Without knowing what Sutskever had in mind, I think that depends on what you mean by "well enough". The word would have to be predicted based on a unified understanding. Current AI is not focused on producing unified models of the universe. Every sentient being has such a model.
here we go again: Sutskever to @NewYorker: "If you train a system which predicts the next word well enough, then it ought to understand"
ML fans when I show GPT-2 is clueless: GPT-2 is just system for predicting next words; not fair to ask more.https://twitter.com/Plinz/status/1194391612735942656 …
-
-
-
agreed this current AI lacks models of the universe; the whole problem is that the field has lost sight of this and thinks that success on some benchmarks means much more than it really does.
- 2 more replies
New conversation -
-
-
Interesting conception of "ought"
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
-
-
current A.I. is an insult to I.
- 1 more reply
New conversation -
-
-
Can someone just train this thing on
#Cleverbot data so we can all see how bad a chatbot it makes?Thanks. Twitter will use this to make your timeline better. UndoUndo
-
-
-
Word to word to word, the best sentences, follow a new pattern. A pliable mind, freshened by novelty, accepts what can’t be predicted.
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
-
-
Also a problem of regularization/ loss function. Most "next word predicting" is very mechanical & based purely on extremely local state (last couple of words). Which is very different to sentences like "the meaning of life, the universe and everything is <?>" or "317 / 87 = <?>".
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.