Gary Marcus

@halhod here’s example of someone walking away from @openai press believing that a network was TRAINED to SOLVE rubik’s and thinking that is a huge deal. but the Solving wasn’t a network and there was no training & it’s not really a huge deal, as most experts would tell you.

FYI I’m senior policy advisor for OpenAI so my assessment of importance of this work is based on watching the project. So maybe I’m an example of someone who thinks it’s a big deal but not of someone who is walking away from OpenAI press with wrong idea about what the project is

Ok, what is that you mean by "this is happening", then? and where do you think the advance is here relative to the extant literature? why is this is a significant advance?

For what it's worth Gillian, it *is* unclear to me from your phrasing here (https://twitter.com/ghadfield/status/1184657579223457792 …) whether when you wrote "this is learned" you meant "this"=cube-solving is learned, or whether "as" in the next clause "...as a pretty difficult feat" means "this"=manipulation.

(Although it *is* pretty obvious, in my read of it, that the thing you're you're impressed with and believe is difficult is the manipulation, not the figuring out of which square goes where on the cube.)

Yes it’s the learning of position in space and how to manipulate. Not the puzzle solving