Indeed, once upon a time, Hinton was interested in the same sorts of unification that I believe to be so important.https://twitter.com/danbri/status/1075653592596377600 …
-
Making good progress! https://twitter.com/santoroAI/status/1017060254855630849 … https://twitter.com/santoroAI/status/1076033843080105986 … I would recommend paying attention to this stuff (which as
@ylecun said, Geoff has been arguing for all along), because my bet: it's gonna settle this long running debate in the next few years.
This does look interesting. But you can’t both train a system “to predict symbolic explanations” and argue that symbols are a waste of time. Over time, Hinton has taken two different, mutually exclusive positions at most only one can be correct.
-
-
In my personal conversations with him, he's always been consistent: you want to be able to do things like symbolic reasoning using distributed systems that learn. I think sometimes his true opinion gets obscured by the natural tendency we all have to frame things in binaries.
-
After all, I don't see how anyone can truly deny that humans do something like symbolic reasoning. What is math if not that?!
- 9 more replies
New conversation -
-
-
I think the argument is that symbolic *methods* are the "waste of time", not symbols (localist representations?) per se.
- 4 more replies
New conversation -
-
-
you can train a system to predict movie ratings and yet argue that movies are a waste of time... Symbolic communication and rules are an important phenomena *in the world*, and like cinema attendance, comes with predictable structure.
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.