Bayesian scaling laws for in-context learning

Aryaman Arora, Dan Jurafsky, Christopher Potts +1

In-context learning (ICL) is a powerful technique for getting language models to perform complex tasks with no training updates. Prior work has established strong correlations betw…