Does @fchollet think that deep reinforcement learning is "intelligent"? No he doesn't. And he's right
https://www.youtube.com/watch?v=VvRz885Un1Q …
-
-
It would still not capture other aspects of human cognition, like perception, behavior, goal-setting, and so on. Any intelligence benchmark should be an ongoing effort, to be refined as the flaws of the previous iteration become more apparent.
-
Fully agree. The more attention the benchmark receives the better it can be improved and bigger its push for new directions in development. Hopefully, current level, while difficult, can be fully solved with basic reasoning capabilities promoting gradual improvement
End of conversation
New conversation -
-
-
Great, I will adapt my Sokoban solver for ARC (it could in principle do both but shortcuts save time). It would require no ‘training’, which I always find a strange thing in ML. It is an open ended system, which works till a solution is found or the solution space exhausted
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
-
-
Have you written anything up on the ARC challenge on Kaggle? Would love to know how it turned out!
-
(Asking
@fchollet I guess) I was also wondering whether he had an idea of how to solve the challenge - Show replies
New conversation -
-
-
are you concerned that solving ARC requires a highly focussed type of response. a baby learns by convention to label a yellow curvy thing as banana but there are many other possible labels. A baby GI would look at ARC challenges and form many other meaningful interpretations.
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
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.