Prompt Engineering: Shaping Responses from Large Language Models

As Google is mired in controversy over the inaccuracies of its AI Large Language Model Gemini’s highly biased output, the fiasco underscores an important aspect of all data science use cases: “The quality of input data decides the performance of the machine learning model and the accuracy of output it would generate.” The same principle is sacrosanct in the output that an LLM generates based on a question/prompt it is given by the user.

What is Prompt Engineering?

The process of tailoring an input to receive a certain response from an LLM is referred to as Prompt Engineering. However, a discussion on Prompt Engineering begets an imperative answer as to why one needs to learn Prompt Engineering. LLMs are trained on “everything from everywhere” and therefore, one has to experiment with the precision of one’s prompts to make sure that they receive the necessary response from the LLM that they intended to.

Simple Prompts vs Complex Prompts

The distinction between a simple and a complex prompt can be articulated based on the amount of subjectivity the user is looking for while curating a prompt. For example, prompts like “What is 2 + 2” or “Who wrote Principia” are examples of simple prompts. However, a prompt which is curated in a way where the LLM has to circumvent a few extraneous details and be very terse yet encompassing with its output forms the bedrock of Complex Prompts. In complex prompts the idea is to contours on the creativity of the LLM. The illustration below distinguishes between the two:

Simple Prompt

Complex Prompt

Applications of Prompt Engineering

1.?????? Fine Tuning Responses: ?By tweaking the words or sentence structure of a prompt, you can influence the model to provide more contextually accurate responses.

2.?????? Mitigating Bias: Prompt engineering can be used to reduce biases in model outputs by leading the questions to return a more objective reply based solely on facts.?

3.?????? Enhancing Accuracy: Crafting specific and targeted prompts will help you generate more accurate outputs.

4.?????? Summarizing: Summarizing the text based on specific details provided in the prompt.

5.?????? Inferring: Inferring sentiments or topics based on a review.

6.?????? Transforming: Transforming Texts, Converting them in a different language, spell and grammar checking, tone adjustment, format conversion.

7.?????? Expanding: Generating customer service Emails tailored for each customers use.