Short Story about Jimmy, Mike and Steve. Mike: "Hey Jimmy, that terran has a Banshee" Jimmy: "No worries Mike, we have Stalkers!" Mike: "Jimmy, where did that Banshee go?" Mike : "Also Steve just died!" Mike: "Jimmy?!"
The short games are really interesting as well! Just it really hurts to watch AlphaStars bad micro against cannon-rushes! Half of the workers it pulls are idle without attacking the pylons that are finishing. 😢 It does that every time, when a cannon finishes, not just in the game we saw now.
Really doubt it. The whole point from Deepminds perspective was to develop an AI playing at human apm, so the AI has been developed with those restrictions from the start. And with no apm restrictions it would be a walk in the park for an AI vs humans. Search for the video "Automaton 2000 Micro - Dodging Siege Tanks" and the see the difference between 100 zerglings (not) perfectly microed vs 20 siege tanks.
more or less, look for "alphastar vs mana game 4" it has like at one point over 1500 apm. keep in mind that giving the AI more apm does not mean it will perform better, the opposite is actually true since it doesnt see the need for a good macro/strategy and instead relies on its micro.
@@DerDevil1998 The version of Alphastar which Mana played against still had restricted apm. And had lower mean apm compared to Mana. See: www.engadget.com/2019/01/24/deepmind-ai-starcraft-ii-demonstration-tlo-mana/ What do you mean when you say it would perform better? If winning games is the performance you are after then whatever way that is achieved (except cheating/breaking the game) should be considered a success. Since the training of Alphastar is performed by letting different agents of itself play against eachother and eliminating the weaker versions I would argue the claim that it would disregard the macro for the micro is also false. Maybe it would tend to excel in micro in the first variants of itself, but when two agents with comparatively similar micro play against eachother the version better at the macro game would win and be kept for the next round. Remember that the versions of Alphastar Mana played against had been in training for 200 years in the virtual world. With that said it might be the case Alphastar without any apm restrictions could be worse. Simply because the number of actions possible would grow tremendously and that could make it harder for the AI to chose actions. I doubt this would happen though and believe it's much more likely it would simply be a much better agent than the current apm restricted one.
@@Doktorn Noice guys, very cool. I just watched the mana game. Was quite nice, exactly what i was looking for, stalker micro play. But every game is fascinating. I really liked the game where it just built more bases when canon rushed. At last i think Doktorn is right, but it is interesting how the ai changes its behavior in apm restricted games. I would love to see more new ideas for alpha star to act like e.g. canon rush :-) Any more videos you guys would recommend? Thank you
@@LaughNgamez Basic Attention Token, it's a crypto currency which Brave browser uses to allow their users to specifically support the content they enjoy and consume. Let me know if you need/want more details.
The last game. --SpoilerStop-- . : : : : I think that was not a ragequit but Grandmasterquit .. very normal thus bit uninteresting and the GM player was savvy enough to recognize he's behind (but maybe not that he's playing against Alphastar). IT may also be like quick show how the Alphastar "normally" plays TvP when it's winning against protosses that play pretty much "standard "which I suppose it must be able to do to get to grandmaster with Terran.. Interesting that alphastar is so big on stealth units like Banshees and DTs .. With the zerg .. Max Queen-> lingflood -> ravager/queen.. 'it might be that it his bit inhuman biling and transfusing skills.. It can read when something needs healing quicker and maybe where a moving target is going to be when a bile lands better than almost any human.
ive noticed Alphastars rav biles are terrible at hitting moving targets. I always observe it shooting a bile at a unit's current location instead of it's projected location. Transfusion however is top notch with it's ability to recognize the frame before a unit dies and act on it.
Well - perhaps AlphaStar does not bile with prediction but it's biling may be more efficient shooting a new bile whenever ready keeping kind of constant barrage and probably dealing more damage per second per ravager than even really skilled human players.. Since bile is not energy ability but cooldown based each second it's unused reduces damage per second and humans hawe hard time keeping track of when every darn ravagers bile is cooled down and ready.. In any case It does seem to me by the few videos I'we watched that ravagers are preferred by AlphaStar and probably more powerful when controlled by it than when humans at least by a bit.
The exploiters never really got to "grandmaster level" in the internal AlphaStar League, if I read the paper right. Probably had trouble learning precisely-optimal building and probe placement, similar to trouble with walling off....
@@AySz88 in other words, insufficient training, as they were only trained for 44 days with limited number of low-level exploiters . They reached 98 percentile and decided to pull the plug, as training costs millions per day in compute resources.
Short Story about Jimmy, Mike and Steve.
Mike: "Hey Jimmy, that terran has a Banshee"
Jimmy: "No worries Mike, we have Stalkers!"
Mike: "Jimmy, where did that Banshee go?"
Mike : "Also Steve just died!"
Mike: "Jimmy?!"
Im always cheering for Alphastar.... But should I be?
Is this not like cheering for future skynet?
Maybe more like Future skynets grandmother. .
Nope fuck all humans. GO ALPHASTAR!
Yeah, I always cheer for Alphastar, especially against cheese. Fuck those guys lol.
Alphastar just plays like no human which makes it so interesting, and hilarious, when it derps.
Can't be number 1 forever.
Cool idea for a cast. Enjoyed all 3 games
The short games are really interesting as well!
Just it really hurts to watch AlphaStars bad micro against cannon-rushes! Half of the workers it pulls are idle without attacking the pylons that are finishing. 😢
It does that every time, when a cannon finishes, not just in the game we saw now.
3rd video rage quit LOL... now needs a new monitor.
Nice. Any games with a non apm restricted alpha 🌟?
Really doubt it. The whole point from Deepminds perspective was to develop an AI playing at human apm, so the AI has been developed with those restrictions from the start. And with no apm restrictions it would be a walk in the park for an AI vs humans. Search for the video "Automaton 2000 Micro - Dodging Siege Tanks" and the see the difference between 100 zerglings (not) perfectly microed vs 20 siege tanks.
Doktorn thanks for the video
more or less, look for "alphastar vs mana game 4"
it has like at one point over 1500 apm.
keep in mind that giving the AI more apm does not mean it will perform better, the opposite is actually true since it doesnt see the need for a good macro/strategy and instead relies on its micro.
@@DerDevil1998 The version of Alphastar which Mana played against still had restricted apm. And had lower mean apm compared to Mana. See: www.engadget.com/2019/01/24/deepmind-ai-starcraft-ii-demonstration-tlo-mana/
What do you mean when you say it would perform better? If winning games is the performance you are after then whatever way that is achieved (except cheating/breaking the game) should be considered a success.
Since the training of Alphastar is performed by letting different agents of itself play against eachother and eliminating the weaker versions I would argue the claim that it would disregard the macro for the micro is also false. Maybe it would tend to excel in micro in the first variants of itself, but when two agents with comparatively similar micro play against eachother the version better at the macro game would win and be kept for the next round. Remember that the versions of Alphastar Mana played against had been in training for 200 years in the virtual world.
With that said it might be the case Alphastar without any apm restrictions could be worse. Simply because the number of actions possible would grow tremendously and that could make it harder for the AI to chose actions. I doubt this would happen though and believe it's much more likely it would simply be a much better agent than the current apm restricted one.
@@Doktorn Noice guys, very cool. I just watched the mana game. Was quite nice, exactly what i was looking for, stalker micro play. But every game is fascinating.
I really liked the game where it just built more bases when canon rushed.
At last i think Doktorn is right, but it is interesting how the ai changes its behavior in apm restricted games.
I would love to see more new ideas for alpha star to act like e.g. canon rush
:-)
Any more videos you guys would recommend? Thank you
Bet that first one was florencio
3 likes or 42? Uhh... Not sure what to do...uhhh... LIKE!
I've been enjoying your casts.. have some BAT
BAT sorry?
@@LaughNgamez Basic Attention Token, it's a crypto currency which Brave browser uses to allow their users to specifically support the content they enjoy and consume. Let me know if you need/want more details.
Gg alphastar
The last game.
--SpoilerStop--
.
:
:
:
:
I think that was not a ragequit but Grandmasterquit .. very normal thus bit uninteresting and the GM player was savvy enough to recognize he's behind (but maybe not that he's playing against Alphastar). IT may also be like quick show how the Alphastar "normally" plays TvP when it's winning against protosses that play pretty much "standard "which I suppose it must be able to do to get to grandmaster with Terran.. Interesting that alphastar is so big on stealth units like Banshees and DTs .. With the zerg .. Max Queen-> lingflood -> ravager/queen.. 'it might be that it his bit inhuman biling and transfusing skills.. It can read when something needs healing quicker and maybe where a moving target is going to be when a bile lands better than almost any human.
ive noticed Alphastars rav biles are terrible at hitting moving targets. I always observe it shooting a bile at a unit's current location instead of it's projected location. Transfusion however is top notch with it's ability to recognize the frame before a unit dies and act on it.
Well - perhaps AlphaStar does not bile with prediction but it's biling may be more efficient shooting a new bile whenever ready keeping kind of constant barrage and probably dealing more damage per second per ravager than even really skilled human players.. Since bile is not energy ability but cooldown based each second it's unused reduces damage per second and humans hawe hard time keeping track of when every darn ravagers bile is cooled down and ready.. In any case It does seem to me by the few videos I'we watched that ravagers are preferred by AlphaStar and probably more powerful when controlled by it than when humans at least by a bit.
@@Jawzah you make a good point. I'll keep an eye on that as i watch more AlphaZerg
deepmind didn't train alphastar w/ grandmaster level canon rush exploiters!?
It only played against itself, therefore if it doesn't do cannon rush itself, it would have no experience with it.
@@TosiakiS th-cam.com/video/KPLYhRBCcvk/w-d-xo.html
The exploiters never really got to "grandmaster level" in the internal AlphaStar League, if I read the paper right. Probably had trouble learning precisely-optimal building and probe placement, similar to trouble with walling off....
@@AySz88 in other words, insufficient training, as they were only trained for 44 days with limited number of low-level exploiters . They reached 98 percentile and decided to pull the plug, as training costs millions per day in compute resources.
Alpha star is like .. Wait Parting is not EU -.-
Please, I like your casts, but really do something about your keyboard, it's like hearing somebody use his nails on a chalkboard...
Is it really that hard to not speculate or hypotesize constantly, and instead do some actual commentary? It is tedious as h*ll to listen to :/ .