If anyone’s going to wilfully misinterpret an AI result for increased visibility, it’s him. He’s become his own hype-generator.

It's much less funny than inventing "pre-AGI" though. Clearly you like manipulating hype than robot hands. So such criticism is healthy.

I guess it's fair since the title was not "Learning to Solve...", more fair is "Learning to Manipulate...". Researchers get that end-to-end combinatorics is a VERY difficult problem, but I wonder how many people think that a neural net came up with a Rubik's cube solver?

Ok, so if you made a robot hand that makes physical moves in the game of go, but the logical moves are provided by alpha zero, would it be fair to say that your task is playing(/solving) go?

He is 100% right though.

I thought the video was about the perturbation theory ;D

You, of course, know the word "solves" implies much more, and leads to viral behavior. Maybe a warning for journalists to read the paper first would be the professional thing to do.

Marcus' criticism is wrong. "Solve the Rubik's Cube" means to change from the first state to the last state. The first state is color misalignment. The final state is that the colors are aligned. This statement is appropriate and not hype.

IMO you're getting blowback from hyping GPT-2 as the most dangerous AI evah(TM). That said, perhaps if you had framed this as using a classical algorithm to plan moves for Rubik's Cube and reinforcement learning to train a robot hand to do those moves, you'd have avoided this?

Is just about not overselling research, feeding the hype just make the actual targe harder to be achieved