Facts: 1. If it was easy to do as pure neural network, OpenAI probably would have done it. 2. The closest I can find is linked, and is a hybrid system, NOT a pure CNN. 3. The current paper misrepresented what they did, as if it were not a hybrid.https://www.nature.com/articles/s42256-019-0070-z …
#DeepLearning #hype alert: @OpenAI’s paper on Rubik’s cube solving starts “We’ve trained a pair of neural networks to solve the Rubik’s Cube with a human-like robot hand”; mentions critical hardwired symbol-manipulating cube solver only at ref. 111https://openai.com/blog/solving-rubiks-cube/ …
-
-
- 1 more reply
New conversation -
-
-
Probably also didn't dwell too much on details of the chips used or the type of flooring in their lab space. They were working on dextrous object manipulation, not solving Rubik's cube.
-
Yeah,
@GaryMarcus seems to be making a bit of a cheap shot here. Missing the achievement in pursuit of decrying hype as part of a narrative that reaffirms his prior views. - 12 more replies
New conversation -
-
-
Also the set up is one hand and like 12 cameras in a non moving glass box. That’s not human like. At all.
-
A very pertinent point. A lot is made here about transferring from simulation to real world, but the physical world setup seems so constrained as to effectively undermine that claim.
End of conversation
New conversation -
-
-
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
-
-
Does the network pair learn to *solve* the Rubik’s cube puzzle or does it only learn how to *manipulate* a Rubik’s cube with a robotic hand?
-
Manipulate only. That's exactly what he's pointing out the paper says when citing ref. 111pic.twitter.com/jjLM7ea9NX
- 2 more replies
New conversation -
-
-
This seems like an odd shot. The linked blog post mentions under the first heading that they use Kociemba’s algorithm for picking solution steps. I never assumed it was about anything but dexterity. I thought the reporting was a step in the right direction for OpenAI.
-
The difficult part was sensing/control policy, but the PR department can still do better. Everything is in the paper (including the high failure rate), but most people will just watch the short promo video with some 'inspirational' statements about future of robotics 1/2
- 3 more replies
New conversation -
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.