Exploring the Importance of Prompt Engineering for Optimizing Language Models

How Prompt Engineering Works

💡 Note: Prompt engineering plays a crucial role in the optimization and design of prompts for language models, which is a key aspect of natural language processing and language generation. The format of prompts serves as a guide for language models and can be utilized for various tasks such as conversational AI or product-related tasks. Although there are dependable prompt formats, it is highly recommended to explore new formats as well.

👍 Rules of thumb and Examples

Rule #1 - Start with clear instructions and use ### or """ to separate instructions or context. Now let's make the text more engaging!
❌ Rewrite the text below in more engaging language.

(Your Input Here)

✔️ Rewrite the following text using more captivating language

Text: """
(Your Input Here)
"""

Rule #2 - Be specific and additionally, provide comprehensive information regarding the specific context, desired outcome, required length, preferred format, and desired style.
❌ Write a short story for kids
✔️ Write a funny soccer story for kids that teaches the kid that persistence is key for success in the style of Rowling.

Rule #3 - Rule #3 - Illustrate the desired output format with compelling examples.
❌ Bad Example

   Extract house pricing data from the following text
   Text: """
      (Your text containing pricing data)
   """

✔️Good example

   Extract house pricing data from the following text.
      Desired format: """
      House 1| $1,000,000 | 100 sqm
      House 2| $   500,000 |   90 sqm
      ... (and so on)
      """
      Text: """
      (Your text containing pricing data)
      """

Rule #4 - First try without examples, then try giving some examples ❌ Extract brand names from the text below.

   Text: (Your text here)
   Brand names:

✔️ Extract brand names from the text below.

   Text 1: Finxter and YouTube both thrive in the dynamic world of technology. Google as well.
   Brand names 2: Finxter, Youtube, Google
   ###

   Text 2: If you like tech, you'll love Finxter!
   Brand names 2: Finxter
   ###

   Text 3: (your text here)
   Brand names 3:
   ###

Rule #5 - Consider fine-tuning the model if Rule #4 fails to give satisfactory results.
By undergoing fine-tuning, the model's performance is enhanced through extensive training on a larger pool of examples, leading to superior results in terms of quality, token efficiency, and reduced latency for requests.
ChatGPT can intuitively generate plausible completions from few examples, known as few shot learning Fine tuning achieve better results on various tasks withour requiring examples in the prompt, saving costs and enabling lower latency requests.

Example Training Data

("prompt": "<input>","completion": "<ideal output>")
("prompt": "<input>","completion": "<ideal output>")
("prompt": "<input>","completion": "<ideal output>")
...

Rule #6 - Be specific. Omit needless words.
❌ ChatGPT, Compose a brief sales page for my desert sand company. Showcase the value and quality of our product in just a few compelling sentences

✔️ Write a 5-sentence sales page, sell sand in the desert.

Rule #7 - Employ persuasive language to subtly guide the model towards recognizing patterns.
❌ Create a Python function that generates a visual representation of the net worth over a span of 10 years, based on various inputs such as the initial investment and a specified RoI (Return on Investment).

✔️ Good Example

# Python function that plots net worth over 10
# years for different inputs on the initial
# investment and a given RoI

import matplotlib

def plot_net_worth(init, roi):

🤖 Bonus Prompt: Let ChatGPT design the Optimal Prompt

You are a robot for creating prompts.
You need to gather information about the user's goals, examples of preferred output and
any other relevant contextual information.
Prompt should contain the necessary information provided to you.
Continuously inquire the user with additional questions until complete certainty is attained for constructing an optimal prompt. Ensure that your response is appropriately structured and tailored to maximize engagement and effectiveness in ChatGPT interactions.
Be sure to start by asking the user about the goals, desired outcome, and any
additional information you may need