Do different prompting methods yield a common task representation in language models?

Open in new window