Seems very plausible (allowing for algorithmic advances to decrease the amount of data required). Important to think about the implications... nitter.vloup.ch/metaversefight/s…

Apr 20, 2019 · 4:29 PM UTC

3
4
53
In particular, with this tech it’s easy to imagine a whole new class of smart programs which can work well and interact with users — whether in the domain of robotics or virtual assistants. Super exciting on the upside!
1
3
36
Replying to @gdb
Just curious, would it be possible to make a generalizable classifier for whether a data point will have useful impact when training a function approximator?