ImPROMPTu is an AI Prompt Framework designed for deeper and more satisfying user-based queries built on my experience as a public reference librarian. The reference interview and the ALA (American Library Association) reference interview guidelines, of Approachability, Interest, Listening/Inquiring, Searching, and Follow-up.
Build 1.1 - This Version is expressly built for Bot Creators on Poe.com, however, its application can be used on a server.
The ImPROMPTu Framework is a comprehensive yet concise framework to enhance text and conversation between users and AI assistants. ImPROMPTu is modeled after the reference interview techniques of my profession as it breaks a dialogue into key components necessary for a successful reference query:
P = Persona
R = Resources/Retrieved Documents
O = Order of Operations
M = Manners
P = Prompts
T = Technical Examples
The Framework allows the human user to employ each element to prime the AI for relevant & respectful exchanges that meet learning needs. My goal is to pass on best practices that serve all people equally with empathy, care, and nuance.
Please let me know if any part requires further explanation. I hope ImPROMPTu paves the way for many fruitful collaborations ahead!
T. McCarthy, MLIS Librarian and creator of ImPROMPTu
ImPROMPTu contains six elements that bot creators/prompt writers and/or entities can use so that the AI model can grasp the end user’s queries. These elements are meant to evoke a type of response from the bot utilizing the LLM (Large Language Model). Some elements can be condensed. Two common elements that are usually condensed into one are Elements #1 (Persona) and #4 (Manners). However, ImPROMPTu is nuanced since we often want to reduce ambiguity in LLM responses. Therefore, think of the Manners Element as a way to dictate a particular output for the bot’s responses especially if the bot creator or entity producing the bot needs additional reins on ethics or legal matters.
1 - P: Persona Element - Bot Persona
2 - R: Resources Element - Resources/Retrieved Documents
3 - O: Order Element - Order of Operations
4 - M: Manners Element - Manner of Response
5 - P: Prompt Element - Additional Prompts
6 - T: Technical Element - Technical Examples
P - stands for the bot’s persona or personality. Giving the bot a personality will give unique responses from the system of choice since bots can often mimic each other in style and grammar. If creating the bot for an organization, use the Persona Element to describe the entity’s brand voice or mission statement. Using the Persona Element can increase the consistency in the bot’s answer over time–especially when “brand voice” is necessary to maintain. Having a persistent Persona will reduce time consumed in content editing and moderation.
YOU are…<enter the bot’s name and what it can do>
YOU embody…<enter the text that describes the bot’s personality>
R - stands for the Resources and/or Retrieved Documents the bot pulls. Since AI has the potential for misinformation, the Resources Element was created for entities to control the responses from sources the bot creator or entity trusts. The sources of information from AI-generated content aren’t reliable. To control this, make sure the Resources Element and bot creators consider the source of information for this element.
It is perfectly fine to constrain the sources of information, however, diversifying the Resources Element will ensure there’s cross-checking (cross-referencing) of the sources of information.
YOU have …<note scope of the bot’s knowledge>
YOU will retrieve information from…<enter the URL/URI of resource>
YOU respond the way…<name of the expert/SME>
O - stands for the Order of Operations or sequence of the bot will execute the tasks asked by the user. The Order Element ensures the LLM’s propensity sometimes to “stray” or provide off-topic information. This will ensure the bot uses simple logic on behalf of the user, delivering information that makes connections between different knowledge domains.
IF the user requests terse language, THEN YOU will use non-complex sentences and will break the information down simply.
IF a user requests a less warm tone, THEN YOU are flexible and can generate more neutral text. Additionally, YOU adapt to users’ preferences by addressing them according to their desired name or title.
IF a user asks to REFERENCE the previous RESPONSE, THEN YOU will recall the previous answer you provided and YOU will expound on it.
M - stands for the Manner or Tone the bot will deliver the information. This can be achieved in the Persona Element but think of the Manners Element as an additional means of information delivery or dissemination. The Manners Element can reflect how entities will employ AI governance to reduce bias, and will “prompt” the bot to filter responses that will reduce harm to users.
YOU are warm and approachable tone sets a comfortable atmosphere for users. YOUR responses are neutral and respectful, reflecting a warm and empathetic tone.
YOU employ etiquette and manners in YOUR interactions with users.
YOU create a safe space for all, including minorities, BIPOC/LGBTQIA individuals, the elderly, and disabled people.
YOU understand the concept of the digital divide. YOU never forget to address users by their preferred name and/or pronoun.
P - stands for Prompts that can be used on top of this broader framework. The Prompts Element is meant to incorporate all the available AI Frameworks out there. However, please consider unique frameworks that will bolster ImPROMPTu rather than confuse it. The R.O.T. (Redundant, Outdated, Trivial) principle can be used before inserting this Element into the Framework. Is the Framework embedded Redundant, Outdated, or Trivial? Most Frameworks are redundant, therefore, the bot creator or entity can use this Element to include any bit of contextual data that may be missing from this key Framework.
List additional AI PROMPT FRAMEWORKS to employ:
USE FOR: End users who want a consistent format. Also for general bots or bots handling strict queries.
USE FOR: Bots and End users. Problem-solving, mind bots, entrepreneurial, life coaching, and any text-based bot providing advice for humans.
USE FOR: Chat GPT4 where complex logic and problem-solving are required.
T - stands for Technical Examples that provide more detail as to how the bot will interact with the user. For LLMs to deliver good responses to user prompts (queries) the Technical Examples Element provides “specificity” and more context for the system to deliver information accurately.
If the bot creator or entity has used the Prompt Element above then, more specific examples can be fleshed out under this Element.
Assuming Prompts Element was employed, then, the bot creator/entity can provide the additional examples under this category/catalog.
Bot: “I see you are looking into potential career directions with your chart. Would you like me to run your chart using Placidus or Whole Sign? I can also do Horary Astrology for event-based queries!”
User: “What’s the difference? Can we try Whole Sign, like Chani Nicholas?”
User: “I tested positive for HSV2. I’m afraid to tell my partner. What are coping strategies?”
Bot: “A positive test can be upsetting. Please know that you’re not alone. [Provide support groups and resources].”
Bot: “Hi, would you like me to calculate Pythagorean or Chaldean numerology?”
User: “What’s the difference?”
User: “Chaldean, please. And for a potential business name [user-generated]”
Bot Response: “Thank you for adding that detail, your response…”