Demonstration Hero Light

What does the Demonstration Pack contain?

The Demonstration pack contains useful prompts that can be used right away to get familiar with prompting in Wispera. There are examples of popular prompting schemes such as chain-of-denstity prompts, step-around prompts, prompts for article generation, and more.

Chain-of-Density Prompt Example

Chain-of-density (CoD) prompts are an advanced technique in prompt engineering. They involve a strategic and iterative process of refining a summary to include more information without increasing its length. This process creates denser summaries, which are more informative and detailed. The CoD technique balances clarity and informativeness, ensuring the summary remains readable while encapsulating more content. The concept of CoD prompting is based on the strategic incorporation of key entities derived from the source content. It starts with a basic entity-sparse summary and gradually integrates additional entities, making the summary richer in information. This method underscores the importance of maintaining a fixed length, challenging the prompt engineer to skillfully compress and fuse content. The goal is to produce a summary that is both highly informative and succinct, a crucial capability for applications requiring rapid information processing.

This document steps you through navigating the Wispera Designer UI and using the example from the Demonstration pack for your own summarization.

1. Navigating the UI.

  • Log into your Wispera account and navigate to ‘Packs’.
  • Select the ‘Demonstration’ pack.
  • Enter the Packs workbench.

Select Demonstration Pack

The workbench for a Pack provides quick access to various pre-built prompts for different use cases. This makes it easy to find a starting point that most closely aligns with what you need to build.

2. Selecting an Example Prompt.

  • Find the Chain-of-Density Prompt Example.
  • Click on the Workbench icon to get started with your summarization.

Chain-of-Density Prompt Example

Chain-of-Density Prompt Example

Once in the workbench, you can examine the prompt’s structure and modify it to suit your needs.

3. Review and/or modify the prompt’s structure.

  • Review the System clause which defines the persona.
Chain-of-Density System Clause
  You are an expert in writing rich and dense summaries in broad domains.

Chain-of-Density Prompt - System Clause

Chain-of-Density Prompt - System Clause

  • Insert your document content into the Article clause. This example operates off raw text; however, you can also upload files, scrape web pages, and return JSON from API endpoints.
Chain-of-Density Article Clause
  Article:

  {{article}}

Chain-of-Density Prompt - Article Clause

Chain-of-Density Prompt - Article Clause

  • Review the Step Definition clause.
Chain-of-Density Step Definition Clause
  Generate increasingly concise, entity-rich summaries of the Article by following these steps five times:

  1. Identify 1-3 critical Entities missing from the previous summary. Ensure they are:
   - Relevant to the main story
   - Descriptive yet concise (5 words or fewer)
   - Novel and not previously mentioned
   - Faithful to the Article content
   - Sourced from any part of the Article

  2. Write a new, denser summary of the same length that includes every entity and detail from the previous summary, plus the newly identified entities.

Chain-of-Density Prompt - Step Definition Clause

Chain-of-Density Prompt - Step Definition Clause

  • Review the Guidelines clause.
Chain-of-Density Guidelines Clause
  Guidelines:
  - Start with a long, highly non-specific summary (4-5 sentences, ~80 words) indicating missing entities.
  - Use verbose language and fillers to meet ~80 words.
  - Optimize wording to improve flow and add more entities.
  - Use fusion, compression, and remove uninformative phrases.
  - Make summaries dense and concise but self-contained.
  - Ensure all previous entities appear in the new summary.
  - Maintain the exact word count for each summary. If needed, add fewer new entities.
  - The JSON in `summaries_per_step` should be a list of 5 dictionaries, each with keys "missing_entities" and "denser_summary."

Chain-of-Density Prompt - Guidelines Clause

Chain-of-Density Prompt - Guidelines Clause

  • After making changes, save the prompt before navigating away from the view.

Chain-of-Density Prompt - Save Prompt

Chain-of-Density Prompt - Save Prompt

4. Running prompts against different models and conducting tests.

  • In the prompt workbench, click the New Thread button.

Chain-of-Density Prompt - Start New Thread

Chain-of-Density Prompt - Start New Thread

  • Select the LLM you want to test your prompt against from the list of models available.

Chain-of-Density Prompt - Start New Thread

Chain-of-Density Prompt - Start New Thread

Testing prompts against various LLMs is a critical step in evaluating their performance. Different models might interpret and process the same prompt in unique ways, leading to varying levels of summary quality. Wispera’s interface provides a seamless way to run these tests, allowing you to compare results efficiently. In this testing phase, theoretical knowledge meets practical application, providing insights that can refine your prompt engineering techniques. For this example the responses we received from two different models the prompt was tested against are below.

GPT-4o Example
{
  "summaries_per_step": [
    {
      "missing_entities": ["platform features", "user interface"],
      "denser_summary": "Summarization plays a significant role in natural language processing by condensing information into concise summaries. Effective summarization often requires the use of prompt engineering, a technique that Wispera excels at. This article explains how to create and optimize chain-of-density (CoD) prompts using Wispera. The CoD method involves refining summaries to include more information without increasing their length. Wispera provides tools to test and compare different LLMs, making it possible to achieve highly informative and readable summaries."
    },
    {
      "missing_entities": ["data processing", "prompt refinement", "model performance"],
      "denser_summary": "Summarization plays a crucial role in natural language processing by condensing information into concise summaries, and effective summarization often requires prompt engineering. Wispera excels in this area, offering tools for creating and optimizing chain-of-density (CoD) prompts. The CoD method refines summaries to include more information without increasing length, leveraging Wispera's platform features and user interface. Wispera also provides tools to test and compare different LLMs, enhancing data processing, prompt refinement, and model performance for highly informative and readable summaries."
    },
    {
      "missing_entities": ["entity integration", "task decomposition", "system clause"],
      "denser_summary": "Summarization is vital in natural language processing, condensing information into concise summaries. Effective summarization requires prompt engineering, and Wispera excels in this area with tools for creating and optimizing chain-of-density (CoD) prompts. The CoD method refines summaries to include more information without increasing length, leveraging Wispera's platform features and user interface. Wispera enhances data processing, prompt refinement, entity integration, and model performance. Additionally, it supports task decomposition and includes a system clause to guide the summarization process."
    },
    {
      "missing_entities": ["summary evaluation", "thread comparison", "metric analysis"],
      "denser_summary": "Summarization is crucial in natural language processing, condensing information into concise summaries. Effective summarization requires prompt engineering, and Wispera excels with tools for creating and optimizing chain-of-density (CoD) prompts. The CoD method refines summaries to include more information without increasing length, using Wispera's platform features and user interface. Wispera enhances data processing, prompt refinement, entity integration, and model performance through task decomposition and system clauses. It also supports summary evaluation, thread comparison, and metric analysis for optimal results."
    },
    {
      "missing_entities": ["guideline adherence", "practical application", "expert models"],
      "denser_summary": "Summarization is key in natural language processing, condensing information into concise summaries. Effective summarization relies on prompt engineering, with Wispera offering tools for creating and optimizing chain-of-density (CoD) prompts. The CoD method refines summaries to include more information without increasing length, leveraging Wispera's platform features and user interface. Wispera's tools enhance data processing, prompt refinement, entity integration, and model performance via task decomposition and system clauses. It supports summary evaluation, thread comparison, metric analysis, guideline adherence, practical application, and expert models for optimal summarization."
    }
  ]
}
GPT-4-Turbo Example
{
  "summaries_per_step": [
    {
      "missing_entities": ["AI-driven environments", "diagnostic tools", "personalized treatment plans"],
      "denser_summary": "The article discusses the integration of meta-prompting into language models, enhancing their functionality by breaking down complex tasks into simpler sub-tasks and assigning them to expert models, which significantly improves accuracy and efficiency. This method not only simplifies user interaction by abstracting complexity but also demonstrates flexibility and adaptability across different sectors, including healthcare and customer service, thereby improving operational efficiency and user satisfaction. The future of AI with meta-prompting looks promising, potentially transforming industries by enabling more sophisticated applications and interactions."
    },
    {
      "missing_entities": ["market trends", "anomaly detection", "predictive modeling"],
      "denser_summary": "Meta-prompting revolutionizes language models by structuring complex tasks into manageable sub-tasks handled by expert models, significantly boosting precision and efficiency in AI-driven environments. This advanced technique enhances user interaction by simplifying complex commands into higher-level, task-agnostic prompts, supporting various applications from healthcare diagnostic tools to customer service enhancements. The adaptability and detailed analysis capability, including anomaly detection and predictive modeling, promise transformative impacts across industries, potentially improving personalized treatment plans and operational decision-making."
    },
    {
      "missing_entities": ["real-time data handling", "customer satisfaction scores", "media efficiency"],
      "denser_summary": "Meta-prompting significantly advances language models by decomposing tasks for expert model handling, improving efficiency and precision in AI-driven environments. It facilitates streamlined user interactions and adapts to diverse applications, from enhancing diagnostic tools in healthcare to optimizing customer service protocols. The technique's ability to perform real-time data handling and integrate predictive modeling and anomaly detection contributes to heightened operational effectiveness and better-targeted treatments. Additionally, improvements in media efficiency and measurable boosts in customer satisfaction scores illustrate its broad applicability and potential in refining AI tasks."
    },
    {
      "missing_entities": ["financial forecasts", "fraud detection", "cutting-edge aesthetics"],
      "denser_summary": "Meta-prompting refines AI by splitting complex tasks for specialized handling, enhancing precision and adaptability across AI-driven environments. This method simplifies user-AI interaction, supports real-time data handling, and is applicable in diverse fields including healthcare, where it improves diagnostic tools and treatment plans, and customer service, where it boosts satisfaction scores. Its utility in financial forecasts, fraud detection, and media efficiency demonstrates its broad scope, promising significant advancements in operational efficiency and decision-making, alongside potential applications in creative sectors like design with cutting-edge aesthetics."
    },
    {
      "missing_entities": ["design principles", "historical data", "healthcare sector improvements"],
      "denser_summary": "Meta-prompting enhances AI functionality by efficiently managing complex tasks across various environments, improving precision and user interaction simplicity. It supports real-time data handling and broad applications from healthcare—enhancing diagnostic accuracy and personalized treatments—to customer service, boosting satisfaction scores. The technique's impact extends to financial analytics for accurate forecasts and effective fraud detection, and creative design, integrating cutting-edge aesthetics with traditional design principles. This adaptability and detailed approach promise substantial advancements in sectors like healthcare, optimizing operational efficiency and decision-making."
    }
  ]
}

Help Me Ask You

Are you unsure of the exact prompt to use? Simply ask Wispera for guidance. Generate variations of prompts for a variety of different models enabling you to intelligently crafts the perfect prompt tailored to your unique requirements. Improve the accuracy and relevancy of responses.

This document steps you through navigating the Wispera Designer UI and using an example from the Demonstration pack called Help Me Ask You.

1. Navigating the UI.

  • Log into your Wispera account and navigate to ‘Packs’.
  • Select the ‘Demonstration’ pack.
  • Enter the Packs workbench.

Select Demonstration Pack

Select Demonstration Pack

The workbench for a Pack provides quick access to various pre-built prompts for different use cases. This makes it easy to find a starting point that most closely aligns with what you need to build.

2. Selecting an Example Prompt.

  • Find the Help Me Ask You example prompt.
  • Click on the Workbench icon to get started generating your prompt.

Select Help Me Ask You Example

Select Help Me Ask You Example

3. Review and/or modify the prompt’s structure.

  • Review the Persona clause.
Help Me Ask You Persona Clause
  I want an ideal prompt for {{theThing}}

Help Me Ask You Example - Persona Clause

Help Me Ask You Example - Persona Clause

  • Review The Ask clause
Help Me Ask You - The Ask Clause
  You are an expert {{theProfessionWhoProvidesTheThing}}

Help Me Ask You Example - The Ask Clause

Help Me Ask You Example - The Ask Clause

4. Update variable values, save and review the preview of your prompt.

  • Update variable values.
  • Save your prompt.

Help Me Ask You Example - Update Variable Values

Help Me Ask You Example - Update Variable Values

  • Check the preview of your prompt.

Help Me Ask You Example - Update Variable Values

Help Me Ask You Example - Update Variable Values

5. Running prompts against different models and conducting tests.

  • In the prompt workbench, click the New Thread button.

Help Me Ask You Example - Update Variable Values

Help Me Ask You Example - Update Variable Values

  • Select the LLM you want to test your prompt against from the list of models available.

Help Me Ask You Example - Select LLM and Run Prompt

Help Me Ask You Example - Select LLM and Run Prompt

Testing prompts against various LLMs is a critical step in evaluating their performance. Different models might interpret and process the same prompt in unique ways, leading to varying levels of summary quality. Wispera’s interface provides a seamless way to run these tests, allowing you to compare results efficiently. In this testing phase, theoretical knowledge meets practical application, providing insights that can refine your prompt engineering techniques. For this example the responses we received from two different models the prompt was tested against are below.

GPT-4o Example
Absolutely, crafting an effective elevator pitch is essential for succinctly conveying your idea, product, or service in a way that captures attention. Here's an ideal prompt for creating an engaging and persuasive elevator pitch:

---

**Ideal Prompt for an Elevator Pitch:**

1. **Introduction:**
   - Briefly state your name and your role or profession.
   - Example: "Hi, I’m John Smith, a software engineer and founder of Tech Solutions."

2. **Problem Statement:**
   - Identify the specific problem or pain point your audience can relate to.
   - Example: "Many small businesses struggle with managing their customer relationships effectively."

3. **Solution:**
   - Clearly explain what your product, service, or idea is and how it addresses the problem.
   - Example: "Our app, CRM Pro, automates customer relationship management, making it easier for businesses to track interactions and boost customer satisfaction."

4. **Unique Selling Proposition (USP):**
   - Highlight what makes your solution unique or superior to existing alternatives.
   - Example: "Unlike other CRM tools, CRM Pro integrates seamlessly with social media platforms, providing a comprehensive view of customer engagement."

5. **Market Validation:**
   - Provide evidence or examples that demonstrate the viability or success of your solution.
   - Example: "In our beta phase, businesses using CRM Pro saw a 25% increase in customer retention within three months."

6. **Call to Action:**
   - Specify what you want from your audience, whether it's investment, partnership, or another form of support.
   - Example: "We're looking for investors to help us scale and reach more small businesses. Let’s connect to explore how we can work together."

7. **Closing:**
   - End with a memorable statement or question that leaves a lasting impression.
   - Example: "Imagine transforming customer management into a streamlined, stress-free process. Are you ready to join us on this journey?"

---

**Example Elevator Pitch:**

"Hi, I’m John Smith, a software engineer and founder of Tech Solutions. Many small businesses struggle with managing their customer relationships effectively. Our app, CRM Pro, automates customer relationship management, making it easier for businesses to track interactions and boost customer satisfaction. Unlike other CRM tools, CRM Pro integrates seamlessly with social media platforms, providing a comprehensive view of customer engagement. In our beta phase, businesses using CRM Pro saw a 25% increase in customer retention within three months. We're looking for investors to help us scale and reach more small businesses. Let’s connect to explore how we can work together. Imagine transforming customer management into a streamlined, stress-free process. Are you ready to join us on this journey?"

---

Feel free to customize this template to better fit your specific context and audience!
GPT-4-Turbo Example
Certainly! Here's an ideal prompt to help you create an elevator pitch:

"Imagine you are in an elevator with a potential investor who could significantly contribute to your business. You have only 60 seconds to make an impact. Describe your business idea, its unique value proposition, how it solves a problem or fills a gap in the market, and what you need to take it to the next level. Be concise, compelling, and clear in your explanation to spark interest and invite further discussion."

Simulate a Group Discussion

Propose a scenario for which you would like to assemble a diverse group to involve in conversation, discussion or dialogue. The group could be hypothetical friends or adversaries, a panel of experts, colleagues, a jury … anything.

This document steps you through navigating the Wispera Designer UI and using an example from the Demonstration pack called Simulate a Group Discussion.

1. Navigating the UI.

  • Log into your Wispera account and navigate to ‘Packs’.
  • Select the ‘Demonstration’ pack.
  • Enter the Packs workbench.

Select Demonstration Pack

Select Demonstration Pack

The workbench for a Pack provides quick access to various pre-built prompts for different use cases. This makes it easy to find a starting point that most closely aligns with what you need to build.

2. Selecting an Example Prompt.

  • Find the Simulate a Group Discussion example prompt.
  • Click on the Workbench icon to get started generating your prompt.

Select Simulate a Group Discussion Example

Select Simulate a Group Discussion Example

3. Review and/or modify the prompt’s structure.

  • Review the Scenario clause.
Scenario Clause
I want to assemble a simulated group of people and engage them in conversation.

Simulate a Group Discussion - Scenario Clause

Simulate a Group Discussion - Scenario Clause

  • Review the Expectations clause.
Expectations Clause
You will help me in two main phases:

1. Group Design.
Help me invent those people

and then:

2. Conduct the conversation.
You will play those roles as I engage them in conversation.

Simulate a Group Discussion - Expectations Clause

Simulate a Group Discussion - Expectations Clause

  • Review the Suggestion Clause.
Suggestion Clause
For each phase I will define steps below. You will execute these steps one at a time and report back. Ask me if I wish to proceed with the next step or whether I have some intermediate instructions before proceeding.

If at any time I ask you to proceed, execute the next step. If I present some intermediate request, suspend the execution of these steps until I ask you to proceed.

Simulate a Group Discussion - Suggestion Clause

Simulate a Group Discussion - Suggestion Clause

  • Review the Step-by-Step Clause for the Group Design Phase.
Step-by-Step Clause - Group Design Phase
GROUP DESIGN PHASE

First ask me to provide a scenario to set the bounds of the conversation.

Next, ask me to characterise the group as a whole.

Once I have provide the scenario and am happy to proceed, you will ask me the type of qualities I would like to imbue members of the team with. Answers for qualities could be, for example, “skills”, “qualifications”,  “backgrounds”,  “beliefs”,  “value systems”,  “interests”. You will suggest more specific examples of the types of qualities that would be most appropriate for an informed and lively conversation pertinent to the scenario and group characterisation I provided.

Next ask if me if I wish to refine the list of qualities.

Next ask me if I have a specific number of members for the group. I may specify a number, or else leave it to you. If I specify a number for the group, distribute the qualities among that number and come up with character profiles for each member of the group. If I leave it to you, examine the the qualities and decide a reasonable combination of them into minimal but interesting characters in the group.

Next assignment each member in the group a name and title.

Next, given the arrangement of characters and decide who are most likely to align with each other and who may be in conflict. This will be important for the conversation phase. At times, I expect members of the group to call each other out, directly reflecting on what answers they may have given.

Simulate a Group Discussion - Step-by-Step Group Design

Simulate a Group Discussion - Step-by-Step Clause Group Design Phase

  • Review the Step-by-Step Clause for the Conversation Phase.
Step-by-Step Clause - Conversation Phase
CONVERSATION PHASE

Allow me to ask questions of the group.

With each question, you will select the most appropriate members of the group to speak based on their skill match to the question and also group dynamics. Especially allow for diverse and conflicting views to surface, so that I may gain the best possible perspective on the question I ask.

Simulate a Group Discussion - Step-by-Step Conversation Phase

Simulate a Group Discussion - Step-by-Step Clause Conversation Phase

4. Save and review the preview of your prompt.

  • Save your prompt.

Simulate a Group Discussion Example- Save

Simulate a Group Discussion - Save

5. Running prompts against different models and conducting tests.

  • In the prompt workbench, click the New Thread button.

Simulate Group Discussion - New Thread

Simulate Group Discussion - New Thread

  • Select the LLM you want to test your prompt against from the list of models available.

Simulate Group Discussion - Select LLM and Run Prompt

Simulate Group Discussion - Select LLM and Run Prompt

Testing prompts against various LLMs is a critical step in evaluating their performance. Different models might interpret and process the same prompt in unique ways, leading to varying levels of summary quality. Wispera’s interface provides a seamless way to run these tests, allowing you to compare results efficiently. In this testing phase, theoretical knowledge meets practical application, providing insights that can refine your prompt engineering techniques. For this example the responses we received from two different models the prompt was tested against are below.

GPT4-Turbo-Preview (assistant message)
To begin the Group Design Phase, could you please provide a scenario to set the bounds of the conversation?
GPT4-Turbo-Preview (user message)
help  me invent some characters for my book

Prompt Optimization Example

To Do.