Eric Michaud on scaling, grokking and quantum interpretability
The Inside View
00:00
The Importance of Generalization in Network Learning
So yeah you've been publishing two papers on in the rocking recently. Why are you so excited about rocking? So I don't know, rocking it kind of exciting just because it's weird and like surprising. In our first paper we like looked at representation learning and how sort of generalization in these networks on like the math operations that they're learning depends on the networkLike learning is like particular like structured representations of their inputs. It seems like maybe the networks learn something similar here.
Transcript
Play full episode
