Rectifying Demonstration Shortcut in In-Context Learning: Introducing In-Context Calibration

Rectifying the reliance of Large Language Models on pre-trained semantic priors in demonstrations through In-Context Calibration.