Fundamentals of Prompt Engineering

Why Prompt Engineering ?

When embarking on the journey of building a machine learning application, the choice of approach can significantly impact development time and efficiency.

In this video we explore three common methods:

Starting from Nothing: Here you build your own LLM from scratch. Process begin by collecting and preparing labeled training data, where each input is associated with the correct output. Followed by training, validating , fine-tuning hyperparameters and retraining until we get our perfect model. And then deploy, serve and monitor. Timeline: This approach can be time-consuming, requiring 9 to 12 months, as it involves extensive data preparation and model training. Complexity : In terms of complexity it is considered High.

Fine-Tuning: The second approach is Fine-tuning on top of a Foundational Model. Process start with a foundational model , preparing fine-tuning data, fine-tuning the selected foundational model until it meet the desired expectation and finally deploy, serve and monitor. Timeline: While this is faster than the first approach, fine-tuning still takes time, typically ranging from 3 to 6 months. Complexity : In terms of complexity it is considered Medium.

Third approach is Prompt-Based Approach also called Instruction fine-tuning on Instruction tunned model: Process starts with Utilizing a pre-trained Instruction tunned model , then Defining Prompt using prompt engineering techniques for instruction fine-tuning, Send this prompt to the LLM and receive LLM response. Remarkably, this approach allows you to start building your application and witnessing prototype results within just a few hours.

Clearly the complexity in this case is Low Thus, when devising a use case and considering the application of LLM and GenAI for business problem-solving, your initial strategy should ultimately focus on leveraging Instruction-tuned LLMs through Prompt Engineering.

What is Prompt Engineering ?

In this video we are going to answer this question. First we will try to understand what is Prompt with some real life analogy and then we will talk about prompt engineering.

What all is covered:

What is Prompt?

What is Prompt Engineering with a real-life analogy

What is Prompt types. Ex: System,Instruction and user prompt

Characteristics of System prompt

Usage of System Prompt

Finally develop a GenAI use case where we will be developing a Digital Assistant for library.

In this video I talked about following Tactics for writing Clear Instructions.

Tactics for writing clear instructions:

Include details in your query to get more relevant answers

Ask the model to adopt a persona

Use delimiters to clearly indicate distinct parts of the input.

Specify the Steps required to complete a task

• Colab Jupyter nootbook is shared for hands-on

In this Video we are going to talk another Prompt Engineering tragegy: ? Breaking Complex Task into Simpler Sub-tasks. In this video I talked about this strategy using a real worl use case "Log Analysis".

Following topics are covered in this Lession:

• Why this strategy - Breaking Complex Task into Simpler Sub-tasks

• Advantages

• Implementing Log Analysis using Language Model (GPT 3.5 turbo).

• Colab Jupyter nootbook is shared for hands-on

In this Video we are going to talk Zero-Shot Prompt Engineering Technique.

We are going to cover the followings:

• What is Zero-Shot prompting

• Benefits of Zero-Shot prompting

• Real word use cases using Zero-Shot Prompting.

1. Implementation of following use cases:

2. Language Translation

3. Sentiment Analysis

4. Text Classification

5. Summarization

6. Question Answering

7. Information Retrieval

8. Text Generation

9. Named Entity Recognition (NER)

10. Content Moderation

11. Recommendation Systems

• Colab Jupyter nootbook is shared for hands-on

In this Video we are going to talk Zero-Shot Prompt Engineering Technique.

We are going to cover the followings:

• What is One-Shot prompting

• Benefits of One-Shot prompting

• Real word use cases using One-Shot Prompting.

Implementation of following use cases:

1. Customer Support Response Generation

2. Social Media Post Creation

3. Creating JSON from Raw Data

4. Complex data formatting

5. Log Analysis Use Case

6. Educational Summaries and Explanations

• Colab Jupyter nootbook is shared for hands-on

In this Video we are going to talk Zero-Shot Prompt Engineering Technique.

We are going to cover the followings:

• What is Few-Shot prompting

• Benefits of Few-Shot prompting

• Real word use cases using Few-Shot Prompting.

  
  

Implementation of following use cases:

1. Legal Document Analysis

2. Customer Support Automation

3. Financial Report Summarization

4. Medical Information Extraction

5. Content Generation for Marketing

6. Recipe Generation

7. Sentiment Analysis with Context

8. Translation with Cultural Nuances

9. Personalized Email Responses

10. Log Analysis

• Colab Jupyter nootbook is shared for hands-on

In this Video we are going to talk Chain-Of-Thought Prompt Engineering Technique.

We are going to cover the followings:

• What is Chain-of-Thought in general

• Understanding Chain-of-Thought in LLMs

• Real word use cases using Chain-Of-Thought Prompting.

• When to apply CoT

• Properties of Chain-of-thought(CoT)

• Why CoT fails in Small Language Models

• How to fine-tune small LM for CoT

• Colab Jupyter nootbook is shared for hands-on