Tuesday, January 3

Learn Prompt Engineering with ChatGPT & Google Gemini : A Simple Guide with Examples

In the era of generative AI, prompt engineering has emerged as one of the most essential skills for effectively interacting with AI.  For beginners 3 popular AI tools are 

1) ChatGPT (by OpenAI) 

2) Gemini (Google's)  and  

3) Claude (Anthropic) and 100+ other similar AI tools

So how does this Artificial Intelligence really work?

1) You know a computer runs a Software and can Search Internet and store data

2) 10000 computers can Search Internet and store data much faster than 1 computer

3) If I tell computer to search data about about Dolphin or Coffee it can search and store all information and when I ask a question it can then reply to any question related to Dolphin or Coffee in seconds

4)  That is how AI works. There are thousands of computers, all connected to each other,  searching information about 'certain words' and storing it in a Database so that when you ask a question they can reply in seconds. (You are right! That is tremendous waste of resources & electricity and that is contributing to global warming but I will write about it in another post)

5)  Prompt Engineering is a way of writing a command so that computer knows what 'exactly you are looking for' and give you best results and not a generic reply as you get in Google Search.

6) If I want to explain how Coffee is made to a 10yrs old I should tell computer in so many words so he gives an answer with examples so that 10yrs old can understand it

7) If I want computer to explain 30yrs old how to 'Brew Coffee' at home then I should write a command so that I get  a relevant answer 

8) Just understand that the more context you provide to AI, the better reply you will get from AI 

Even before you start reading you should know the answer to following basic questions 

  1. AI is fantastic but can a layman talk to these AI tools & get a good reply?

  2. Is there any use of Artificial Intelligence in life of a wife, mother, student, lawyer, doctor, chef?

  3. Can I start using AI from today?

  4. I am 79yrs old, can AI help me in any way and can AI learn to use AI? 

Answer to all the above questions is YES

Prompt engineering is the technique of writing questions (prompts) to get desired outputs from AI systems or in simpler words. Prompt Engineering is way of writing enough details in your question to AI to get best results. This guide is aimed at beginners who are curious about prompt engineering, offering a comprehensive overview of the fundamentals, techniques, and practical examples. If you want to read a 'Marathi' version of this blog please click on this link प्रॉम्प्ट इंजिनिअरिंग: सविस्तर मार्गदर्शक (उदाहरणांसह)

What is Prompt Engineering?

Prompt engineering tells you what details you should put in your question to AI in a way that gives the most useful, relevant, and accurate results. Because ChatGPT generate responses based on patterns learned from massive datasets, the way you ask a question can significantly influence the answer.

In essence, prompt engineering is about:

  • Understanding how ChatGPT or Gemini interpret and respond to input.

  • Designing prompts to guide the model's behavior.

  • Iterating and refining prompts to improve outcomes.

Why Prompt Engineering Matters

AI models are highly capable, but they are not mind readers. They depend entirely on the text provided. Subtle variations in phrasing, tone, specificity, or structure can change the results dramatically.

Benefits of good prompt engineering include:

  • More accurate and relevant outputs.

  • Reduced hallucinations or fabricated content.

  • Increased efficiency in achieving results.

  • Better alignment with business, educational, or creative goals.

Basic Principles of Prompt Engineering

  1. Clarity

    • Clear prompts produce clearer responses.

    • Avoid ambiguity.

  2. Specificity

    • The more specific the prompt, the better the output.

    • Specify the format, tone, length, or point of view if needed.

  3. Contextualization

    • Provide background or context to help the model generate more informed responses.

  4. Instructional Language

    • Use imperative or guiding language: "List", "Summarize", "Compare", etc.

  5. Iteration

    • Refine and reword prompts based on outputs.

    • Use feedback loops.

Types of Prompts - 

By the way, if you want you can open https://chatgpt.com, register with Google id and get started  using these ready made prompts (blue text)

  1. Descriptive Prompts

    • Example: "Describe the atmosphere of Mars."

      Example: "Describe climate at Hawaii in September 2026 to plan a family holiday " 

  2. Instructional Prompts

    • Example: "Explain how a blockchain works in simple terms."

      Example: "Explain how a Aeroplane works in simple terms in 2 paragraphs."    

  3. Creative Prompts

    • Example: "Write a poem in Marathi language about a 10yrs old girl enjoying rains."

  4. Comparative Prompts

    • Example: "Compare the economic policies of USA and India in a tabular format."

      Example: "Explain why per capita income of Srilanka is more than India when India's GDP is much higher than Srilanka." 

  5. Conversational Prompts

    • Example: "Pretend you're a tour guide in ancient Rome. Walk me through a day in the city."

Common Techniques in Prompt Engineering

  1. Zero-Shot Prompting

    • Asking the model to perform a task without providing examples.

    • Example: "Translate this sentence into French: 'The sky is blue.'"

  2. Few-Shot Prompting

    • Providing a few examples to guide the model.

    • Translate the following sentences to French:  1. The apple is red. 
      -> La pomme est rouge. 

  3. Chain-of-Thought Prompting

    • Encouraging the model to reason step by step.

    • Example: "If there are 3 apples and you take away 2, how many are left? Explain your reasoning."

  4. Role-based Prompting

    • Asking the model to adopt a specific role or persona.

    • Example: "Act as a professional career coach and give resume tips."

      Example: "As a doctor with 10yrs experience analyze attached CBC report of a CML patient , compare it with reports of patient with same age and provide a summary" 

      ( You will have to attach a image or pdf of the report for above prompt using + sign next to typing area in ChatGPT ) 

      How to add files

  5. Prompt Templates

    • Predefined prompt formats to standardize input.

    • Templates are useful in automation and large-scale tasks for example running a advertisement campaign or sending mails to large number of invitee etc 

Tips and Best Practices

  1. Be Iterative

    • Start simple and refine as needed.

  2. Use Constraints

    • Limit word count, specify format (e.g., bullet points), or define tone (e.g., formal, friendly).

  3. Test for Edge Cases

    • See how the model responds to unexpected inputs.

  4. Break Down Complex Tasks

    • Use a series of prompts for step-by-step tasks.

  5. Utilize System Messages (if supported)

    • Many APIs allow for system-level instructions to guide behavior consistently.

Examples of Effective Prompting

  1. Basic to Advanced Prompting

    • Basic: "Tell me about Newton's laws."

    • Better: "Summarize Newton's three laws of motion in simple language for a 10-year-old."

  2. Formatting Output

    • Prompt: "List the benefits of solar energy in bullet points."

  3. Using Roles

    • Prompt: "You are a chef. Give me a quick, healthy dinner recipe using spinach and chickpeas."

  4. Creative Prompting

    • Prompt: "Write a short science fiction story about AI taking over Mars colonies."

  5. Chained Reasoning

    • Prompt: "Solve this math problem step-by-step: What is 25% of 240?"

Challenges in Prompt Engineering

  1. Ambiguity in Prompts

    • Unclear inputs lead to unpredictable outputs.

  2. Hallucinations

    • Models may generate false or fabricated information.

  3. Token Limitations

    • Each model has a maximum context window (measured in tokens).

  4. Bias and Ethics

    • Outputs can reflect biases present in training data.

  5. Consistency

    • Responses may vary between runs even with the same prompt.

Applications of Prompt Engineering

  1. Software Development

    • Code generation, debugging, documentation.

  2. Marketing

    • Ad copy, email campaigns, content ideas.

  3. Education

    • Personalized tutoring, lesson planning, quiz generation.

  4. Research

    • Summarizing papers, generating hypotheses.

  5. Creative Arts

    • Poetry, storytelling, idea generation.

Future of Prompt Engineering

As AI models grow more sophisticated, the role of prompt engineering will evolve. The future may include:

  • Prompt programming languages: Tools or DSLs for structured prompting.

  • Multi-modal prompting: Integrating text with image, audio, or video inputs.

  • Automated prompt optimization: AI optimizing prompts for best results.

  • Embedded prompt layers: Built into apps and workflows seamlessly.

Conclusion

Prompt engineering is the bridge between human intent and machine response. It's a powerful tool that unlocks the potential of AI, enabling users to tailor outputs to their specific needs. By understanding the fundamentals, practicing different techniques, and learning through iteration, anyone can become proficient in this modern skill.

Useful links
By -

No comments:

Post a Comment

Subscribe to: Post Comments (Atom)

प्रॉम्प्ट इंजिनिअरिंग: सविस्तर मार्गदर्शक (उदाहरणांसह)

मी हा लेख कृत्रिम बुद्धिमत्ता (Artificial Intelligence) आणि तिचा वापर कसा करावा याबद्दल लिहिला आहे, जेणेकरून इंग्रजीत सहज बोलू न शकणाऱ्या आप...