A researcher says modern neural networks are far more interpretable than believed, with most neurons understandable in models up to trillions of parameters (Nick Cammarata/Goodfire)

Nick Cammarata / Goodfire:
A researcher says modern neural networks are far more interpretable than believed, with most neurons understandable in models up to trillions of parameters  —  The cause of polysemanticity in early model organisms wasn't superposition, but several ordinary, mostly fixable issues.

原文：https://www.a8q9.cn/en/article/a-researcher-says-modern-neural-networks-are-far-more-interpretable-than-believe-6ca4d7.html