romptypes

Promptypes is a term used to describe a taxonomy of prompt types used to elicit targeted responses from large language models. The concept groups prompts by structure, intent, and the constraints they impose on model behavior, with the aim of guiding output quality and reliability.

Common categories include instruction prompts (clear tasks with explicit instructions), zero-shot prompts (no examples provided), few-shot prompts (a handful of exemplars), chain-of-thought prompts (requests for step-by-step reasoning), role prompts (assigning a persona or authority), and constraint prompts (imposing formatting, length, or stylistic constraints). Template or pattern prompts, which provide reusable structures, are also used to promote consistency across tasks.

Designers evaluate prompts by clarity, brevity, and the balance between guidance and model freedom. Other considerations include robustness across domains, sensitivity to minor wording changes, alignment with safety policies, and the potential for prompt leakage or unintended biases. Prompts are often iteratively tested and benchmarked to compare performance on specific tasks such as accuracy, helpfulness, and factuality.

Promptypes are used in research and practice to organize prompt design, compare approaches, and share reusable patterns. The taxonomy is not fixed; new categories and patterns emerge as models evolve and new applications appear. As prompting methods mature, practitioners increasingly document preferred promptypes, correlate them with outcomes, and develop standardized evaluation frameworks.

Related topics include prompt engineering, natural language processing, and AI safety and alignment.