\n\n

Researchers are making progress in mechanistic interpretability, the field that aims to open the black box of large language models. Recent work shows that even at the scale …