Can Input Attributions Explain Inductive Reasoning in In-Context Learning?

Mengyu Ye, Tatsuki Kuribayashi, Goro Kobayashi +1

Interpreting the internal process of neural models has long been a challenge. This challenge remains relevant in the era of large language models (LLMs) and in-context learning (IC…