4 ms·

Are you so sure? APM limits on the AI aren't necessary, which is a real insight FTA: >In its games against TLO and MaNa, AlphaStar had an average APM of aroun

Are you so sure? APM limits on the AI aren't necessary, which is a real insight FTA: >In its games against TLO and MaNa, AlphaStar had an average APM of around 280, significantly lower than the professional players, although its actions may be more precise. This lower APM is, in part, because AlphaStar starts its training using replays and thus mimics the way humans play the game. https://deepmind.com/blog/alphastar-mastering-real-time-strategy-game-starcraft-ii/#image-34426 https://deepmind.com/blog/alphastar-mastering-real-time-stra...

vkou 8y ago
AlphaStar had an average APM of 280, but during battles, it's APM would spike up to ~1500. It's like saying that my car has an average speed of 2mph. Sure, it does - but that's because it spends 23 hours a day sitting in a garage.
- bduerst 8y ago
  Where does it say the APM spiked during battles? If it's using it head-to-head with the human player, then that's bad, but how do you know it's not just herding workers and units individually? I get the analogy but based on the distribution, 75% of the APM for alphastar is below the mean of APM for the human during gameplay. Following that analogy, both cars are on the racetrack, except one is applying acceleration with more precision - i.e. accelerating out of a turn rather than always accelerating. Capping acceleration wouldn't matter in this case.
  - saberience 8y ago
    The APM maxed out at 1500 at one point, far in excess of any human, especially considering many human actions are meaningless actions or spam clicks. The AI could win based on inhuman micro with stalkers and blink. I would be far more interested in seeing the agent with a fully realistic camera and fully realistic apm limit.
    - bduerst 8y ago
      It does make a difference, because if the APM spike is during a head-on interaction with the human then it provides a competitive advantage, if it's randomly during another part of the match, then it doesn't provide a competitive advantage. Where did they say the spike was during a battle?

vkou 8y ago

AlphaStar had an average APM of 280, but during battles, it's APM would spike up to ~1500. It's like saying that my car has an average speed of 2mph. Sure, it does - but that's because it spends 23 hours a day sitting in a garage.

bduerst 8y ago
Where does it say the APM spiked during battles? If it's using it head-to-head with the human player, then that's bad, but how do you know it's not just herding workers and units individually? I get the analogy but based on the distribution, 75% of the APM for alphastar is below the mean of APM for the human during gameplay. Following that analogy, both cars are on the racetrack, except one is applying acceleration with more precision - i.e. accelerating out of a turn rather than always accelerating. Capping acceleration wouldn't matter in this case.
- saberience 8y ago
  The APM maxed out at 1500 at one point, far in excess of any human, especially considering many human actions are meaningless actions or spam clicks. The AI could win based on inhuman micro with stalkers and blink. I would be far more interested in seeing the agent with a fully realistic camera and fully realistic apm limit.
  - bduerst 8y ago
    It does make a difference, because if the APM spike is during a head-on interaction with the human then it provides a competitive advantage, if it's randomly during another part of the match, then it doesn't provide a competitive advantage. Where did they say the spike was during a battle?