> Question for OpenAI: Is it more accurate to think of the bots as 5 separate minds, or a single mind controlling 5 heroes? They released their architecture: h

> Question for OpenAI: Is it more accurate to think of the bots as 5 separate minds, or a single mind controlling 5 heroes? They released their architecture: https://www.reddit.com/r/MachineLearning/comments/9533g8/n_openai_five_benchmark_results/e3q0f9l/?context=1 https://www.reddit.com/r/MachineLearning/comments/9533g8/n_o... In the above thread, a reddit user noted that 512 out of the 2048 unit input into the LSTM of each bot is shared (max pooled across players). This means they are telepathically linked and never need to worry about communicating, disagreeing, etc. They know how the others are interpreting their local inputs because it's explicitly shared. So it's not really fair to call it 5 separate minds. You can't really call it a single mind either because if you assume the LSTM is the "mind" (since it's the only place that has memory of previous state), that state isn't read by any other bots.