Prompt Engineering For Developers - In a Nutshell

#ai #llm #beginners

Two type of LLM

Base LLM (Pre-trained LLMs) - Base LLMs undergo pre-training on massive unlabeled text datasets to predict the next word or token, gaining broad language understanding and knowledge. These models excel at tasks like text completion or generation but often respond loosely to prompts, producing generalized or unpredictable outputs without specific guidance.
Instruction Tuned LLM - Instruction-tuned LLMs start from a base model and undergo supervised fine-tuning on datasets of instruction-response pairs, teaching them to follow explicit user directives like “summarize” or “explain.” This process, often enhanced by reinforcement learning from human feedback (RLHF), improves precision, reduces hallucinations, and enables better handling of diverse tasks such as question-answering or step-by-step reasoning.

Guidelines

Principle-1: Write clear and specific instructions

Use delimiters - such as triple quotes, backticks, angle brackets etc. to define what are instructions and what is the actual text on which task is to be performed in your prompt
Ask for structured output
Check whether conditions are satisfied, check assumptions are required to do the task
Few-shot prompting - Give successful examples of completing tasks, then ask model to perform the task

Principle-2: Give model time to think

1. Specify the steps to complete a task

Step-1: ______

Step-2: ______

……

Step-N: ______

2. Instruct the model to workout its own solution before rushing to a conclusion

Ask model to solve the problem itself first, then compare with your given solution.

For example, to solve a maths problem give the question first then ask model to solve the question first and also give your solution for comparison in your prompt

Model Limitations

LLM might not have memorized the information it has seen while training. So sometimes it gives plausible response but not actually true
Hallucinations → Make statements that sound plausible but not true

Iterative Prompt Development

Source: deeplearning.ai

Prompt guidelines

Be clear and specific
Analyze why result does not give desired output
Refine the idea and the prompt
Repeat

→ Iteratively refine the prompt for the length and style

Iterative Process

Try something
Analyze where the result does not give what you want
Clarify instructions, give more time to think
Refine prompts with a batch of examples

LLM Capabilities

Summarizing
Inferring → Identifying sentiments, identifying emotions, identifying anger, Extracting something, inferring topics from a text
Transforming → Translation, Tone transformation, Format conversion (e.g. - JSON to HTML), Spellcheck/grammar check
Expanding → e.g. Customize the automated reply to a customer email
Chatbot → Setting Role: system/assitant/user while chatting with the model

Source: deeplearning.ai