HoagyC

sparse_coding. Using sparse coding to find distributed representations used by neural networks.

muz. Attempting to replicate EfficientZero

hiddeninfo. Experiments for creating interpretable latent spaces by screening off hidden info.

AutoInterpret. Attempt to interpret neurons automatically using GPT-4. Now obsolete due to openai/automated-interpretability

steering_directions. Python

distrib_feats. Code for "Distributed Features and Non-linearities" post.

compress_check. Python

ARC. Work on the Abstraction and Reasoning Challenge at Kaggle