Tomorrow the @OpenAI Five bots compete against some of the world's best @DOTA2 players in a benchmark match. Many friends asked me what to pay attention to during these match, so here's a beginner friendly collection of what's most likely in store!http://smerity.com/articles/2018/n_things_to_look_out_for_in_openai_benchmark.html …
-
I wonder if
@GaryMarcus,@yudapearl etc would care to predict if winning an imperfect information game like this is achievable without the progress they advocate in their papers and books? It seems like a perfect situation to test the necessity of eg counter-factual reasoning.
one key variable (i have never played DoTA) is how open-ended the game is. poker has incomplete information, yet is solvable, because the world is still closed-end; brute force can thus still be used.
-
-
A rough idea of task complexity would be controlling the players in a real world game of basketball, if your and your opponent could select your 5 players from a roster of 18 available athletes. Each AI controls a single player, so they have to cooperate with each other etc.
-
Now add in only being able to see in a ten foot circle around your player, and constant pitch invasions from local cats and dogs. It is so absurdly beyond brute force, it isn't even worth mentioning. Every single instant must offer millions+ of options for each player.
- 1 more reply
New conversation -
-
-
Ask
@Smerity how much training time would have been needed to come up with a brute force solution given all the combinatorial variables.Thanks. Twitter will use this to make your timeline better. UndoUndo
-
Loading seems to be taking a while.
Twitter may be over capacity or experiencing a momentary hiccup. Try again or visit Twitter Status for more information.