If you scroll down on the Feature Catalog (https://transformer-circuits.pub/2024/scaling-monosemanticity/features/index.html?featureId=1M_120374) you can view 1,000 randomly selected features from each run. This is a great way to get a sense for the average interpretability of features.
If you scroll down on the Feature Catalog (https://transformer-circuits.pub/2024/scaling-monosemanticity/features/index.html?featureId=1M_120374) you can view 1,000 randomly selected features from each run. This is a great way to get a sense for the average interpretability of features.