Fundamentals of Effective Prompt Engineering

The launch of foundational models, popularly called Large Language Models (LLMs), created new ways of working – not just for the enterprises redefining the legacy ways of doing business, but also for the developers leveraging these models.

The remarkable ability of these models to comprehend and respond in human-like language has given rise to a new skill of prompt engineering.

In this post, we will learn the need to learn prompt engineering and utilize it efficiently to get the best outcome from LLMs.

Let’s get started.

Why Learn Prompting?

 
Who does not want to achieve more with less? Think of when you can produce better results or drive increased outcomes with equal or lesser effort.

Feels fascinating, right?

That’s what the large language models can do for you – by letting you optimize your time to help you achieve efficiencies, which in turn rolls up to business growth.

A win-win for you and the organization, both. 

Do not just go by my words, I’ll let the numbers speak.

As per a study by Harvard Business School, “ The AI-powered group completed 12.2% more tasks on average than their peers while completing tasks 25.1% faster with 40% higher quality results than those without, per one of the study's authors.”

Now that we understand the benefits of prompting, let’s see how these models are trained fundamentally.

The Foundation of these Models

 
LLMs are trained on data from almost all of the internet, which makes them good at, well, a lot of tasks. 

As much as it sounds like an exciting proposition, it quickly becomes a challenge when enterprises hope to leverage them to solve their specific business problems. 

The reason is that the model needs to be shown and taught those task-specific examples to be exemplary at performing them at enterprise-grade.

Before we proceed further, let’s see some of the most sought-after uses of LLMs:

• Language Translation
• Summarization
• Sentiment Analysis
• Question-answering
• Content creation
• Code generation and debugging

So, for a model to accurately respond to such specific tasks, it needs the right nudge to solicit the contextual response.

The 3C Framework of Prompting

 
The right prompting saves you several hours, yielding faster business outcomes. But, what if you need to prompt iteratively to get to those outcomes – that ends up spending hours learning to prompt effectively?

As much as prompting is a science, it is equally an art of structuring your query in a manner that gives enough context to the model.

And, how much is enough?

Giving too much context might confuse the model, making it fuzzy for it to pay attention to the relevant part. While too little context might deprive it of to better understanding of the intent.

To better understand the context length, it is best to consider it as an optimization problem, wherein you aim to achieve an objective, called a “task”.

However, there are certain constraints under which you often operate, that make it an optimization problem.

That is, how do you plan to achieve an objective given the list of constraints that make up your context space and window?

Hence, I have designed the 3C framework of prompting, wherein you give the model clear and concise instructions with enough context to get accurate and desired outcomes.

To summarize, the key pillars of prompting include answering questions like:

• Who am I?
• What do I want to achieve?
• Where do I need the model’s help?
• What makes a successful output?

In the next article, I will share a detailed step-by-step approach to writing an effective prompt using six elements of effective prompting.
 
 

Vidhi Chugh is an AI strategist and a digital transformation leader working at the intersection of product, sciences, and engineering to build scalable machine learning systems. She is an award-winning innovation leader, an author, and an international speaker. She is on a mission to democratize machine learning and break the jargon for everyone to be a part of this transformation.