Few Shot learning enables adaptive argument parsing

One can use a language model in a few-shot regime to extract items specified in a natural language command. First, concatenate many examples of commands followed by the type of item being sought and its actual value from the text. Second, append the query and the new sought item type, and generate the answer.