For most problems where deep learning has enabled transformationally better solutions (vision, speech), we've entered diminishing returns territory in 2016-2017
Show this thread
The next wave of progress will deal with what deep learning cannot handle today -- reasoning and abstraction. But the thing is, human-level general capabilities are not one wave of innovation away, nor three
-
-
I wonder if there is an incremental path to reach such a goal, or it’s a matter of discovering a totally new way of orchestrating neural networks.
-
I think it will happen more suddenly than incrementally, but the build-up to get there will involve a long series of discontinuous breakthroughs, each shining light on one important aspect of the problem
End of conversation
New conversation -
-
-
If I am an investor who just throws hundreds of million dollars into startups, like
#elementai, betting some breakthrough, this statement sure sounds bad.Thanks. Twitter will use this to make your timeline better. UndoUndo
-
-
-
Maybe what we need to improve more than neural nets is simulation, which can be implemented as neural nets or something else, like, games and physical models. Simulation is essential for an agent to act intelligently and reason over the external state of the environment.
-
You are correct in that developing AI will require developing its learning environment as much as its learning algorithm
End of conversation
New conversation -
-
-
Can we model acoustic and olfactory senses in our DL Models?
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
-
-
Indeed. I’m intrigued by
@demishassabis and@DeepMindAI’s strategy of attacking this through first narrow vertical challenges, simulation, and then seeking to use the algo in parallel verticals. AlphaGo to AlphaZero, as an example. -
Agree, AlphaZero's success was surprising and perhaps unintended progress towards more generalized intelligence
End of conversation
New conversation -
-
-
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
-
-
Or things like keras and fastai that make the state of the art accessible to a wider audience
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.