Recently, a boom of papers has shown extraordinary progress in zero-shot and
few-shot learning with various prompt-based models. It is commonly argued that
prompts help models to learn faster in the same way that humans learn faster
when provided with task instructions expressed in nat