Completely agree with this. #AIhttps://twitter.com/GaryMarcus/status/1186068158697758720 …
-
The biggest irony for me is that there *is* at least one RL solver for Rubik's Cube but
@OpenAI apparently couldn't be bothered to lit search for it (or even google "reinforcement learning rubik's cube").https://www.engadget.com/2019/07/17/ai-rubiks-cube-machine-learning-neural-network/?guccounter=1&guce_referrer=aHR0cHM6Ly93d3cuZ29vZ2xlLmNvbS8&guce_referrer_sig=AQAAALb8aedjf6p55T-chbV4jRrSx_a51_L2RRITDgzSXLL2FQ6KAqFEy4Jvo8Gy3TYrRnrt0zXUeEiU-Mtp2jpZ-nE7wdT20havXMpTLscu1rf0crvbCb6EA08978orJXUFAHEAy77Oz8sbhlUpS-dhQU64hA0N5iqoi2OSgRHFJsRy …
If that's the paper I think it is, it's actually a hybrid model, with a symbolic MCTS algorithm as a key component.
