ICLR Debate with Leslie Kaelbling (ICLR 2019) 

21:06 Jeff Clune at 100% cpu

Neural networks didn't start working well until priors were built into the architecture.

That panel was amazing. I hope there will be more of these! Personally, I like all of the research directions taken here.

At 30:30 -- models can't learn about causality. Hume and Kant already went over this. If you want causality, you have to supply it yourself.

Suchi Saria seems a little out of her depth here.

Joshua Tenenbaum work free for google, same as we people mostly do work for google without payment