In-Context Learning (and Unlearning) of Length Biases

Stephanie Schoch, Yangfeng Ji

Large language models have demonstrated strong capabilities to learn in-context, where exemplar input-output pairings are appended to the prompt for demonstration. However, existin…