Why a "database"? How is all of the incredible work on embeddings not storage and representation of knowledge? Just because it's in a form that we humans can't directly interpret (like, need I point out, is true for our OWN brain's knowledge "database") doesn't mean it's garbage
Bingo. So many people think they can do NLU (or robotics, or even driving) without having a commonsense knowledge database, and I just don’t see how that’s remotely possible.https://twitter.com/punkstrategy/status/1071560512662515712 …
-
-
-
a. did i say it was garbage? or just not adequate for the task? b. have a look at BERT, best embeddings Google can buy, still can’t do Winograd Schemas or discourse comprehension c. knowledge base would have been better; i don’t literally mean SQL etc
- 3 more replies
New conversation -
-
-
I think this was forgivable 3-5 years ago. Even I was open to the possibility that appropriately guided machine learning could create human-equivalent automated driving. But now it's clear that some unknown breakthrough is needed for that to happen.
-
It has been known since AI Winter #1 but it gets ignored a lot. Plenty of examples. Researchers foolishly think "If only I had more data!"
- 2 more replies
New conversation -
-
-
Until AGI “knows” that gravity falls down and no matter what strange image is presented. Certain fact are imputable we will not proceed with AGI. After AGI knows about the concrete physical world then it can be trained to imagine a world with different gravity, but not until
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
-
-
also figuring in possibility that evolution has had millions of years to give us some commonsense priors
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
-
-
I can see why DL researchers want to capture these relationships somehow without an explicit database. For practical purposes in NLU agree it is necessary today. There may be some ways to encode “common sense” knowledge into networks as forbidden or highly unlikely patterns...
- End of conversation
New conversation -
-
-
SQL'99 or SQL:2016?
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
-
-
I’d say the KB may even be secondary, but there’s an overlooked need for a representational mechanism that many “neural” NLP systems lack; a positive exception is work at Groningen by the group of Johan Bos and their DRT work (Groningen Meaning Bank).
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.