Meme 1: Marcus was foolish to claim in 2015 that pure #deeplearning systems can't understand novel sentences; they already do
Meme 2: Marcus is just saying what we already know: DL doesn't yet understand language
These can't both be right
-
Another "understanding is binary" fallacy. Deep Learning systems can understand novel sentences well enough to translate them into multiple foreign languages. But it turns out this understanding is very shallow. Much more work to do!
i am fine with deep learning has achieved a “very superficial but nonzero level of understanding” but then the Marcus 2015 quote stands, since that is certainly not at all what i was talking about.
-
-
I feel even translation lacks understanding as it lacks meaning based translation. Even the ranslation accuracy depends on direction of translation. Eg. In many cases lang A to B translation is not the same as B to A for same sentense.
-
Secondly understanding is not application specific. If a system understands a sentense, then it should be able to answer question based on it as well as should be able to translate. We cant say the system understands enough to translate but cant answer questions over it.
- 9 more replies
New conversation -
-
-
Just completed reading your book, the most resonating idea in your book is that true AI will be an ensemble of methods, including deep learning/classical ML, causal analysis and common sense..., there will be no A master algorithm to fit all, a truly AI would be hierarchical.
- 1 more reply
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.