[jefb]

Sure, take some pot shots - some are valid criticisms - but OpenAI's Rubic's cube solver being lame does not mean AI needs to be re-evaluated.

Sure, AI has its faults; the tantalizing cost savings of automation has created some negative feedback loops - might that be more deserving of the question "what in the hell are we trying to accomplish and how exactly did we get here in the first place?"

A Rubic's cube solver is the problem? Really?

OpenAI (of now infamous Rubic's cube failure :p) released a hide-and-seek demo a few months back that gave me literal goosebumps. Little AI agents facing off in a game of hide and seek start evolving with seriously clever strategies. According to the author's bio (dynamic, time-aware ML systems, etc.) that sort of thing should be right up their ally!

Instead we get some sort of selective self-promotion hit piece - highlighting anecdotal failures while claiming some better AI based robotics startup is coming soon(tm).

[JonathanFly]

>Sure, take some pot shots - some are valid criticisms - but OpenAI's Rubic's cube solver being lame does not mean AI needs to be re-evaluated.

There may be genuine criticisms of that particular project, but 'only the actual solving is done via symbolic methods' is a non-sequitur. The Rubik's cube is just a generic physical task that requires dexterity, they could have done the same research with dominoes or blocks or playing Tic-Tac-Toe with random pens in various adverse conditions -- the point wouldn't be that the ML solves or doesn't solve the actual Tic Tac Toe!