Advanced Prompting Guide Unique FinanceGPT Chat

1. Asking for Examples or Illustrations

Explanation:

Requesting examples or illustrations can significantly improve the clarity and understanding of a concept. It allows the user to make abstract information tangible and vivid.

Example Prompt:

"Can you give me an example of how companies use machine learning to improve their customer service processes, including a brief description of the technique used?"

2. Conditional Prompting

Explanation:

Conditional prompting involves giving an AI model a condition or set of conditions that guide its response. This technique is useful when you need the model to consider certain constraints or preferences before generating its output.

Example Prompt:

"Assuming today is a business day and the customer has passed all necessary KYC (Know Your Customer) checks, what would be the next steps to open a new business account?"

This prompt directs the AI to consider specific conditions (business day and KYC checks) before suggesting the next steps. It ensures that the response is relevant and adheres to regulatory and operational guidelines.

3. Few-Shot Prompts

Explanation

Few-Shot Prompts use a few examples to demonstrate the desired task, helping the AI model to better understand the requirements and generate more precise responses. This technique is helpful when specific, but limited examples are available that the AI model can use to calibrate its responses.

Example Prompt

Situation:

A bank employee needs an assessment of how customer complaints about online banking services might affect customer retention.

Prompt:
"Example 1: A customer complaint about slow transaction times led to a quick response from customer service, which resolved the issue.
Example 2: A customer complaint about unclear fees was ignored, resulting in the termination of the account.
Question: How might similar complaints about our online banking affect customer retention?"

By utilizing these two prompting techniques, bank employees can effectively interact with AI systems to extract relevant information and support decision-making processes. Zero-Shot Prompts offer flexibility in new or unexpected situations, while Few-Shot Prompts improve accuracy in more familiar contexts. Both methods are valuable tools for enhancing efficiency and effectiveness in the banking industry.

4. Chain of Thought Prompting

Explanation:

Chain of Thought Prompting involves guiding an AI model through a logical sequence of intermediate steps or thoughts that lead to a final answer or decision. This method helps the AI to process complex tasks more effectively by breaking them down into simpler, manageable parts. It is particularly useful in scenarios where decisions are based on multiple factors or where the reasoning process needs to be transparent.

Example Prompt:

Task:

Calculate the total annual expense for maintaining multiple bank branches, including rent, utilities, and staff salaries.

Prompt:
"To calculate the total annual expense for maintaining our bank branches, we first need to itemize the costs for each branch. Let's start by listing the expenses: rent, utilities, and staff salaries. For each branch, we will sum these costs to get the total per branch. Finally, we will add up the totals from each branch to arrive at the overall annual expense. Let's begin with the first branch located at [Branch Location]. What is the monthly rent, utility costs, and total staff salaries for this branch?"

5. Tree of Thought Prompting

Explanation:

Tree of Thought Prompting is an advanced prompting technique where the prompt encourages the AI to explore multiple branching possibilities or scenarios before arriving at a conclusion. This approach is useful in decision-making processes where different outcomes need to be evaluated based on varying inputs or conditions. It allows the AI to consider a wider range of factors and their possible inter dependencies.

Example Prompt:

Task:

Determine whether to approve a loan application based on various financial metrics and external economic conditions.

Prompt:
"To decide on the loan application, we need to consider several factors: the applicant's credit score, current income level, existing debts, and the current economic conditions. Let's analyze each scenario:

1. If the credit score is above 700 and the income level is stable, evaluate the impact of existing debts.

2. If the credit score is below 700 but the income level is high, assess the risk considering the current economic downturn.

3. If the economic conditions are favorable and the applicant's debts are manageable, consider a conditional approval based on a re-evaluation in six months.
   For each scenario, outline the potential risks and benefits. Begin with the first scenario: what is the applicant’s credit score, and how does it compare to their income and debt levels?"

6. Flip Interaction Pattern

Explanation:

The Flip Interaction Pattern changes the usual dynamic where the user asks all the questions. Instead, the LLM takes an active role in querying the user to achieve a deeper understanding or to fulfill a specific requirement. This method is particularly useful when the user might not know what information is necessary or when detailed clarification is needed to proceed.

Example Prompt:

"Suppose you need to prepare a report on daily banking transactions but are unsure what specific data is required. I will help by asking you a series of questions to clarify your needs. First, could you specify which types of transactions (e.g., deposits, withdrawals, transfers) should be included in the report? Second, do you need this report to include comparisons to previous periods?"

7. Format of the Game Pattern

Explanation:

This pattern turns the interaction into a game format, where the AI is asked to invent or play a game based on specific rules. This can be engaging and encourage creative thinking.

Example Prompt:

Situation:

Imagine you are an office manager tasked with streamlining administrative tasks. You want to use the Game Format Pattern to make the process of scheduling meetings more engaging and efficient. Here’s how you might craft your prompt:

Prompt:

"Let's play a game called 'The Ultimate Scheduler'. In this game, you are an AI assistant tasked with organizing a perfect week of meetings for a busy team. The rules are simple:

1. No two meetings can overlap.

2. Each meeting must allow for at least a 15-minute break between other meetings.

3. Try to accommodate personal preferences for morning or afternoon meetings where possible.

Your goal is to fit all requested meetings into the weekly calendar without conflicts and while maximizing satisfaction among team members. Ready to play? Here are the meeting requests you need to schedule: [Insert meeting details here]."

8. Recipe Technique Pattern

Explanation:

The Recipe Technique Pattern is a structured approach used to extract a detailed process or set of instructions from a given task. By using this technique, the prompt directs the AI to outline a complete sequence of steps necessary to achieve a specific goal, fill in any missing steps, and optionally identify any unnecessary steps. This is particularly useful in administrative tasks where precision and step-by-step guidance are required.

Example Prompt:

"I want to organize the annual compliance training for all employees in the bank. I know I need to schedule the training sessions, notify the employees, and prepare the training materials. Please provide me with a complete sequence of steps for organizing this training. Also, fill in any missing steps that I might not have considered, and identify any steps that might not be necessary."

9. Alternative Solution Pattern

Explanation:

The Alternative Solution Pattern is used to explore different methods or approaches to accomplish a task. This pattern encourages the AI to list possible alternatives to a proposed method, optionally compare the pros and cons of each, and refer back to the original method if required.

Example Prompt:

"If there are alternative ways to process a client's loan application faster than the current method I proposed, please list the best alternative approaches. Compare and contrast the advantages and disadvantages of each approach. Also, mention the original method I asked about, and ask me which approach I would prefer to use."

10. Alternative Prompting Pattern

Explanation:

The Alternative Prompting Pattern involves asking the user for specific input or clarification, which can help tailor the response more accurately to the user's needs. This is particularly useful in administrative tasks where details often vary from one task to another, and precise input is necessary to generate relevant and customized responses.

Example Prompt:

"As part of our quarterly financial review, I need to compile a report on branch performance across the region. Please ask me for the specific data points you need to include in this report, such as total deposits, loan volumes, or customer satisfaction scores. Once you have the necessary data, help me organize it into a comprehensive analysis."

11. Outline Expansion Pattern

Explanation:

The Outline Expansion Pattern is designed to help structure information in a clear and organized manner. Initially, you request bullet points to summarize the main topics. Once the bullet points are established, you delve deeper into a selected point by asking for more detailed information about it. This iterative process helps in breaking down complex information into manageable parts.

Example Prompt:

"Please act as an outline expander. Based on the following details about our bank's administrative tasks: account management, customer inquiries, and compliance reporting, create an initial outline. After reviewing the outline, I will tell you which point to expand further. Can you start by outlining these topics in bullet points?"

12. Menu Action Pattern

Explanation:

The Menu Action Pattern allows users to interact with the AI through a menu-driven interface, where specific commands trigger corresponding actions. This pattern is useful for streamlining interactions and making the prompting process more user-friendly, especially for repetitive tasks.

Example Prompt:

"Whenever I type 'List today’s tasks', you will provide a summary of all scheduled tasks for today. Whenever I type 'Reminder setup', you will ask me for details about the reminder I wish to set, such as time and topic, and then confirm once the reminder is set. After each task, please ask me what you should do next. Let's start by listing today's tasks or setting up a new reminder. What would you like to do first?"

13. Semantic Filter Pattern:

Explanation:

The semantic filter pattern involves extracting specific types of information from a text, such as names, locations, or other identifiable details, and then reproducing the text without these elements. This technique is particularly useful in contexts where privacy concerns or data sensitivity require the redaction of personal information from communications or documents.

Challenges:

It's important to note that this method may not always work perfectly. The accuracy of the extraction and filtering can vary based on the complexity of the text and the specificity of the information being targeted.

Example Prompt:

Suppose you are tasked with removing all personal names and locations from a set of administrative emails to ensure privacy before archiving. Your prompt to the AI might look like this:

Prompt:

"Here is a set of emails. Please filter out all personal names and geographic locations from these emails and provide the redacted versions."

For further techniques also see:

https://jmservera.github.io/miscdemos/prompt-engineering#additional-resources

https://www.coursera.org/learn/prompt-engineering