Semantic Anchors in In-Context Learning: Why Small LLMs Cannot Flip Their Labels

Open in new window