How much processing power is needed to run the agent once it is trained? It would be interesting for game developers to use reinforcement learning to train in-game AI's for different difficulty levels. Could also be useful for testing game balance.

Has restricting the hero pool helped train faster?

Is there any restriction in the game? If does, can I see reference? And Will openai five be modified b4 the tmr match?

Why didn't the bots draft like previously?

You will be playing one of VG or Winstrike? who got eliminated today..

With hero picking or without? Can you tweet somewhere the win probability through the game?