Hi — I saw you're working on sparse autoencoders for mechanistic interpretability.
I’ve been experimenting with a small idea that makes internal signals in CNNs “locatable” (A0 → A1 → A2).
Not sure if it connects, but I’d be curious what you think.
https://github.com/luolearning/luoshu_kit
Hi — I saw you're working on sparse autoencoders for mechanistic interpretability.
I’ve been experimenting with a small idea that makes internal signals in CNNs “locatable” (A0 → A1 → A2).
Not sure if it connects, but I’d be curious what you think.
https://github.com/luolearning/luoshu_kit