I know your not that dumb that you don’t understand the difference between long term goals and the specific goal they had in mind for Alphastart which they CLEARLY state in their paper.
Take the loss and move on, stop trying to sound so smart.
Yet you are dumb enough to think that I am referring to their short-term goals even though I’ve specifically linked you to their long-term goals. This AI accomplishes virtually nothing towards their long term goal and linking to their short-term goals does nothing to change that fact.
You even stated it in your OP, Alphastar is not them claiming success in their long term goal.
Except their marketing team did, before issuing a retraction, which is what this is referring to. When it beat TLO, they claimed it had “mastered” sc2. You can’t even claim that this thing is Grandmaster like they have.
1 Like
AlphaStar is the first agent to achieve Grandmaster level in StarCraft II, and the first to reach
the highest league of human players in a widespread professional esport without
simplification of the game. Like StarCraft, real-world domains such as personal assistants,
self-driving cars, or robotics require real-time decisions, over combinatorial or structured
action spaces, given imperfectly observed information. Furthermore, similar to StarCraft,
many applications have complex strategy spaces that contain cycles or hard exploration
landscapes, and agents may encounter unexpected strategies or complex edge cases when
deployed in the real world. The success of AlphaStar in StarCraft II suggests that generalpurpose machine learning algorithms may have a substantial effect on complex real-world
problems.```Yes apparently you haven’t seen the part of my argument which details that such a small selection of games isn’t enough to support their claim that it is Grandmaster level. It lost to a masters level player while being shredded by lurkers and refusing to make an observer. No rational person would look at this bot’s play and conclude that it is 6k mmr.
This is one of the dumbest things I’ve ever heard on the forums.
The account is literally grandmasters, yet you claim its not? Get over yourself, these people made some real progress, stop trying to bash them just to sound smart on the internet.
Uh oh, someone doesn’t understand that win-rates fluctuate with the meta and that peaking in Grandmaster doesn’t mean someone is Grandmaster level. Too bad he didn’t bother reading my arguments before deciding that I was wrong!
3 accounts in grandmasters
not grandmaster level
I read your dumb argument, I wish I could get those 10 seconds of my life back.
I also read the published paper, something you obviously didn’t do.
Ok, so you legitimately don’t understand that performance fluctuates and that 100 games isn’t enough to get a good average given the fact that it fluctuates.
You could easily do some hypothesis testing if you wanted to prove that their MMR/winrate/rank is not actually grandmasters. Then just figure out the odds of this occurring 3/3 times. I’m not going to do this because its obviously not by chance, your welcome to try though.
Ok, lets hold hands and do some basic logic. The meta fluctuates. The meta is the total sum of all possible strategies and how likely they are to be used. The meta fluctuates meaning the composition of strategies is always changing as players are trying to counter one another. This means a given strategy’s win-rate will also fluctuate depending on the composition of strategies that it faces vs. When the meta is full of strategies that are weak vs this strategy, the win-rate will go up, and when the meta is full of strategies that are strong vs this strategy, the win-rate will go down.
Win-rates also fluctuate based on your opponents. This thing has 6k mmr, but losses to masters-level (<5.3k mmr) players who go lurkers. This clearly tells us their sample has not included enough games vs lurkers. When this strategy faces vs opponents who do a strategy that is strong vs this strategy, the win-rate will go down, and when it faces vs opponents who do a strategy that is weak vs this strategy its winrate will go up.
Win-rates fluctuate and peaking in GM after 100 games is not a large enough sample.
1 Like
Obviously the winrate fluctuates, which is why its possible that the true winrate is good enough to be GM level while still losing to masters level opponent.
You are literally arguing against yourself.
Ok, then we agree that their claim that it is GM level is not substantiated by the merit of a couple hundred games.
Lol, you and I both know that you could easily disprove this if it wasn’t true. So quit dodging.
You are the one who is claiming that a bot is 6k mmr when it doesn’t know how to make an observer and walks his entire army over lurkers vs a 5.3k opponent. You need to realize that their sample didn’t include enough games vs players who go lurkers.
“You could easily do some hypothesis testing if you wanted to prove that their MMR/winrate/rank is not actually grandmasters. Then just figure out the odds of this occurring 3/3 times. I’m not going to do this because its obviously not by chance, your welcome to try though.”
You need to realize that their sample didn’t include enough games vs players who go lurkers. If it did, the bot would have <5.3k mmr.
"The Battle net matchmaking procedure selected maps and opponents. Matches
were played under blind conditions: AlphaStar was not provided with the opponent’s identity,
and played under an anonymous account. "
This is why its important to read the paper.