance, Grant's opponent Eris plays the same safe opener on every map, including a long scouting trip to search for proxies on its side of the map) and trading efficiently in a long grind of a game.
One way the difference between bot types manifests is in decisiveness; AlphaStar will press the advantage after favorable engagements, presumably based on its database of pro games (the games vs Serral from a few years ago show this well as AlphaStar holds off timing attacks from Serral and immediately counters to win.) For an example of the more indecisive style, see the Eris vs Terran bot MicroMachine showmatch Harstem casted earlier this year where Eris had a very sizable advantage throughout the midgame but never presses its advantage with a large scale attack and was eventually bled dry by MicroMachine (who got a huge lead over Eris in the lategame but took over 5 minutes to finish off an opponent with 100 less supply.)
I would enjoy for you to play against more AI please 👍