The Importance of Roles.

TL;DR:

In the context of Large Language Models (LLMs), recognizing the importance of roles involves utilizing directive words, specificity words, contextual words, clarifying words, and comparative words in more complex prompts. Directive words like “analyze” or “compare” guide the LLM’s response approach. Specificity words such as “in the 20th century” or “for beginners” narrow down the focus. Contextual words like “within the field of biology” provide necessary context. Clarifying words like “specifically” or “precisely” highlight the need for precise information. Comparative words like “similarly” or “in contrast” prompt the LLM to draw comparisons. By incorporating these types of words strategically, users can shape the LLM’s response style, narrow down the scope, provide clarity, and guide the LLM’s focus, resulting in more targeted and contextually appropriate interactions.

We touched on the Anatomy of Roles in the Lesson: https://tokenomics.ie/topics/anatomy-of-roles/

Whilst the basic syntax is Role:{Storyteller}Prompt: {the detailed request} we can go beyond this using ‘word types / categories’ and ‘power words.’

While AI doesn’t experience emotions, certain words can indeed guide the AI’s responses more effectively. Here are some categories of such words:

Word Types / Categories

Directive Words: These are words that give clear instructions to the AI. They include words like “describe”, “explain”, “tell”, “show”, etc. They can be used to guide the AI’s task in the response.

Specificity Words: These are words that provide specific details or constraints. They include words like “five”, “yesterday”, “in the future”, “from the perspective of”, etc. They can be used to guide or constrain the AI’s context or perspective in the response.

Contextual Words: These are words that set the context or scenario for the AI. They include words like “imagine”, “suppose”, “assume”, etc. They can be used to guide the AI’s approach to the response.

Clarifying Words: These are words that ask the AI to provide more detail or clarification. They include words like “why”, “how”, “what do you mean”, etc. They can be used to guide the AI to provide more detailed or explanatory responses.

Comparative Words: These are words that ask the AI to make a comparison or contrast. They include words like “compare”, “contrast”, “similar”, “different”, etc. They can be used to guide the AI to evaluate or analyze in the response.

These categories of words can be highly effective in guiding the AI’s responses to be more aligned with your specific goals or expectations.

Power / Special Words

In the context of prompts for an AI there are several words and phrases that can be considered “special” because they imply a certain type of interaction or response. Here are a few examples:

“Imagine“: This word can guide the AI to generate a creative or hypothetical response. For example, “Imagine you’re a tour guide on Mars. What would you show your tourists?”

“Suppose“: Like “imagine”, this word can guide the AI to consider a hypothetical scenario. For example, “Suppose you had to explain the internet to a time traveler from the 1800s. How would you do it?”

“Why don’t we”: This phrase implies a collaborative action or thought process. For example, “Why don’t we brainstorm some ideas for a new novel?”

“What if“: This phrase can guide the AI to consider a hypothetical scenario and its implications. For example, “What if gravity suddenly stopped working?”

“How about“: This phrase can guide the AI to consider a suggestion or proposal. For example, “How about we design a menu for a vegetarian restaurant?”

“Can you“: This phrase can guide the AI to perform a specific task or answer a specific question. For example, “Can you summarize the main points of this article?”

These words and phrases can be highly effective in guiding the AI’s responses to be more aligned with your specific goals or expectations. However, it’s important to note that while these words and phrases can guide the AI’s response, the AI doesn’t actually engage in thought processes or have experiences in the way humans do. The AI simply generates responses based on patterns it learned during training.

20 Examples of Further Interactions

• “Imagine you’re a historian. What would you say is the most significant event of the 20th century?”
• “Suppose you were a chef. What would be your signature dish?”
• “Why don’t we come up with a list of the top 10 science fiction books of all time?”
• “What if we had to design a city from scratch? What features would it have?”
• “How about we brainstorm some ideas for a new mobile app?”
• “Can you explain the theory of relativity in simple terms?”
• “Let’s plan a virtual tour of the Louvre Museum.”
• “Imagine you could interview any person from history. Who would it be and why?”
• “Suppose you were an astronaut on the International Space Station. What would your daily routine look like?”
• “Why don’t we create a timeline of major events in the evolution of technology?”
• “What if we had to create a new language? What would be some of its words and rules?”
• “How about we design a new board game? What would be its rules and objectives?”
• “Can you provide a summary of the plot of ‘War and Peace’?”
• “Let’s create a list of the most influential works of literature in the world.”
• “Imagine you could travel anywhere in the universe. Where would you go and what would you hope to see?”
• “Suppose you were a character in your favorite book. What would you do differently?”
• “Why don’t we come up with a new concept for a reality TV show?”
• “What if we had to invent a new sport? What would be the rules and how would it be played?”
• “How about we create a guide for learning a new language effectively?”
• “Can you describe what life might be like on a habitable exoplanet?”

Each of these prompts is designed to guide the AI’s response in a specific way, whether that’s by specifying the format of the response (e.g., a list, a plan, a design, a summary), the perspective (e.g., a historian, a chef, an astronaut), or the content (e.g., a significant event, a signature dish, a daily routine).

When used in a prompt for an AI words that imply collaboration, such as “what if we” or “why don’t we”, are used to set a cooperative tone and guide the AI to generate a response that continues the proposed action or thought process. For example, if you start a prompt with “why don’t we plan a trip to Paris”, it guides the AI to generate a response that contributes to planning a trip to Paris.

Statistical Patterns

Statistical patterns in the data are the foundational basis of how machine learning models like ChatGPT learn and make predictions. These patterns are essentially relationships between different elements in the data that the model has learned to recognize. Here’s a high-level overview of how these patterns might manifest:

Word Associations: The model learns which words tend to appear together. For example, in English text, the word “rain” is often associated with words like “umbrella”, “wet”, “clouds”, etc. This is a simple form of pattern that the model can learn.

Syntax and Grammar: The model learns the rules of syntax and grammar implicitly by recognizing the patterns in which words are arranged in sentences. For example, in English, adjectives usually come before the nouns they modify, and subjects usually come before verbs.

Semantic Relationships: The model learns relationships between words and phrases that have similar meanings or that are related in some way. For example, it might learn that “cat” and “kitten” are related, or that “running” is related to “jogging”.

Contextual Understanding: The model learns to understand the context in which words or phrases are used. For example, it might learn that the word “bank” has different meanings depending on whether it’s used in a financial context (“I went to the bank to deposit money”) or a geographical context (“I sat by the bank of the river”).

Longer Text Structures: The model learns patterns in how longer pieces of text are structured. For example, it might learn that stories often have a beginning, middle, and end, or that arguments often start with a claim, followed by evidence, and then a conclusion.

Long prompts that use Power Words & Role

“Imagine you’re a historian in the year 2100, looking back at the development and impact of large language models like GPT-3. Describe the key milestones in their evolution, the challenges they faced, and the ways they transformed how information is created and consumed.”

“Suppose you were a technology expert tasked with writing a comprehensive review for a technology magazine about the evolution of information in the era of large language models. Discuss the transition from traditional search engines to AI-driven content generation, the implications for data privacy and security, and the potential future developments in this field.”

“Let’s create a detailed timeline of the evolution of information in the context of large language models. Start from the inception of machine learning, move through the development of natural language processing and the launch of models like GPT-3, and project into the future based on current trends and advancements.”

“Why don’t we brainstorm some ideas for a documentary about the evolution of information in the age of large language models? Consider key events, influential figures, major breakthroughs, societal impacts, and potential future scenarios that should be included.”

“What if you were an educator designing a course on ‘Information Evolution in the Large Language Model Era’? Outline the key modules of the course, the learning objectives for each module, and the types of assignments and assessments you would include to engage students and evaluate their understanding.”

Each of these prompts is designed to guide the AI’s response in a specific way, whether that’s by specifying the format of the response (e.g., a description, a review, a timeline, a brainstorm, a course outline), the perspective (e.g., a historian, a magazine writer, an educator), or the content (e.g., key milestones, challenges, transformations, implications, future developments).

Evolution in a large language model world.

“Imagine** you’re a journalist in the year 2050, writing an article about the impact of large language models on the evolution of online communication. Discuss the changes in how people interact, the emergence of new forms of digital literacy, and the societal implications of these shifts.”**

“Suppose** you were an AI ethicist examining the role of large language models in the evolution of information dissemination. Analyze the ethical considerations, potential risks, and proposed guidelines for responsible use of such technology.”**

“Let’s** develop a comprehensive report on the evolution of information in the age of large language models. Include a historical overview, a discussion of the current state of the field, and projections for future developments.”**

“Why don’t we** create a visual timeline that traces the evolution of information from the invention of writing to the development of large language models? Highlight key milestones, breakthrough technologies, and influential figures along the way.”**

“What if** you were a science fiction writer crafting a story set in a world where large language models have revolutionized the way information is created and shared? Outline the plot, characters, and key themes of your story.”**

“Can you** provide a detailed analysis of the role of large language models in the evolution of digital journalism? Discuss the advantages, challenges, and potential future trends.”**

“How about** we design a museum exhibit about the evolution of information in the era of large language models? Describe the sections of the exhibit, the artifacts or displays in each section, and the educational goals of the exhibit.”**

“Consider** the impact of large language models on the evolution of education. How has AI-driven content generation changed the way students learn and teachers teach? What are the potential implications for educational equity and accessibility?”**

“Explore** the role of large language models in the evolution of creative writing. How have these AI models influenced the process and product of writing in genres such as poetry, fiction, and scriptwriting?”**

“Explain** the influence of large language models on the evolution of customer service. How has AI-driven communication transformed the customer experience, and what might the future of AI in customer service look like?”**

Each of these prompts is designed to guide the AI’s response in a specific way, whether that’s by specifying the format of the response (e.g., an article, an analysis, a report, a timeline, a story outline), the perspective (e.g., a journalist, an ethicist, a science fiction writer), or the content (e.g., online communication, ethical considerations, digital journalism, education, creative writing).

Note: when the text is enclosed within double asterisks (**), it is typically used to emphasize or highlight a specific word or phrase in a prompt. It acts as a form of formatting to draw attention to the particular word or phrase. It does not have a specific meaning beyond this formatting purpose. The use of double asterisks (text) is similar to formatting techniques like bold or caps.

Collaborative Action or Thought Process

It is important to note that while words can guide the AI’s response, the AI doesn’t actually engage in collaborative actions or thought processes in the way humans do. The AI doesn’t have thoughts or intentions, and it doesn’t experience collaboration. It simply generates responses based on patterns it learned during training.

In more complex terms, when the AI receives a prompt that implies collaboration, it doesn’t understand the concept of collaboration or have a sense of shared purpose. Instead, it uses its training data to predict the most likely response given the input. This prediction is based on statistical patterns in the data, not on any understanding of the task or any intention to cooperate. The AI doesn’t have a concept of “we” or a sense of shared action. It simply generates text based on patterns it has learned.

Please Note: we provide specific examples of relevant prompts in the final lesson of this module.

Tip: Use the Tech

Creating prompts that include a combination of Roles, Word Categories and Power Words is an iterative process. In this context is helpful to engage the AI in a pre-prompt exercise. One way to do this is to work back from your primary objective by breaking the anticipated outcome into blocks and then using question & answer with the AI to iterate.

Conclusion

Roles are a most important aspect. Whilst you can use them in isolation when you combine them with certain categories / types of words and also with ‘power words’ the results can be impressive. Statistical patterns in the data are the foundational basis of how machine learning models like ChatGPT learn and make predictions. Consequently whilst the inclusion of Directive Words, Specificity Words, Contextual Words, Clarifying Words, Comparative Words might be a big ask, the combination of Roles, Word Categories and Power Words are crucial in Prompt evolution.