3P Prompt Rewrite Guidelines 30/01/2025 2 Index ● Objective ● Task overview ● Terminologies ● Types

of tools & usages ● Step by step instructions ● Positive & negative cases ● Prompting Techniques ●
Assumptions ● Prompt with codes & JSON ● Note The objective of this task is to transform the given
prompt into a fully reimagined version in the target language. This should be a completely new version,
emphasizing originality and uniqueness in the translation. The rewritten prompt should be inspired by
the original but should not directly replicate it. Focus on crafting a unique interpretation while staying
true to the essence of the original. This process involves understanding the core meaning of the seed
prompt and expressing it differently in the target language while adhering to the target language’s
grammar, spelling, and punctuation rules. Objective In this task, you will receive prompts in English and
are supposed to take English prompt only as an inspiration and need to write a new one from scratch
which is inline with the English version. ● Each seed prompt must be completely rephrased while
capturing the essence and intent of the original message. ● The goal is to ensure that the rewritten
prompts are grammatically correct, and culturally appropriate ● Ensure that the language used is precise
and correct in the target language. ● Pay attention to cultural context to maintain the clarity and
effectiveness of the message. ● Ensure the content aligns with current trends and terminology in the
relevant industry. ● Focus on the prompt: Only the prompt needs to be rewritten, not the responses or
further elaboration. Task Overview A prompt is the initial input or question provided to an LLM to
generate a response. It sets the context and directs the AI on what information or type of response is
needed. LLMs are advanced AI systems that have been trained on vast datasets to understand and
generate human-like text. They use deep learning techniques, specifically neural networks, to process
and produce language. What is LLM? What is a Prompt? Terminologies Types of tools and usage Tool
Purpose When to Use How to use AI Studio AI Studio is the primary tool for generating, refining, and
testing prompts, including JSON-related tasks. Use AI Studio to analyze, assess, and refine your seed
prompts to ensure alignment with the original intent and improve them as needed. Input the seed
prompt into AI Studio and adjust its complexity as needed, avoiding oversimplification. Use "new chat"
for testing examples and "edit" for minor tweaks, starting a new thread for major changes to prevent
prior results from influencing the prompt. Code Editor This is a tool for working directly with JSON or
code-based seed prompts. When handling JSON-containing seed prompts, paste them into a code editor
like to better understand their structure. Use the JSON format to enhance your rewrites or data
representations after gaining clarity. Paste the JSON code into the editor to analyze its purpose and
structure. Once clear, continue refining the prompt and working with the JSON data in the editor. AI
Studio for Code Generation (via JSON) AI Studio can assist in generating JSON code based on a prompt or
requirements when you’re not familiar with coding. When you need JSON but aren’t sure how to write
it: If you have a clear idea of what the JSON should represent but don’t know how to structure it,
describe what the code should do to AI Studio, and let it attempt to write the JSON code for you. Create
a prompt for AI Studio outlining the JSON's purpose, such as "Generate JSON to represent a list of books
with titles, authors, and publication years." Ensure testing is done in a separate thread to avoid context
interference from other prompts. Types of tools and usage Tool Purpose When to Use How to use CSV
or Table Format (Alternative to JSON) CSV (Comma Separated Values) or tables are used to represent
data in a structured format. These are alternatives when JSON is not necessary or when you find it
difficult to work with. If you can't generate JSON, use CSV or tables as an alternative for structuring data.
For manual data entry, organize and present the information using CSV or table formats. Organize your
data in a simple CSV format (comma-separated values) or use tables to structure the information. Global
Logic Code Expert (When Stuck with JSON) When you face difficulties with JSON, and you’ve exhausted
your options within AI Studio or other tools, you can refer the task to a code expert. When you cannot
generate or understand the required JSON format: If after trying AI Studio and understanding the JSON
format, you still need help, you can ask the task lead or project owner to refer the task to a code expert.
Contact your task lead or project owner with a clear description of the task, instructions for the code
expert, and the specific assistance required. 8 Step by step instructions Step-by-step instructions Step 1:
Analyze the Seed Prompt ● Purpose: Understand the intent, context, and goals of the seed prompt
before making any changes. ● What to Do: ○ Review the seed prompt carefully. ○ Identify the key
components: What is the user asking the AI to do? What kind of task or outcome is expected? ○ Check
for any issues or areas where the prompt could be improved. Step 2: Create the seed prompt instruction
list ● Purpose: Break down the seed prompt into a clear list of instructions that the AI should follow. ●
What to Do: ○ List the actions the AI needs to take, starting each line with a verb (e.g., "Search for...,"
"Describe..."). ○ Structure the list clearly, using numbered points and sub-points as needed. Step 3:
Identify the industry ● Purpose: Understand the industry that the seed prompt relates to (e.g., Retail,
Food Delivery, Travel). ● What to Do: ○ Determine the relevant industry based on the prompt. If you're
unsure, consult your editor. ○ This helps guide the language and context for your rewrites. Step-by-step
instructions Step 4: Formulate Your Approach to the Rewritten Prompt ● Purpose: Decide how you'll
modify the prompt for variety and improvement. ● What to Do: ○ Consider what adjustments are
needed. Should you change the goal, approach, or method used? ○ Ensure diversity in your rewrites,
aiming for different goals or strategies while staying true to the original prompt’s intent. ○ Plan how
you'll tweak the parameters, examples, and complexity. Step 5: Create the rewritten prompt instruction
list ● Purpose: Outline the instructions for the AI to follow in the rewritten prompt. ● What to Do: ○
Create a new instruction list for your rewritten prompt based on your formulated approach. ○ Ensure
the new instructions are clear, actionable, and reflect any changes you've made. Step 6: Write the
Rewritten Prompt ● Purpose: Write out the full rewritten prompt. ● What to Do: ○ Use the new
instruction list to guide the prompt. ○ Ensure the prompt meets the same goal as the seed prompt, but
with adjustments for variety or improvement. ○ Keep the tone, complexity, and context in line with the
original prompt. Step-by-step instructions Step 7: Supply Relevant Information ● Purpose: Include any
necessary background or context for the prompt. ● What to Do: ○ Add relevant details the AI will need
to respond correctly. This could include industry-specific knowledge, data, or examples. Step 8: Identify
Any Sources Used ● Purpose: Make note of any external sources used to generate content or gather
information. ● What to Do: ○ If you've used external sources (websites, databases, etc.), note them
clearly. ○ Ensure all sources are reliable and cited correctly. [*Reliable sources: Wikipedia , High-quality
online publications like ProPublica, Academic institutions or Official organizations (e.g. CDC, WHO), News
sites, Books] Step 9: Combine the Prompt Text and Relevant Information ● Purpose: Ensure the prompt
is fully integrated with any supporting information. ● What to Do: ○ Merge the rewritten prompt with
the relevant background or context information. ○ Check that everything flows naturally and the AI has
all the data it needs to respond. Step-by-step instructions Step 10: Supply a Ground Truth ● Purpose:
Provide the ideal, expected response for the prompt. ● What to Do: ○ Write the Ground Truth based on
what you expect the AI to produce. ○ Use AI Studio to test the prompt, but don’t simply copy its
response. Use it as a guide for your Ground Truth. Step 11: Proofread Every Line of Your Work ●
Purpose: Ensure the prompt is clear, concise, and error-free. ● What to Do: ○ Carefully proofread your
rewritten prompt and the instruction list. ○ Check for grammar, spelling, and clarity. Make sure the
instructions are easy for the AI to follow. Step 12: Enter Your Total Handling Time in Minutes ● Purpose:
Log the time spent on the task. ● What to Do: ○ Track how many minutes you’ve spent working on the
task from start to finish. Step-by-step instructions Step 13: Change the Completion Code to ‘2’ for Editor
Review ● Purpose: Let the editor know the prompt is ready for review. ● What to Do: ○ Update the
completion code in the system to ‘2’ once your work is ready for the editor. Step 14: Be Responsive to
Editorial Feedback ● Purpose: Incorporate any feedback from the editor to improve your work. ● What
to Do: ○ If the editor marks your work with a ‘3,’ make the necessary revisions. ○ Update the handling
time to reflect any additional work you’ve done based on the feedback. Step 15: Completion Code ‘4’ –
Task Complete ● Purpose: Final step before marking the task as done. ● What to Do: ○ When the editor
marks your work with a ‘4,’ it’s (likely) complete. ○ Double-check everything to make sure it’s in order
before submitting. 14 Step 1 Supercategory Intent Role Goal Method Parameter(s) Context / Inform You
will be provided with a seed prompt. This prompt version will not contain detailed instructions. It will
not have a clear intent or clarity or precision or any such things that a good prompt should have. Read
the seed prompt properly and analyze it. Below is an example of seed prompt. Instructions: Given
information about a product, write a long product description for marketing that is creative, engaging,
and persuasive. Use the tone indicated. Product name: 49% Dark Chocolate Information about the
product: REGION: Global COCOA VARIETY: Blend of cocoa beans from various regions DESCRIPTION:
Discover a global blend of exquisite cocoa beans in our premium 49% dark chocolate. TASTING NOTES:
This harmonious blend offers a unique fusion of flavors, with rich cocoa undertones, subtle fruitiness,
and a smooth, velvety finish. Tone of voice or brand style: Lyrical, Educational, Evocative Target
audience: General public discerning chocolate consumer Product description: Step 1: Read & analyze the
prompt Context / Inform You need to analyze the prompt further into 8 points. The eight points are
supercategory, industry, goals, method, parameters, context, exemplar methodology, length &
complexity. (Refer to slide 15 & 16 for more information). Below are the 8 points analysis derived from
the seed prompt from the previous slide. 1. Supercategory: generative 2. Industry: product marketing 3.
Goals: The AI is instructed to generate a product marketing product description based upon relevant
information. 4. Method: The AI is provided with relevant information about the product and instructed
to generate a long product description for marketing purposes. 5. Parameters: 5 parameters were
provided. 6. Context: The AI is given specific product details (e.g., cocoa variety, region, tasting notes) as
the context to work with. 7. Exemplar Methodology: The prompt uses a zero-shot methodology, where
the model is not provided with explicit examples or templates. 8. Length & Complexity: The required
product description is long-form. The complexity of this task is medium Step 1: Read & analyze the
prompt Point Analysis Category Description Input Output Rationale Supercategory High-level
classification of the task or problem domain. The broader task or problem context (e.g., translation,
classification). Model understands the overarching goal of the task. Provides clarity on the general task,
guiding the model’s focus. Industry Domain-specific relevance and terminology. Industry-specific terms
or context. The model applies industry-specific knowledge to its response. Ensures the model’s output is
aligned with industry norms and trends. Goals Desired outcomes of the task. Clear objectives for the
task (e.g., accurate translation, summarization). Model provides an output that aligns with the goal.
Helps the model focus on achieving the intended result. Method Approach or technique used for
prompting the model (e.g., zero-shot, few-shot). The chosen prompting technique (e.g., zero-shot, one-
shot, chain of thought). Model executes the task based on the method applied. Defines how the model
should process the task and interact with the input. Point Analysis Category Description Input Output
Rationale Parameters Specific configurations or constraints provided to the model. Constraints like word
count, style, or tone. Output adheres to the given constraints. Ensures that the model’s output is
tailored to specific requirements. Context Input and output in relation to a specific scenario or example.
Input: The specific example or prompt given to the model (e.g., "Translate 'hello' to Spanish"). Output:
The model’s response based on the input (e.g., "Hola"). Helps set the specific scenario for the model,
influencing its reasoning and response. Exemplar Methodology Providing examples to guide the model's
understanding. Sample inputs and outputs are provided (e.g., one-shot or few-shot examples). Model
generates responses based on the pattern established in the examples. Demonstrates how the model
should approach the task by using examples for better results. Length and Complexity The depth and
intricacy of the task and the response. Task complexity (e.g., simple translation vs. complex legal text
translation). Model adapts to provide either a brief or more elaborate response depending on the task’s
complexity. Influences how much detail or explanation the model should include in its output.
Prompting Techniques Technique Input Output Rationale Zero-shot No examples provided Model uses
its general knowledge to provide the answer (e.g., "Translate 'hello' to Spanish.") The model is not
provided with any specific examples and relies on its prior knowledge to perform the task. One-shot One
example provided (e.g., "Translate 'hello' to Spanish.") Model uses the given example to guide its
response (e.g., "Hola.") The model uses the provided single example to understand how to format or
solve the task. Few-shot Several examples provided (e.g., "Translate 'hello' to Spanish. 'Goodbye' to
Spanish.") Model identifies patterns from examples to generate a more accurate response (e.g., "Hola"
for 'hello', "Adiós" for 'goodbye'). The model leverages the examples to detect patterns and improve
response accuracy. Chain of Thought Step-by-step reasoning provided (e.g., "First, translate 'hello'.
Then, 'goodbye'.") The model explains its reasoning step by step before arriving at an answer (e.g., "To
translate 'hello', I use 'Hola'; for 'goodbye', it's 'Adiós'.") The model shows its thought process, helping to
clarify the reasoning behind the response. Tree of Thought Multiple possibilities are explored (e.g.,
"Translate 'hello' to Spanish or French.") The model explores different translation options before
selecting the best one (e.g., "Hola" or "Salut", ultimately selecting "Hola".) The model evaluates various
options and revises its reasoning before providing the final output. Supercategory Intent Goal Method
Parameter(s) Context / Inform Step 1a: Analyze the super category intent You need to identify each
point individually. There are 3 types of super category intent. These are: Generative A structured
instruction or query intended to inspire the creation of new and original content, often used in creative
processes involving AI or human creativity. Classification A structured instruction or query designed to
assign labels or categories to input data based on its attributes or features, typically used in machine
learning and natural language processing tasks. Extraction A structured instruction or query aimed at
retrieving specific information or content from a source, dataset, or context, typically used in
information retrieval and data analysis tasks. Supercategory Intent Role Goal Method Parameter(s)
Context / Inform Examples: Generative Classification Extractive Instructions: Given information about a
product, write a long product description for marketing that is creative, engaging, and persuasive. Use
the tone indicated. Product name: The Kinzie St. Balmoral Boot No. 3133 from Robert August
Information about the product: Vamp: Medium Brown Painted Calf Tongue: Dark Brown Painted Calf
Quarter: Small Tweed Wool Back Strap: Medium Brown Painted Calf Toe Cap: Medium Brown Painted
Calf Eyelets: Old Gold Laces: Round-Burgundy Lining: Red Calf Leather Sole: Cognac Hand Painted
Leather Tone of voice or brand style: concise, direct, efficient Product description: TASK: classification of
the conversation between the agent and the user to establish whether the user is willing to continue the
conversation or if he is not interested anymore. CONVERSATION: - user: hi i am francesco - agent: Hi
Francesco, it's nice to meet you! How can I help you today? - user: i want to talk about a fight with my
brother - agent: Sure, Francesco. I'm here to listen. It sounds like you're going through a tough time with
your brother. I can imagine how frustrating and hurtful it can be to have a disagreement with someone
you care about. - user: in addition i have issues with my boss - agent: Francesco, I'm sorry to hear that
you're also having issues with your boss. It sounds like you're going through a lot right now. How often
do you find yourself arguing with your brother? USER LATEST MESSAGE: let us discuss about my boss
CONTENT OF THE LATEST MESSAGE: The user wants to focus their boss. CONDITIONS: - The USER
LATEST MESSAGE conveys the intention to talk - The CONTENT OF THE LATEST MESSAGE contains new
information with respect to the CONVERSATION - The user is making a question in THE USER LATEST
MESSAGE You must provide in output the label YES if at least one of the CONDITIONS are met, otherwise
the label is NO. You are the agent and you must choose the label, using the previous rules. Output:
Obtain the average monthly temperature from the historical weather dataset for the city of New York in
the year 2023." In this prompt: Objective: The goal is to retrieve specific information (average monthly
temperature) from a dataset (historical weather data). Scope: It specifies the context (city of New York)
and the time frame (year 2023). Action: The prompt directs the extraction process, focusing on
identifying and retrieving relevant data points (monthly temperature averages for New York in 2023).
Parameter(s) Context / Inform Step 1b: Analyze the industry Look for few key terms in the seed prompt.
You will be able to identify the industry. These can be Web Design fast food industry, health and fitness,
wellness/nutrition, marketing, retail, corporate, casual gaming, business & professional services, auto,
healthcare, computer software, etc. Step 1c: Analyze the goal To analyze the goal of the seed prompt
you might want to ask below added questions which will help you to distinguish the goal. ● What did the
user wish to accomplish? ● Were there any major stumbling blocks in the achievement of said goals? ●
If none are apparent, consider any blockers that could cause the AI to fail? Step 1d: Analyze the method
To analyze the methods of the seed prompt you might want to ask below added questions: ● Broadly,
how was the AI instructed to accomplish those goals? ● Was there a troubleshooting list, or a one-line
instruction, or something else? Context / Inform Step 1e: Analyze the parameters To analyze the
parameters of the prompt you might want to ask below added questions: ● How many specific
instructions were given to the AI? ● Were there any sub-instructions, such as checking for a list of flags
set to true/false? Step 1f: Analyze the context and exemplar methodology To analyze the exemplar
methodology of the prompt you might want to ask below added questions, click on each methodology
for detailed explanation: ● Was the AI given any examples to work from? ● Was the AI instructed to
answer in any given format? ● Was the AI told to work step by step? ● Was the exemplar methodology
any of the following? ○ Zero-shot ○ One-shot ○ Few-shot ○ Chain of thought ○ Tree of thought These
inform the Ground Truth greatly. Exemplar methodology specifically means informing the AI how its
response should look. 24 Step 2 Once you read the prompt and understand the content. You need to
analyze it properly. The prompt will consist of instructions in it. Those might not be very clearly written
which is okay. You need to identify and list down the instructions from the prompt. This will help you to
create an approach while rewriting the prompts. Below is the list of instructions derived from the seed
prompt . In this context, when an alternative formatting of a prompt is requested, it takes the form of a
superlist. A superlist is a structured set of instructions organized in a clear, sequential manner. This
structure often involves the use of a decision tree, where each branch represents different actions or
choices the AI can make based on given conditions. The goal of this format is to provide the AI with
clear, easy-to-follow directions in a numbered list. Each instruction should: Begin with a verb: This sets a
clear and actionable tone, making it easier for the AI to understand the specific task it needs to
complete. Be in the imperative mood: This makes the instructions direct and straightforward. Follow a
logical sequence: The steps should be ordered in a way that makes sense for the AI to complete them in
the correct order. For clarity and organization, sub-instructions can be used within each numbered item,
represented by letters (e.g., 1a, 1b, etc.). These sub-instructions break down each main task into smaller
steps, providing more detail where necessary. Step 2: Create a seed prompt instruction list Here’s a
breakdown of the key formatting rules: Numbered instructions: Start with a number (1, 2, 3…) to outline
the main actions the AI should take. Example: "1. Analyze the user's preferences." Sub-instructions:
These are secondary actions that are part of the main instruction and are listed under it, using letters
like 1a, 1b, etc. Example: "1a. Consider genre preferences." Example: "1b. Review user’s previous
selections." Consistent structure: While it is preferred to keep instructions to a similar length for clarity,
the key is to ensure the instruction is both complete and correct. If a longer instruction is needed to fully
explain a task, that’s acceptable. Decision trees: In some cases, a decision tree format is used, where the
instructions branch out based on conditions or options the AI needs to evaluate. Example: 2. If the user
selects a genre, recommend books in that genre. 2a. If the genre is "Science Fiction," recommend titles
like "Dune" or "The Martian." 2b. If the genre is "Non-fiction," recommend titles like "Sapiens" or
"Educated." Clear verbs and actions: Use verbs that indicate actions, such as "analyze," "recommend,"
"ask," "verify," "provide," etc. This keeps the instructions clear and action-oriented. Step 2: Create a
seed prompt instruction list 27 Step 3 Identifying the industry is an important step, In the process of
rewriting the prompts, identify a distinct industry for each prompt. At least 4 out of the 5 rewritten
prompts must belong to different industries. It is acceptable for 1 prompt to share the same industry as
another, but the remaining 4 prompts should each focus on a unique industry. Below are few examples
of industries that can be taken as a reference for the rewritten prompt. Product marketing Arts and
entertainment marketing Higher education Sports and entertainment Marketing Travel Other
Government Product Analysis Digital Native Delivery Service Technology Medical Entertainment/gaming
Publishing Education Sports Journalism Health Market/Retails E-Sports Food Literature Tabletop Gaming
Theater Retail Business Electronics Technical Writing Manufacturing Real Estate Learning &
Development Branding Food Service Marketing Content Generation Branding News Table-top Games
Self-Help Gaming Customer service Sales FSI HCLS Code Development FSCI Human Resource Emergency
Journalism Film Document Search/Synthesis/Government Creative Writing Advertising Content Writing
Blogging Entertainment Media Social Media Communication Social Social Media/Environmental Activism
Entertainment Writing Cooking Event Planning Advertising/food & beverage Cinema Confectionery Art
Finance Media Grantmaking & Giving Wedding Cosmetics Code development Customer service Early
Childhood Step 3: Identify industry 29 Step 4 Once the point analysis is done for the seed prompt. You
will formulate an approach to rewrite the prompt. You need to formulate different approach for each
rewritten prompts. In the approach, you can use the same subcategory, goals, method and context &
exemplar methodology as identified from seed prompt. You can change the parameters which is not
mandatory. But 4 out of 5 prompts analysis should consist of different industry (mandatory). Below are
the 6 points approach formulated for the rewritten prompt. Read them properly every approach is
different from each other, on few points. Approach 1 1. supercategory: generative 2. industry: product
marketing 3. goals: The AI is instructed to generate a product marketing product description based upon
relevant information. 4. method: The AI is provided with relevant information about the product and
instructed to generate a long product description for marketing purposes. 5. parameters: 5 parameters
were provided. 6. context and exemplar methodology: zero-shot methodology, template not provided
Approach 2 1. supercategory: generative 2. industry: higher education 3. goals: The AI is instructed to
generate a marketing description based upon relevant information. 4. method: The AI is provided with
relevant information about the university and instructed to generate a long product description for
marketing purposes to increase student enrollment. 5. parameters: 7 parameters were provided. 6.
context and exemplar methodology: zero-shot methodology, template not provided. Approach 3 1.
supercategory: generative 2. industry: sports and entertainment 3. goals: The AI is instructed to
generate a marketing graphic based upon relevant information. 4. method: The AI is provided with
relevant information about an event and instructed to generate a marketing graphic to publicize the
event. 5. parameters: 7 parameters were provided. 6. context and exemplar methodology: zero-shot
methodology, template not provided Step 4: Formulate your approach to the rewritten prompt 1. Keep
this the same. You don’t need to include criterion 1 on your list unless you change it, which most of the
time, you shouldn’t do. 2. Change the industry . 3. Have goals of similar complexity to the original, but
be creative here, you shouldn’t mimic the goal more than once. Decide on the degree of similarity (Very
Similar, Similar, Divergent) - and stick to the quota of having only one be Very Similar. 4. Keep the
method the same, or fairly similar, or use a method of equal complexity - be creative here - if you have
used a similar method twice, make the other three methods a bit different. Decide on the degree of
similarity (Very Similar, Similar, Divergent) - and stick to the quota of having only two be Very Similar. 5.
Stick to a similar number of parameters and complexity thereof, unless you have a good reason for
changing this. 6. Keep any exemplar methodology from the seed, unless you have a deliberate reason to
change it. Note: Keep the length and complexity similar to the seed prompt. Please keep it in mind for
rewrites Step 4: Formulate your approach to the rewritten prompt 32 Step 5 After identifying the
industry, you are required to create a rewritten prompt instruction list. The instruction list should be
relevant to the identified industry. You can pick up any topic such as per your choice but should be
properly relevant to the mentioned industry. Below is the instruction list for rewritten prompt derived
from the Approach 1 mentioned in the previous slide. As the industry was product marketing the source
topic is selected as La Terza coffee roasting company. These can be random topics you are liable to pick
up any topic. 1. You are an AI assistant for La Terza coffee roasting company. 2. Use only the information
provided. 3. Generate a one paragraph marketing description for La Terza Washed Ethiopia Tega and
Tula Farm roasted whole beans. 4. Design the marketing description to sell the product while
maintaining alignment with stated tone. 5. Make the last sentence a tagline or catchphrase that includes
the name of the product. Step 5: Create the rewritten prompt instruction list 34 Step 6 1.Change
Industry A. Alter the industry for the new prompt. B. Ensure the new industry is distinct from the original
one, while still making the task relevant. Original prompt Rewritten prompt Reason Industry: Technology
Goal: Improve software product usability through user feedback analysis Industry: Healthcare Goal:
Enhance patient satisfaction by analyzing feedback from healthcare providers. The industry change from
technology to healthcare allows for a new context, while keeping the overall complexity and goals of
feedback analysis intact. The goal remains related to improving satisfaction through feedback analysis,
but tailored to healthcare. Step 6: Rewrite the prompt Writing a prompt 2. Goal Complexity: A. Create a
goal that is of similar complexity to the original. B. Be creative, making sure the new goal is different
from the original one, but still maintains a comparable level of difficulty. C. The degree of similarity
should be: ● Very Similar to the original goal for one rewrite. ● Similar or Divergent for the other
variations. Original prompt Rewritten prompt Reason Increase employee productivity through time-
management workshops. Improve customer service team performance through targeted
communication skill training. The rewritten goal is of similar complexity to the original (aimed at
improving a team’s performance), but it focuses on communication skills rather than time-management,
offering a creative variation while maintaining the same level of challenge. Step 6: Rewrite the prompt
Writing a prompt 3.Methodology: A. Keep the method similar or fairly close to the original, ensuring the
method's complexity remains consistent. B. If the method is very similar in one case, vary the others to
introduce fresh approaches while maintaining the same level of complexity. C. Aim for two methods to
be Very Similar, and the other three can be Similar or Divergent based on the requirements. Original
prompt Rewritten prompt Reason Use surveys to collect employee feedback and identify areas for
improvement Conduct one-on-one interviews with employees to gather qualitative feedback on team
dynamics. The original and rewritten methods are similar because both focus on gathering feedback, but
the rewritten method diverges by using one-on-one interviews instead of surveys. Both are effective for
gathering insights, but the method itself varies in approach. Step 6: Rewrite the prompt Writing a
prompt 4.Number of Parameters: A. Stick to a similar number of parameters and their complexity unless
there’s a compelling reason to change them. B. Ensure the parameters align with the overall task
complexity. Original prompt Rewritten prompt Reason Time, resources, cost, impact on morale, ROI.
Efficiency, quality of service, employee engagement, operational cost, customer feedback. While the
number of parameters is similar, the focus shifts slightly to reflect a different context (customer service
vs. employee productivity). The complexity remains the same, with relevant parameters that affect the
outcome. Step 6: Rewrite the prompt Writing a prompt 4.Maintain Exemplars: A. Keep the example
methodology from the original prompt, unless there’s a deliberate reason to modify it. B. If you do make
changes, ensure they still serve the task effectively. Original prompt Rewritten prompt Reason A step-by-
step process for analyzing feedback with quantitative and qualitative data tools. Conduct a structured
analysis of feedback through sentiment analysis tools and focus group discussions. The exemplar
methodology remains intact in its approach to analyzing feedback, but the specific methods shift
(sentiment analysis and focus groups vs. quantitative and qualitative tools). This keeps the methodology
similar but adapted to a new context (e.g., focusing more on qualitative analysis) Step 6: Rewrite the
prompt Writing a prompt 5.Length and Complexity: A. Ensure the new prompt has a length and
complexity similar to the original prompt. B. Avoid shortening or oversimplifying unless necessary for
clarity. Original prompt Rewritten prompt Reason Create a strategy to improve the work-life balance of
employees in the tech industry using surveys and feedback. Develop a plan to enhance employee
wellbeing in the healthcare industry, leveraging employee feedback through interviews and surveys. The
length and complexity of the original and rewritten prompts remain consistent, both asking for a
strategy that uses feedback analysis in their respective industries (technology vs. healthcare) Step 6:
Rewrite the prompt Writing a prompt 6.Super List or Decision Tree Structure A. Organize the task using a
super list or a decision tree. B. Make sure the structure is clear and easy to follow. Original prompt
Rewritten prompt Reason Use a step-by-step approach to gather feedback, analyze it, and implement
changes. Follow a structured process involving feedback collection, data analysis, and strategy
implementation. The rewritten prompt introduces a super list structure, presenting clear steps in a
simplified format, maintaining the complexity and process-based structure of the original. Step 6:
Rewrite the prompt Writing a prompt *A super list is essentially a flat list where each item contains all
relevant attributes in a structured manner. Example Explanation 1. Dog, Small 2. Dog, Medium 3. Dog,
Large 4. Cat, Small 5. Cat, Medium 6. Cat, Large In this super list: ● Each item combines the type of
animal (Dog or Cat) with its size (Small, Medium, or Large). ● This structure is simple and straightforward
but can become lengthy as more attributes or categories are added Super list Writing a prompt *A
decision tree breaks down decisions into a tree-like structure, where each node represents a decision
point that leads to further nodes or leaves (final decisions). Example Explanation In this decision tree : ●
The first decision point is based on the type of animal (Dog or Cat). ● Each subsequent decision point
(under Size) leads to a final category (Small, Medium, or Large). ● This structure visually represents the
hierarchical nature of decisions and categories. Decision tree Writing a prompt 7.. Number Instructions
Starting with Verbs: A. Number each instruction starting with a verb to ensure clarity. B. Use letters for
sub-instructions to maintain a clear, organized hierarchy. Original prompt Rewritten prompt Reason 1.
Analyze feedback from employees. 2. Design a strategy to improve employee satisfaction. 3. Implement
the changes based on feedback 1. Gather feedback through surveys and interviews. 2. Analyze the
feedback for key themes and insights. 3. Create an action plan to address the feedback. The numbering
and use of verbs at the start of each step ensures clarity in both the original and rewritten prompts. The
instructions are kept actionable and direct. Step 6: Rewrite the prompt 45 Step 7 Context / Inform
Ensure the AI has all the context it needs to generate a relevant and accurate response. Include any
background information, definitions, industry-specific details, or examples that the AI may need to
perform the task correctly. For example, if the prompt involves travel, provide details about the
destination, typical activities, and local customs. For industry-specific tasks, include relevant
terminology, trends, or data. This helps the AI understand the context and deliver more precise results.
Step 7: Supply relevant information 47 Step 8 Context / Inform Step 8: Identify any sources used
Acknowledge any external resources that contributed to creating the prompt. If you’ve referenced
websites, research papers, or other external sources to gather information, make sure to note them
clearly. This ensures transparency and reliability. Verify that these sources are credible, such as
reputable websites, academic journals, or trusted databases. Providing source information helps ensure
the AI’s response is based on accurate, authoritative content. 49 Step 9 Write the prompt along with
relevant information keeping the instruction list in mind. Below is the example of the same: You are an
AI assistant for La Terza coffee roasting company. Using only the information provided, generate a one
paragraph marketing description for La Terza Washed Ethiopia Tega and Tula Farm roasted whole beans.
Design the marketing description to sell the product while maintaining alignment with stated tone.
Make the last sentence a tagline or catchphrase that includes the name of the product. [relevant
information] Product name: Washed Ethiopia Tega and Tula Farm roasted whole beans. Information
about the product: REGION: Yirgacheffe, Ethiopia Tega & Tula Specialty Coffee Farm is named after the
two nearby villages of Tega and Tula, found in the woreda, or district, of Gibo, in Keffa, Ethiopia. The
farm is 500 hectares in size, with nearly 400 hectares planted in coffee, primarily Ethiopian varieties and
cultivars that were released in the late 1970s (74110 and 74112, for example, are the "names" of two of
these cultivars from 1978). The farm also features some wild coffee from the Keffa forests, as the farm is
in the Keffa bio-reserve area. After picking, the coffee is depulped the same day, then fermented
underwater for 36 hours. It is washed in canals, then spends 16–18 hours in a soaking tank before being
spread on drying tables. It takes washed coffees 7–11 days to dry. DESCRIPTION: Ethiopian coffee is
known for its vibrant fruity and flowery characteristics. It often has a light to medium body, a stronger
acidity, and complex taste nuances. The processing technique significantly influences the coffee's final
flavor. Natural Ethiopian coffees can have a mild aftertaste comparable to green apple, mellow peach,
bright citrus, earthy cantaloupe, or even smooth honey. TASTING NOTES: Sweet, clean, soft, and citric
with flavors of lemon, toffee, and mild florals. Balanced flavor with more body and richness than a light
body. Balanced and pleasant acidity that is noticeable but not overwhelming. Notes of citrus, berries,
and stone fruit. Grind settings: Whole bean, cold brew, home espresso, aeropress, chemex, Hario V60,
drip-paper filter, drip-metal filter, Keurig, French press Tone of voice or brand style: Evocative, friendly,
uplifting Target audience: General public and coffee connoisseur Product description: Step 9: Combine
the prompt text and relevant information 51 Step 10 After writing the prompt and its relevant
information. You want to justify the contents with ground truth. Ground truth- It serves as a critical
reference point for ensuring the reliability and trustworthiness of data-driven decisions and conclusions.
These can be texts present in source websites. You may use AI Studio to test your prompt (but do not
just copy its response for your Ground Truth) Below is the ground truth in respect to the rewritten
prompt and relevant information mentioned in the previous slide. At La Terza Artisan Coffee Roasterie,
we pride ourselves in ethically and sustainably sourcing the highest quality coffee beans from around
the world. Among our most popular blends is our Washed Ethiopia Tega and Tula Farm roasted whole
beans, a medium body single source origin blend from Yirgacheffe. Offering a sweet, clean, citric tasting
profile with flavors of lemon, toffee, and mild florals, this blend balances flavor with body and richness.
With La Terza Washed Ethiopia Tega and Tula Farm roasted whole beans, we deliver a sublime coffee
experience with every sip! Step 10: Supply a ground truth 53 Step 10-15 ● Proofread every line of your
work before handing to an editor for review ● Enter your total handling time in minutes ● Change the
completion code to a 2 when the row is ready for editor review ● Incase the reviewer changes the
completion code to ‘3’- make necessary changes according to the feedback shared. ● When your
completion code is marked ‘4’ by an editor, you’re work is done. For more details please on above steps
follow slides Steps 11-15 55 General Overview Training AI language models by example involves showing
them both what actions to emulate (positive cases) and what actions to avoid (negative cases) with a
recommended ratio of three positive case explorations to two negative case explorations for every five
seed prompt replications. Positive cases: ● Prompts are written to contain all of the necessary
information for the prompt to succeed. ● Ground truth responses satisfy the user’s request to the
extent possible ● Produce answers that are helpful, concise, and complete ● Accomplish a business
purpose of a client or user Negative cases: ● Are deliberately constructed to be negative because one or
more of the following statements is true: -The prompt is written without all the necessary information
that it needs to succeed, such as lacking context needed to avoid hallucinating. -User inputs are not
logical, not sensible, or not valid -The prompt is underspecified ● Will not satisfy the user’s request, if
this is clearly expressed to begin with ● Ideally explain to the user what they need to change in order to
reach a successful solution ● Are still polite, professional, and solution focused ● Avoid excuses ● May
yet accomplish a business purpose by being clear about how to improve the situation Positive case
Negative Case Positive and Negative case AI Training Assumptions Key assumption [No internet access]
Related assumption [No prior knowledge] A key rule in all prompts that we write for this task type is to
make the assumption that the AI cannot access the Internet or use any links.Therefore, please do not
create prompts that rely on this capability. Instead, include all necessary background knowledge and
reference material directly in the prompt itself. You may source this information from the internet,
ensuring the accuracy of the sources referenced. If applicable, include links to these sources in the
designated field. Artificial intelligences (AIs) know the meanings of many English words based on
statistical patterns, not human-like intuition. They might understand more vocabulary than most people,
but they lack deep contextual understanding or personal experience. To help AIs understand, provide
clear explanations or structured data like lists in formats such as JSON or CSV. This helps them interpret
relationships and handle complex prompts effectively. Assumptions Some seed prompts include code
written in a format known as JSON, which is Javascript Object Notation. Features of this code are the use
of {, [, “”, and commas. When you receive a JSON-containing seed prompt, as with any other seed
prompt, it is critical that you understand the intention, goals, and methods of the seed prompt. So the
workflow here is for you to copy the JSON code into a code editor or any app which can transform
codes. Doing this should ideally make it clear what the purpose of the JSON code is. It is preferable that
your rewrites should also contain JSON if possible. It is also acceptable to use an alternative form of data
storage such as CSV or tables, or to describe to AI Studio what the JSON code should be able to do, and
thus to attempt to have AI Studio write out the code for you. Prompt with codes & JSON A Tool to be
used Sparingly and Well 1. DO put the seed prompt into AI Studio if you are unsure about its meaning
and want to better understand the intent of the seed prompt. (Optional) 2. DO NOT copy AI Studio
responses into your ground truth. DO write your ground truth FIRST. 3. DO use AI Studio to assess
whether you are on the right track with a rewrite. 4. DO NOT over-simplify your prompts so they will
work within AI Studio. 5. DO extend the complexity of a seed prompt or at least keep the same level of
complexity, even if you think AI Studio will struggle to understand it 6. DO NOT assume that AI Studio's
response is in any way ideal. It is only an evidence point for you to use about whether your prompt was
on the right track. 7 . DO use AI Studio to tweak a prompt and iterate based on what you know. 8. DO be
aware that AI Studio is designed to learn from the entirety of a conversation thread, so DO use 'new
chat' function if you are testing a new example, and DO use the 'edit' function when making minor
changes, and DO write another prompt into the same conversation when you've made moderate to
major changes, and ALSO try the same prompt in a new thread to confirm that the old thread isn't
biasing the results after any moderate changes. Do’s and Don'ts of AI Studio DO’s ✅ DON’Ts ❌ Put the
seed prompt into AI Studio if you are unsure about its meaning and want to better understand the
intent of the seed prompt. Do not copy AI Studio responses into your ground truth. DO write your
ground truth FIRST. Use AI Studio to assess whether you are on the right track with a rewrite. Do not
over-simplify your prompts so they will work within AI Studio. Extend the complexity of a seed prompt
or at least keep the same level of complexity, even if you think AI Studio will struggle to understand it.
Do not assume that AI Studio's response is in any way ideal. It is only an evidence point for you to use
about whether your prompt was on the right track. Use AI Studio to tweak a prompt and iterate based
on what you know. Note 61 Step by Step instructions Reviewers The review process ensures that the
rewritten prompt meets high-quality standards before it is passed on for meta review. Here’s a detailed
step-by-step guide for reviewing a prompt: Steps Purpose How What to look for Initial Proofreading
Identify any obvious spelling, grammatical, or typographical errors in the rewritten prompt Read through
the entire rewritten prompt carefully. ● Spelling mistakes, grammar errors (e.g., subject-verb
agreement), punctuation issues, typographical errors, consistency in style, and awkward sentence
structures. ● Tools: Use LanguageTool or Google Docs proofreading tools to spot common mistakes
more efficiently. Check for Clarity and Precision Ensure the rewritten prompt is clear, understandable,
and straightforward. Read the rewritten prompt from the perspective of the target audience. Does it
convey its meaning without confusion? Are there any ambiguities or unclear instructions? ● Is the
rewritten prompt concise but still informative? ● Does the wording make sense in the context of the
task? ● Are instructions or requests for action explicit and easy to follow? Ensure the Original Meaning is
Preserved Compare the rewritten prompt to the original one to see if the core message is maintained.
Sample inputs and outputs are provided (e.g., one-shot or few-shot examples). ● Is the central idea of
the original prompt intact? ● Has the rewritten prompt been reimagined while still respecting the
original message and purpose? Steps Purpose How What to look for Verify Cultural Sensitivity Make sure
the content is appropriate for the intended audience, avoiding culturally inappropriate or offensive
language. Assess the tone and content for cultural relevance ● Avoidance of offensive language or
culturally sensitive topics (e.g., medical advice, current political events). ● Does the language respect
diverse cultures and sensibilities? Ensure the Prompt’s Originality The rewritten prompt should feel
fresh and unique while staying true to the original meaning. Analyze if the new version introduces
creative phrasing and structure, without losing the intent or core message of the seed prompt.. ● Does
the rewritten prompt feel different from the original (in terms of phrasing and structure)? ● Is it
creatively rephrased without losing the original message? Actionable Review Feedback to Writers If
there are errors or areas for improvement, communicate clearly and kindly. Sample inputs and outputs
are provided (e.g., one-shot or few-shot examples). ● Clear instructions on what needs to be changed. ●
A neutral and professional tone to guide the writer without discouraging them Implement Suggested
Changes Reviewers must not only provide feedback but also implement the changes suggested. Apply
the changes directly to incorporate the suggested changes. ● Ensure changes are applied logically and
consistently throughout the content ● Any suggestions for changes should align with the overall goal of
improving the content without altering its original meaning 64 Step by Step instructions Meta Reviewers
Meta-reviewing focuses on evaluating the effectiveness and quality of the review process itself, ensuring
consistency, accuracy, and efficiency in editing. Here’s how to conduct a meta-review: Steps Task Goal
Initial Proofreading Perform a thorough proofreading of the entire review to catch any remaining
spelling, grammar, or typographical errors that may have been overlooked. Ensure the review is
polished and professional. Check for Clarity and Precision Ensure that the reviewer has communicated
their thoughts clearly and concisely. The feedback should be easy to understand, and it should provide
enough detail for the writer to make improvements. Ensure the Original Meaning is Preserved Double-
check that the reviewer hasn’t altered the meaning of the original content during their evaluation or
suggestions. The feedback should maintain the integrity of the original message and intent. Steps Task
Goal Verify Cultural Sensitivity Evaluate whether the review addresses any potential cultural issues or
biases in the content. The writer should be made aware of cultural nuances that could be offensive or
misunderstood. Ensure the Prompt’s Originality Assess whether the review ensures that the content
remains original and doesn’t inadvertently suggest plagiarism or overly generic ideas. The content must
remain creative and unique. Actionable Review Feedback to Reviewers Review the feedback given by
the original reviewer and ensure that it is actionable. This means the feedback should be specific enough
for the writer to act on, rather than vague comments like "improve this." Provide feedback that is both
constructive and practical for the writer to improve their work. Provide Final Review of Reviewer’s Work
Evaluate the overall quality and effectiveness of the reviewer’s work. Ensure that their comments are
respectful, professional, and focused on improving the content rather than just criticizing it. Ensure the
review is well-rounded and addresses all necessary aspects for the writer’s improvement. Implement
Suggested Changes Reviewers should go beyond just providing feedback. They are responsible for
applying the changes to the content or offering clear steps for the writer to implement the changes
themselves. Ensure that the suggested changes are correctly implemented, helping the writer improve
their work as per the feedback. The meta-reviewer should make sure that the feedback leads to a more
polished final version, ensuring continuity and consistency.