Prompt Development: Beyond the Basics

Watch The Podcast on Acast | Spotify | Amazon | Apple

The essence of prompt development transcends the mere act of instructing an AI. It encompasses a thoughtful exploration of frameworks and methodologies borrowed from diverse fields, including software development, linguistics, psychology, and design thinking. This multidisciplinary approach not only enriches the prompt development process but also ensures a more robust and versatile interaction with AI systems.

Diving into the realm of prompt development, we enter a nuanced space where the construction of prompts transforms into a strategic endeavor, echoing the complexities of architectural design. This segment, therefore, delves into the concept of "Prompt Architecture" or the art of structuring and linking prompts and strategies in a coherent, purposeful manner. Drawing from a comprehensive survey, we see that in Figure 4.1.1 and 4.1.2 that a significant majority, 57% of 164 respondents, favor the term "Prompt Architecture" over "Virtual Brain," with only 24% leaning towards the latter. This preference underscores the industry's inclination towards a more structured, methodical approach to crafting AI interactions.

Figure 4.1.1: Preferences for Terminology in Prompt Arrangement

Participants' Predilection for Terminology Describing Prompt Arrangement and Strategy

Note. Represents survey results from "Exploring the Landscape of Prompting" survey conducted by Jonathan Kyle Hobson. Data illustrates participants' preferred terminology for the specific arrangement and connection of prompts and their strategies. The survey captures the community's leaning towards standardized lexicon in the field. For a detailed exploration of the data, interact with the visualization. Chart created using Datawrapper, an online visualization tool.

Prompt Architecture vs. The Brain

Preference for the term 'prompt architecture' (57%) over 'the brain' (23%) when participants define the systematic arrangement and connections of prompts. This preference suggests a leaning towards more formalized and structural descriptors in the conceptualization of prompt strategies

Figure 4.1.2: Professional Levels and Terminology Preferences

Distribution of Terminology Preferences by Professional Expertise in Prompt Strategy

Note. Represents survey results from "Exploring the Landscape of Prompting" survey conducted by Jonathan Kyle Hobson. Data breaks down by expertise level the preference for 'prompt architecture' versus 'the brain' in the context of prompt strategy. The distribution across expert professionals, experienced users, and casual users underscores the diverse perspectives within the field. For a detailed exploration of the data, interact with the visualization. Chart created using Datawrapper, an online visualization tool.

Prompt Architecture vs. The Brain

Across all levels of professional expertise, 'prompt architecture' is consistently favored, with 66.7% of expert professionals, 59% of experienced users, and 61% of casual users selecting it over 'the brain.' This consistency across professional categories suggests a broad consensus on the preferred term.

Prompt Architecture: Constructing the Virtual Scaffold

Prompt Architecture, at its essence, involves the meticulous arrangement and interconnection of prompts and their underlying strategies. This concept transcends the simplicity of a single prompt, advocating for a more layered, complex system that mirrors human cognitive processes. Whether the objective is to establish an infinite generation model—designed to consistently yield a uniform output—or to develop an adaptive system, prompt architecture serves as the blueprint for achieving these goals.

The Essence of Prompt Architecture

At the heart of prompt architecture lies the endeavor to simulate or mirror human thought processes within AI models. This encompasses not just the mere delivery of instructions but an orchestrated assembly of prompts that, collectively, can simulate reasoning, decision-making, and even creative thought. Through the application of specific arrangements, orders, and strategies within prompts, one can architect a dialogue that closely resembles human reasoning models—from chain of thought to more elaborate planning and problem-solving frameworks.

The application of prompt architecture requires a confluence of diverse thinking structures, including but not limited to:

• Iterative Design: This approach, inspired by an agile mindset, emphasizes continuous refinement and testing. It advocates for the evolution of prompt design through successive iterations, allowing for adaptive improvements based on AI feedback.
• Design Thinking: Borrowing from the realms of design and user experience, this methodology places a premium on empathy with the AI's processing capabilities. It encourages the creation of prompts that resonate with the AI's interpretative mechanisms, ensuring alignment between human intentions and AI responses.

Constraints and Limitations Awareness: A crucial aspect of prompt architecture is the mindful acknowledgment of AI's limitations, such as context understanding, reasoning biases, and the propensity for hallucinations. By incorporating sayings or mnemonic devices that encapsulate these limitations (e.g., "Context is King" or "Clarity over Creativity"), prompt developers can maintain a focused approach that navigates around these potential pitfalls.

The Role of Prompt Architecture in AI Interaction

Prompt Architecture is not merely about the technical execution of commands. It embodies a strategic process aimed at enriching the interaction between humans and AI. By judiciously employing prompt architecture, developers can:

• Simulate complex reasoning models, thereby enabling AI to undertake sophisticated analytical tasks.
• Enhance the AI's understanding of human nuances, bridging gaps in context and intention.
• Foster a more dynamic, flexible dialogue with AI, moving beyond static interactions to more meaningful exchanges.

Prompt Architecture heralds a paradigm shift in how we conceive of and engage with AI systems. It calls for a departure from the simplistic notion of single-prompt interactions, urging us to consider the broader, more complex landscape of human-AI dialogue. By embracing this architectural approach, we pave the way for more nuanced, intelligent, and ultimately human-like interactions with AI, marking a significant step forward in the ongoing journey of AI development and application.

Prompt Development Techniques: Frameworks and Methodologies

Understanding Prompt Architecture: The Foundation

Before we dissect the various frameworks and methodologies that facilitate prompt planning, it's crucial to acknowledge the essence of prompt architecture. Prompt architecture—or the systematic arrangement and interconnection of prompts and their strategies—is akin to constructing a building. Each prompt is a brick, each strategy a mortar, coming together to form a structure that is both resilient and functional. This concept finds resonance among professionals, with a significant majority, 57% of 164 respondents from the research, expressing a preference for "Prompt Architecture" over other terminologies. This preference underscores a shift towards a more structured, thoughtful approach to crafting AI interactions.

Methodologies and Frameworks: A Kaleidoscope of Strategies

Delving into prompt planning, we encounter a plethora of methodologies and frameworks borrowed from various disciplines, each bringing its unique lens to the task of prompt construction. These range from the clarity and goal orientation of the CLEAR methodologies—be it the Clear Path Forward Framework (Lo, 2023), the Clear Prompting Method, or the Clear Framework—to strategic planning tools like SWOT Analysis and creative processes such as the AIDA Model and the Golden Circle Framework.

• CLEAR Path Forward Framework emphasizes a concise, logical, explicit, adaptive, and reflective approach, optimizing the interaction between humans and AI.?

“Brevity and clarity in prompts guide AI language models to concentrate on the core elements of the task" (Lo, 2023).

• CLEAR Prompting Method (Clarity, Length, Empathy, Actionability, Relevance) optimizes AI interactions by focusing on the essential elements of communication, ensuring prompts are efficient and effective.
• CLEAR Framework (Challenge, Limitation, Effect, Action, Result) this framework provides a strategic approach to prompt engineering by outlining challenges, describing limitations, proposing actions, and predicting outcomes, enhancing the predictive power of AI interactions.
• RTF Framework (Role, Task, Format) categorizes requests into roles, tasks, and formats, providing a structured approach for clearer communication in diverse scenarios.
• RISEN Framework (Role, Instructions, Steps, End goal, Narrowing) offers a detailed structure to dissect complex tasks into manageable actions, enhancing focus and execution in project management or content creation.

“Structured and coherent prompts enable AI models to grasp the context and the interconnections between various concepts." (Lo, 2023)

• RATIO Framework (Role, Audience, Task, Instructions, Output) ensures comprehensive coverage of the communication process, from defining the role to specifying the desired output, promoting a thorough understanding between the AI and the user.
• RODES Framework (Role, Objective, Details, Examples, Sense Check) employs examples to clarify expectations, aiding users in crafting more precise prompts by providing a sense of check for comprehensiveness and relevance.

“Clearly stating the expected output format, content, or scope minimizes the risk of unsuitable or irrelevant responses from the AI model" (Lo, 2023)

• SWOT Analysis (Strengths, Weaknesses, Opportunities, and Threats), traditionally used in strategic planning, offers a structured way to evaluate—applicable in gauging the context and potential challenges in prompt design.
• AIDA Model (Attention, Interest, Desire, Action) guides the creation of engaging and compelling content, principles equally vital in prompt construction to capture and maintain AI's focus.
• PAR Method (Problem, Action, Result) focuses on problem-solving by structuring prompts to delineate the issue clearly, propose an action, and anticipate the desired outcome.
• COSTAR Framework (Context, Objective, Style, Tone, Audience, Response) aligns AI responses with the nuanced needs of different scenarios by detailing the context, desired style, tone, and the target audience.

“Experimenting with a variety of prompt formations, phrasings, and temperature settings to find a balance between creativity and specificity" (Lo, 2023)

• The Golden Circle Framework (What, Why, How) (Sinek, 2011) delves into the purpose behind a prompt, ensuring that each inquiry is rooted in a clear objective and rationale.

Figure 5.1.1: Golden Circle Framework

Adapting Simon Sinek’s Golden Circle to AI Prompting Strategies

Note. This figure adapts Simon Sinek’s renowned Golden Circle framework (Sinek, 2011) to the context of AI communication, illustrating the “what”, “how”, and “why” of AI interaction, with an added dimension of “mission” inspired by David Shapiro's work (Shapiro, 2023a) Crafted by researcher Jonathan Kyle Hobson in Figma, the diagram serves as a strategic guide for integrating AI within one’s personal or organizational mission, ensuring that the deployment of AI technologies aligns with underlying motivations and desired outcomes.

The figure visually encapsulates the essence of mission-driven AI usage, emphasizing the alignment of AI capabilities with one’s overarching goals and beliefs. It serves as a conceptual roadmap, prompting users to consider not just the practicalities of “what” they want to achieve with AI and 'how' to implement it, but also to reflect deeply on “why”'—the motivations and values that drive their interaction with technology. By integrating Shapiro’s concept of mission into the framework (Shapiro, 2023a), the figure highlights the potential of AI to serve as a force multiplier for individual aspirations and collective initiatives, acting as a transformative agent that extends beyond a mere facilitator of tasks to become an enabler of broader visions and a collaborator in community building. This adaptation beckons prompters and users alike to craft AI engagements that are not only effective but resonate with their mission (Shapiro, 2023a), ensuring that AI’s profound capabilities are harnessed to bolster a purposeful and values-driven trajectory.

In the realm of prompt development, the meticulous construction and arrangement of prompts stand as a testament to the intricacy of human-AI dialogue. Beyond the strategic frameworks and methodologies that guide the crafting of prompts, there lies an equally critical facet of the process: the evaluation of AI outputs. The RACCCA framework (Maynard, 2023a) emerges as a beacon in this pursuit, offering a streamlined approach to ensure the outputs generated by AI not only meet but exceed our expectations for quality and relevance. “Through careful evaluation of AI-generated responses, users can improve their future prompts and increase the effectiveness of their AI engagements" (Lo, 2023).

The RACCCA Framework: A Pillar of Quality Assurance

RACCCA—an acronym standing for Relevancy, Accuracy, Completeness, Clarity, Coherence, and Appropriateness (Maynard, 2023a)—serves as a multifaceted lens through which to assess the fruits of our prompt engineering labor. Developed by Andrew Maynard (2023a), this framework encapsulates the essential qualities that define the efficacy of AI-generated content:

• Relevancy: Does the output align with the initial prompt's objectives, effectively addressing the issue or question at hand?
• Accuracy: Is the information provided by the AI both reliable and factually correct, verifiable through additional research or logical deduction?
• Completeness: Does the AI's response comprehensively cover the topic, leaving no stone unturned in its pursuit of thoroughness?
• Clarity: Is the output articulated in a manner that is easily digestible and understandable to its intended audience?
• Coherence: Does the content exhibit a logical flow, seamlessly transitioning from one point to the next with grace and precision?
• Appropriateness: Is the tone and context of the AI's response suited to the scenario, respecting the nuances of the situation and the medium of communication?

The RACCCA framework not only stands as a testament to the principles of quality control but also underscores the importance of a tailored, user-centric approach to AI interaction. By adhering to these guidelines, we can sieve through AI outputs, filtering for content that truly resonates with our objectives and expectations.

“The bottom line here is that by becoming familiar with the RACCCA framework, you will be able to develop more effective prompts that lead to more effective and useful outputs” (Maynard, 2023a)

Figure 5.1.2: RACCCA Framework

A Prompter’s Method for Evaluating the Output of GenAI?

Note. The RACCCA Framework (Maynard, 2023a) depicted here represents a methodology developed for evaluating the output of generative AI (GenAI). This visual representation, created by Jonathan Kyle Hobson using Figma, is based on the Arizona State University’s conceptual framework proposed by Andrew Maynard (2023a). It illustrates the essential criteria of Relevance, Accuracy, Completeness, Clarity, Coherence, and Appropriateness—collectively forming a gold standard for AI output evaluation in the field of prompt engineering. This figure serves as a guide for practitioners aiming to assess the quality and utility of GenAI-generated content.

Bridging Frameworks with Execution

The introduction of the RACCCA framework preceding our exploration of core themes in prompt development is not merely a segue but a foundational pillar that reinforces the essence of our methodologies. It highlights a universal truth in our interaction with generative AI: while the art of prompt construction is paramount, the ability to critically evaluate and refine the outputs is what truly closes the loop of effective AI communication.

This evaluation is not a one-off checkpoint but a continuous cycle of feedback and improvement, a dance between human intuition and AI capabilities. As we venture into discussions on Goal Orientation, Contextualization, Adaptability, and Values the RACCCA framework serves as a reminder that our endeavors in prompt engineering are as much about the destination as they are about the journey. Each output from AI is a reflection of our ability to communicate complex ideas and objectives, demanding not just creation but also careful consideration and refinement.

Towards a More Inclusive and Dynamic Approach

In weaving the RACCCA framework into the fabric of our methodologies, we pave the way for a more dynamic, inclusive approach to prompt development. It underscores the need for prompts that are not just well-crafted but are also designed with an acute awareness of their potential impact and reception. As we proceed, let this framework remind us that in the ever-evolving landscape of AI, our strategies, too, must adapt, always striving for outputs that are relevant, accurate, complete, clear, coherent, and appropriate to our diverse and changing needs.

Figure 5.1.2: RACCCA Framework Use by Expertise Level

Usage Patterns of RACCCA Among AI Prompting Professionals

Note. Represents survey results from "Exploring the Landscape of Prompting" survey conducted by Jonathan Kyle Hobson. Data presents the usage of RACCCA Framework categorized by professional expertise. The chart differentiates between those who have not and have used RACCCA, reflecting variations in engagement across different expertise levels. For a detailed exploration of the data, interact with the visualization. Chart created using Datawrapper, an online visualization tool.

The Core Themes: Goal Orientation, Contextualization, and Adaptability

While the diversity of these frameworks is vast, they share core themes essential to effective prompt development:

Contextualization: Perhaps the most critical aspect is ensuring that prompts are rich in context. From the environmental considerations of the PEAS Framework to the empathetic approach of CLEAR Prompting, the inclusion of relevant, detailed context guides AI towards producing accurate, relevant responses.?

This includes:

• What information is essential?
• Why is this task important to you?
• How will this enhance your outcome?
• Who is the intended audience?

Goal Orientation: Central to all methodologies is the emphasis on clear, actionable objectives. Whether defining a specific problem in the PAR Method or setting a task in the STAR Technique, the focus is invariably on achieving a desired outcome.

This includes:

• What is the primary task to be accomplished?
• Why are these tasks critical to your objectives?
• How will you measure success?
• Who is responsible for task execution?

Adaptability and Flexibility: Each framework advocates for a flexible approach to prompt planning, acknowledging the dynamic nature of AI and the need for prompts to evolve based on feedback and changing circumstances.

This includes:

• What discipline or subject area are you addressing?
• Why is a balance between structure and dynamic adaptation necessary?
• How can you use divergent thinking?
• Who benefits from adopting these thinking structures?

Value-orientation: These three themes all boil down to value-orientation. How can we create alignment with inputs we write and the outputs we desire or need.

Moving Forward: Towards a Unified Approach

As we navigate the complexities of prompt development, it becomes evident that no single framework offers a panacea. Instead, the richness lies in the amalgamation of these methodologies, tailored to the specific needs of each prompt development endeavor. The challenge, therefore, lies in selecting, adapting, and combining elements from these diverse frameworks to construct prompts that are not only effective but also reflective of the nuanced interplay between human intent and AI capabilities.

In essence, the methodologies and frameworks outlined here serve as a compass, guiding prompt engineers through the multifaceted landscape of AI interaction. They remind us that the art of prompt development is not just about instructing AI but engaging in a dialogue—a dialogue that is structured, purposeful, and, above all, human-centric.

The Essence of Prompting Captured in Sayings

As we venture beyond the structured methodologies and dive into the artistry of prompt development, a collection of pithy sayings emerges, encapsulating the wisdom and strategies of engaging with AI in a manner both profound and accessible. These adages, akin to navigational stars in the vast expanse of AI interaction, offer immediate, memorable guidance that enriches our understanding and application of prompt engineering principles.

Easy for You, Easy for It; Hard for You, Hard for It (Maeda & Schillace, 2023): This Schillace Laws of Semantic AI maxim underscores the fundamental notion that complexity for the prompter often translates to complexity for the AI. It serves as a litmus test for the feasibility of tasks, suggesting that if a task seems overly complex for a human to articulate or accomplish in one go, it may require decomposition into more manageable, focused prompts for the AI to handle effectively.

“So like often one of the best ways that we have to think about how you build an output using a large language model is to think about how you would do it as a person” (Maeda & Schillace, 2023).

Put Smart, Get Smart; Ask Smart to Get Smart (Maeda & Schillace, 2023) : The essence of this Schillace Laws of Semantic AI lies in the direct correlation between the quality of input and the quality of output. It reminds us that the precision, clarity, and thoughtfulness we invest in crafting our prompts directly influence the relevance, accuracy, and utility of the AI's responses.

The Divergent Intern: This analogy humanizes AI by comparing it to an intern with a unique, divergent way of processing information. It highlights the necessity of providing detailed context, explanations, and clarifications to guide the AI through tasks as we would with a human intern unfamiliar with our implicit expectations and unspoken norms.

The Collaborative Partner: By framing AI as a collaborative partner, we're reminded that the process of prompt engineering is a dialogue, not a monologue. This perspective fosters a more dynamic interaction, where feedback loops and iterative refinements are integral to achieving desired outcomes.

The 10,000 Experts: Viewing AI as a conglomerate of countless experts encapsulates its vast, diverse knowledge base. It encourages us to specify the domain or perspective we wish to engage with, ensuring that the AI's responses are aligned with our specific needs and context.

Figure 5.2.1: 10,000 Experts

Multidisciplinary Perspectives on AI and Subconscious Mind References

Note. This illustrative figure, created by researcher Jonathan Kyle Hobson, encapsulates the ambiguity of AI prompts that lack specificity. The central “bad prompt” highlighted in the figure is critiqued for its outdated references and vague language, underscoring the importance of clarity and precision in AI prompting. Arrows point to terms within the prompt, illuminating the issues of unclear training data scope and undefined terms, while the surrounding cartoon figures from various disciplines—art, psychology, philosophy, and neuroscience—provide divergent interpretations of the term "subconscious mind," reflecting the broad spectrum of meanings the phrase can convey.

The figure demonstrates the complexity of language and the critical need for context in AI prompting. By illustrating how a term like "subconscious mind" can be variously construed by professionals from different fields, the visual highlights how AI might struggle with processing such a prompt due to its inherent training on a multitude of contexts. The misconstrued "whole AI thing" serves as a caution against using jargon that could reflect outdated understanding or broad generalizations. This visualization serves as an educational tool to underscore the challenges of semantic interpretation in AI and the value of crafting prompts that are informed by up-to-date, specific, and unambiguous language.

The Three Cs: Critical, Creative, and Compassionate Thinking: The Three Cs—Critical thinking, Creative thinking, and Compassionate thinking—represent a framework for the essential skills and mindsets required for effective prompting in the evolving AI landscape. This interpretation of the Three Cs deviates from the traditional inclusion of Collaboration, instead emphasizing the importance of Compassionate thinking as a crucial component alongside Critical and Creative thinking.

Moving Forward with Wisdom and Strategy

As we integrate these sayings and insights into our approach, we pave the way for a more nuanced, effective engagement with AI. They serve not just as reminders of best practices but as beacons that guide us through the complexities of prompt engineering. By embodying the principles encapsulated in these proverbs, we embrace a holistic, adaptable, and ultimately more human-centric approach to AI interaction. This enriched perspective ensures that our journey through the landscape of AI is not only informed by structured methodologies but also by the distilled wisdom of collective experience.