This was an absolutely fascinating piece of work. I can only imagine how it must have felt at the beginning, with the scripted bots and how daunting the prospect of playing against human pros must have felt.
I really think the standout point is the fact that the OpenAI team is approaching this immensely complex task in terms of achieving ‘milestones’. Rather than train the bots on a fully complex 5v5 ‘real-world’ environment, they are choosing to progressively increase the complexity of the world the bots see. It shows that the OpenAI team are not greedy for short-term glory, but are rather focusing on pushing the system to its limits and learning what works best, before pushing on to the next level of difficulty. It’s a really heartening approach, and I can’t wait for next year’s international! Who knows where they’ll be??? :D
