Check here our recent #ICML2024 paper on mechanistic interpretability!