Lmao...I was just thinking "dang it why do i think that is programmers everytime at 1st glance?" as i scrolled down to look at connents and this was first comment i see lol
That 3rd game is such a dominant win. All the tactical decisions of Alphastar (retreating from its early aggression, immediately counter-attacking after manhandling Mana's army, even the scouting Phoenix) seemed spot on.
Alphastar uses a different style in each game (in these pre-ladder matches) because it is actually 5 different agents (selected from a single Alphastar league). In contrast, the ladder matches had only 3 agents (one per race) playing 50 games each. I mean three leagues (TLO, MaNa5, and MaNa1) pre-ladder and then three leagues (early, mid, late) ladder.
I love watching Bronze League games. They might not be up your alley, but Winter & Husky seemed to have a great time with them. Perhaps you might consider doing them as well?
Is it possible to play bots and bots in Starcraft? If not, can you play you vs bot and review the game The goal: explain what's different between Alphastar vs the bot that come with the game, and how to beat the bot.
Anyone know how much hardware is needed to run AlphaStar? Wondering if it would be possible to run on a gaming desktop if they would release a version for us to openly play against.
From alphastar blog; "The AlphaStar league was run for 14 days, using 16 TPUs for each agent. During training, each agent experienced up to 200 years of real-time StarCraft play." That's for training though, running a trained agent is much easier, again quoting from alphastar blog; "The final AlphaStar agent consists of the components of the Nash distribution of the league - in other words, the most effective mixture of strategies that have been discovered - that run on a single desktop GPU."
@@KCM25NJL Running the agent once it's trained is like running a 2020 game with everything on Ultra. Neural network AI like this is extremely GPU intensive and requires top of the line hardware to run.
@@michaelbuckers I mean, you just couldn't be more wrong. But since you are arguing against the point, I'm willing to accept your evidence as proof hitherto. Alphastar's trained agent could literally play on a decent spec laptop.
![AlphaStar vs MaNa - Starcraft 2 [Part 2]](http://i.ytimg.com/vi/SV2oHK7jFak/mqdefault.jpg)
![AlphaStar vs MaNa - Starcraft 2 [Part 2]](/img/tr.png)
I always read that title as "alphastar vs programmer"
Lmao...I was just thinking "dang it why do i think that is programmers everytime at 1st glance?" as i scrolled down to look at connents and this was first comment i see lol
That 3rd game is such a dominant win. All the tactical decisions of Alphastar (retreating from its early aggression, immediately counter-attacking after manhandling Mana's army, even the scouting Phoenix) seemed spot on.
Alphastar uses a different style in each game (in these pre-ladder matches) because it is actually 5 different agents (selected from a single Alphastar league).
In contrast, the ladder matches had only 3 agents (one per race) playing 50 games each.
I mean three leagues (TLO, MaNa5, and MaNa1) pre-ladder and then three leagues (early, mid, late) ladder.
I'm out of mana
I love watching Bronze League games. They might not be up your alley, but Winter & Husky seemed to have a great time with them. Perhaps you might consider doing them as well?
Great replay!
I need more pro matches and Mana! are there any more matches with Alphastar other than TLO and Mana in the pro scene?
It's been a great series. I hope Deepmind builds another AI or sells it to Blizzard. It is very entertaining to watch it play.
Build a complete wall off and go for air. That's like 101. Fr.
In my best Diablo voice " Not enough MaNa"
Is it possible to play bots and bots in Starcraft? If not, can you play you vs bot and review the game
The goal: explain what's different between Alphastar vs the bot that come with the game, and how to beat the bot.
Maybe interview programers? If you have the connections, you might get the connections if get one interview going.
I NEED MOAR MANA
Need more mana!
Anyone know how much hardware is needed to run AlphaStar? Wondering if it would be possible to run on a gaming desktop if they would release a version for us to openly play against.
From alphastar blog;
"The AlphaStar league was run for 14 days, using 16 TPUs for each agent. During training, each agent experienced up to 200 years of real-time StarCraft play."
That's for training though, running a trained agent is much easier, again quoting from alphastar blog;
"The final AlphaStar agent consists of the components of the Nash distribution of the league - in other words, the most effective mixture of strategies that have been discovered - that run on a single desktop GPU."
RTX or AS? Choices choices.
Running the agent once trained is no more compute intensive than the regular in-game bot. Training an agent however, is hugely resource intensive.
@@KCM25NJL Running the agent once it's trained is like running a 2020 game with everything on Ultra. Neural network AI like this is extremely GPU intensive and requires top of the line hardware to run.
@@michaelbuckers I mean, you just couldn't be more wrong. But since you are arguing against the point, I'm willing to accept your evidence as proof hitherto. Alphastar's trained agent could literally play on a decent spec laptop.
With only 3 games in this part we should aim for 3 likes only! You became too greedy to expect us to give you 6 likes!
I like an idea to stream replays of another bots
we need more mana!
I need more mana!