41 - Lee Sharkey on Attribution-based Parameter Decomposition

YouTube link What’s the next step forward in interpretability? In this episode, I chat with Lee Sharkey about his proposal for detecting computational mechanisms within neural networks: Attribution-based Parameter Decomposition, or APD for short. Topics we discuss: APD basics Faithfulness Minimality Simplicity Concrete-ish examples of APD Which parts of APD are canonical Hyperparameter selection APD in toy models of superposition APD and compressed computation Mechanisms vs representations Futur...