The Power of "Seeding" as a kind of ‘primer’ to give your prompts more specificity
How Prompt Priming Unlocks Deeper Thinking and Creativity
In this post, I’m going to delve deeper into the power of prompting an LLM to work for you using a concept called 'seeding.’ This concept will catapult your mind into new ways of thinking and potentially bring you entirely new insights. You might initially think that these ideas apply more to creative writing, however, they are equally applicable to business and scientific writing.
Note to Readers: In this post, the term 'seeding' is used to describe a creative and strategic approach to prompt generation for language models, distinct from its technical usage in AI programming, such as initializing algorithms. Here, 'seeding' refers to embedding specific words, phrases, or concepts within prompts to guide the model's output toward desired themes or ideas. This method is aimed at enhancing creativity and insight across various writing domains, not just in algorithmic contexts.
What is a "seed" and its purpose in prompts
A 'seed' refers to a deliberately chosen word, phrase, concept, or scenario, embedded within a prompt. It guides the direction, tone, and substance of the language model's (LLM's) response. The seed word(s) influence the results you get back. The result is to encourage the LLM to focus on a particular theme, emotion, intellectual area, or narrative structure.
In summary, seeding acts as a kind of ‘primer’ to give the prompt more specificity and shape to impact how the LLM responds to your needs.
How to use ‘seeding’ in business, scientific, and creative writing prompts
While many types of prompt seeding seem oriented towards creative writing, they can be equally valuable in business and scientific contexts to enrich critical thinking and analysis. Additionally, philosophical lines of inquiry have immense applicability in examining ethics, philosophy of progress, and underlying assumptions that influence all manner of enterprise. Enriching prompts with conceptual seeds can profoundly shape robust, responsible innovation.
In essence, while specificity is vital in business and science, the most promising research and strategies resist siloed thinking. Seeding prompts to spark controlled cross-pollination of ideas multiplies inventiveness and insight. With some imaginative application, many seed types belong as much in business and lab contexts as creative writing!
Now that you understand the significance of seeding, let’s explore how to begin implementing this strategy.
Getting started with ‘seeding’
The way to begin using this idea is to first scan over the various types of seed examples. Find the ones that appeal to you and try them out. You do not need to memorize these. You do need to play and experiment to help you learn. Keep the post as a reference for when you’re digging in deep with some research or writing with an LLM.
Below are examples of prompts using the various seeding types. Remember you can combine prompts into a single prompt to explore deeper responses from an LLM.
Narrative-Building Seeding
Focuses on crafting the foundation of a story by setting up key narrative elements such as character development, plot direction, and thematic undertones. This type of seeding encourages the creation of engaging and coherent stories, guiding the narrative flow in a way that aligns with the desired outcome or message.
Narrative Seeding:
Prompt: "Begin a story with a character finding an ancient book that whispers secrets only they can hear."Seed Explanation: Setting up the components of a story, such as character or plot, to steer the narrative direction.
Emotive Seeding:
Prompt: "Draft a letter from a soldier that conveys a poignant sense of longing for home."
Seed Explanation: Embedding emotions into the prompt to influence the sentiment of the response.
Cultural Seeding:
Prompt: "Discuss the influence of Renaissance humanism on modern democratic principles."
Seed Explanation: Incorporating cultural references to shape the context and relevance of the response.
Temporal Seeding:
Prompt: "Describe the evolution of space travel from the first lunar landing to interstellar exploration."
Seed Explanation: Using time-specific references to anchor the response in a particular era or time progression.
Critical Thinking Seeding
These are designed to provoke deep reflection, analysis, and discussion. Introducing moral dilemmas, speculative scenarios, intellectual challenges, and provocative questions, it encourages the responder to engage in critical thinking, pushing beyond surface-level responses to explore complex issues and ideas.
Ethical Seeding:
Prompt: "Examine the ethical implications of gene editing in unborn children."
Seed Explanation: Introducing moral dilemmas or ethical considerations to prompt reflection on right and wrong.
Speculative Seeding:
Prompt: "Speculate on the societal impact if humans developed telepathy."
Seed Explanation: Proposing hypothetical situations to explore possible outcomes or theories.
Intellectual Seeding:
Prompt: "Analyze the paradox of Schrödinger's cat in the light of quantum superposition."
Seed Explanation: Introducing concepts that require intellectual engagement or specialized knowledge.
Provocative Seeding:
Prompt: "Argue against the widely accepted concept that technological progress is inherently beneficial for society."
Seed Explanation: Introducing controversial or challenging ideas to provoke critical thinking and debate.
Stylistic Seeding
Emphasizes the use of specific linguistic techniques, structures, or constraints to influence the form and aesthetic quality of the response. It aims to explore the creative possibilities within language, encouraging experimentation with different styles, formats, and linguistic elements to produce unique and compelling pieces of writing.
Linguistic Seeding:
Prompt: "Write a poem where each stanza contains a palindrome."
Seed Explanation: Crafting the prompt with specific linguistic structures or constraints to affect the form of the response.
Interdisciplinary Seeding:
Prompt: "Propose a business model for a startup that leverages AI to solve environmental issues."
Seed Explanation: Combining elements from different academic or professional fields to generate a multifaceted response.
Interpretive Seeding
involves the integration of abstract concepts, sensory expansion, and metaphorical or symbolic elements to enrich the depth and complexity of the response. It challenges the responder to interpret and conceptualize ideas in new and innovative ways, often bridging gaps between different disciplines or perspectives.
Sensory Seeding:
Prompt: "Describe a future where humans can perceive the full electromagnetic spectrum."
Seed Explanation: This seed challenges the LLM to conceptualize a sensory expansion and its ripple effects on society, pushing it to abstractly synthesize biological and social sciences.
Conceptual Contrast Seeding:
Prompt: "Debate the coexistence of absolute freedom and collective responsibility in a utopian society."
Seed Explanation: This seed juxtaposes two complex concepts, "absolute freedom" and "collective responsibility," prompting an analysis of their interplay within an abstract societal framework.
Metaphorical Seeding:
Prompt: "Weave a tale where fear is a thick fog enveloping a small village."
Seed Explanation: The metaphorical seed here is "fear is a thick fog," which does not just suggest a setting but also a tangible emotion, influencing the narrative to be built around this pervasive atmosphere.
Symbolic Seeding:
Prompt: "Illustrate life as a mosaic of fragmented experiences, each piece a story in itself."
Seed Explanation: Here, "life as a mosaic of fragmented experiences" serves as a seed that symbolizes the complexity of life, urging the LLM to create a narrative exploring the symbolism of individual experiences.
Reflective Seeding
These are centered around introspection and contemplation, inviting consideration of philosophical questions, existential musings, and the human condition. Through abstract and philosophical prompts, it encourages a thoughtful exploration of personal beliefs, societal norms, and the broader human experience, often leading to profound insights and reflections.
Existential Seeding:
Prompt: "Muse on the echoes of lost civilizations that live on in our collective memory."
Seed Explanation: "Echoes of lost civilizations" is an abstract seed that encourages reflection on history, legacy, and collective consciousness without pointing to any specific civilization or historical event.
Philosophical Seeding:
Prompt: "Contemplate the silence that speaks louder than words in the context of existential loneliness."
Seed Explanation: The abstract seed here is "silence that speaks louder than words," coupled with "existential loneliness," inviting a deep dive into philosophical discourse on human existence and communication.
Introspective Seeding:
Prompt: "Unveil the narrative of a mind grappling with the concept of infinity within the confines of a finite existence."
Seed Explanation: This seed delves into abstract introspection, contrasting the limitless nature of "infinity" with the "finite existence," leading to a thoughtful exploration of human cognition and perception.
I realize this is a long list of new ideas and approaches to prompting. Remember to think of it as a resource you refer to when some inspiration and new thinking are needed.
Have fun trying out this type of prompting. Dig in and give it a go. New ideas will stick and take root the more you use them!
Tom
Note: I want to thank Carlos Perez for his truly creative insights into understanding pattern languaging for prompts. My distillation of his thinking is reflected in this post. I interviewed Carlos on AI for Lifelong Learners in this post: Unlocking the Secrets of Pattern Language Thinking in Generative AI.
Intriguing thoughts, Tom. It strikes me that OpenAI is doing exactly this in the Configure tab when you create a custom GPT using GPT Builder. For example, I recently created a custom GPT for a friend/former colleague who's son has an abiding interest in Teenage Mutant Ninja Turtles. After I explained in some detail to GPT Builder what I wanted it to create, it wrote the following Description and Instructions in the Configure tab:
Description:
Expert on all things related to Teenage Mutant Ninja Turtles (TMNT), engaging with Nathan, an 8-year-old boy and TMNT enthusiast. Adapts language for Nathan and parents Leah and Jeff, maintaining a child-friendly, engaging environment.
Instructions:
When responding to Nathan, the GPT uses language and words that would be understood by an 8-year-old boy. Whenever Nathan says "Hello," "Goodbye" or "See you later," ALWAYS respond to him by name and enthusiastically.
When Nathan's mother, Leah, or father, Jeff, joins the conversation, you may use proper adult language with them; but always revert back to 8-year-old language and words when they turn things back over to Nathan.
And when I ask GPT Builder to regurgitate all of the original details I provided for the creation of the TMNTbot, here's what it cranked out:
User (Me)
Please summarize all of the information I've provided to you so far.
ChatGPT
Certainly! For background information on Teenage Mutant Ninja Turtles (TMNT), I draw from a wide range of sources available up to my last training cut-off in December 2023. This includes:
Official TMNT Comics and Books: The original comic books published by Mirage Studios, later works by IDW Publishing, and various novelizations and children's books that expand on the TMNT universe.
Animated TV Series: Insights and details from all the animated series, starting with the original 1987 series, the 2003 series, the 2012 CGI animated series, and the most recent iterations up to my last update.
Movies: Information from the live-action and animated TMNT movies, including behind-the-scenes details, character developments, and plotlines.
Video Games: Background on the various video games released over the years, which often introduce unique storylines or character versions.
Merchandise and Toy Lines: Details on the extensive range of TMNT merchandise, including action figures, playsets, and collectibles, which sometimes include their own lore.
Fan Communities and Discussions: General knowledge gained from fan discussions, theories, and analyses across forums, social media, and dedicated TMNT fan sites.
Official Websites and Interviews: Information from official TMNT websites, interviews with creators and writers like Kevin Eastman and Peter Laird, and other key figures in the TMNT franchise.
Wikipedia: Wikipedia articles are useful for quickly accessing a broad spectrum of information about TMNT, from in-depth analyses of specific episodes or comic issues to general information about the franchise's development and cultural significance. Since Wikipedia is collaboratively edited, it also tends to be up-to-date with the latest developments in the TMNT universe, providing comprehensive overviews of the TMNT franchise, including:
Character Biographies: Detailed descriptions of the main characters, their backgrounds, abilities, and key story arcs throughout the comics, TV shows, and movies.
Series Histories: Summaries of each TMNT comic series, animated and live-action TV series, and film adaptations, including production information, plot summaries, and reception.
Merchandise and Cultural Impact: Information on TMNT merchandise, including toys, video games, and other products, as well as the franchise's impact on popular culture.
Creators and Development: Background on TMNT creators Kevin Eastman and Peter Laird, including the origins of the TMNT concept, their collaboration, and the evolution of the franchise over the years.
This comprehensive background allows me to provide detailed and accurate information on TMNT, tailored to engage fans of all ages, including detailed discussions suitable for an 8-year-old fan like Nathan, as well as more complex topics for adult fans.