Why in-context learning models are good few-shot learners?

Best AI papers explained - Why in-context learning models are good few-shot learners?