flâneur

2501.14926

arxiv.org · 8,138 words · saved by 1 readers

N/A

Interpretability in Parameter Space: Minimizing Mechanistic Description Length with Attribution-based Parameter Decomposition Dan Braun∗ Lucius Bushnaq∗ Stefan Heimersheim∗ Jake Mendel arXiv:2501.14926v4 [cs.LG] 7 Feb 2025 Lee Sharkey† Apollo Research‡…

saved by

related reading