Lecture 2 of 5β±οΈ 45 Minutesπ» Practice Heavy
π‘ Inspired by: Sal Khan & AI4ED β focusing on AI as a Socratic tutor, not an answer machine
AI as Your Research & Study Partner
AI doesn't "understand" your subject β but it can be an incredibly powerful thinking partner if you know how to talk to it.
This lecture teaches you prompt engineering without code: how to ask AI better questions and get dramatically better answers.
Concept Discussion β 20 Min
The Shift: From Search to Conversation
For 25 years, we trained ourselves to speak "Google"βusing keywords like "dhaka university admission 2024 dates". AI requires entirely different interaction. You must speak to it like an intelligent, but highly literal, undergraduate assistant.
40%
Higher quality output when using Role-based prompting
10x
Faster workflow when using AI as a research indexer block
The CREATE Framework
A structured approach to prompt engineering guarantees higher quality, academically rigorous outputs.
Context: "I am a 3rd-year Economics student studying macro-trends."
Role: "Act as a rigorous Oxford debate coach."
Explanation: "Analyze this thesis statement for logical fallacies."
Action: "Provide a 3-point critique..."
Tone: "...in a formal, academic tone without using jargon."
Extras: "Do not write the essay for me, only provide the critique."
Zero-Shot vs. Few-Shot Prompting
LLMs are highly sensitive to "In-Context Learning". If you just ask a question, that's "Zero-Shot". To get high-level academic results, you must use "Few-Shot" prompting.
Zero-Shot: "Write a thesis summary." (Results will be generic and uncalibrated).
One-Shot: "Here is an example of an excellent summary: [Paste Example]. Now write a summary for this new thesis."
Few-Shot: Providing 3-5 examples of inputs and desired outputs forces the model to mathematically map to your exact formatting and nuance.
The Persona Pattern
Why does telling the AI to "Act like a PhD Professor" actually make it smarter?
In high-dimensional latent space, words are connected by mathematical weights. By assigning a persona, you restrict the model's probability distribution to access only dense, expert-level vocabulary and reasoning structures, cutting out the 'average internet user' noise from its responses.
Academic Integrity in the AI Era
Using AI is not inherently cheating, but ghostwriting is plagiarism. The paradigm shift is using AI to augment your cognition, not replace it.
The "Whiteboard Test"
If you submit an assignment and your professor asks you to step up to the whiteboard and defend its arguments, can you do it? If you used AI to write it for you, you will fail the whiteboard test. Use AI for outlining, brainstorming, overcoming writer's block, and proofreading β but the core thesis must be yours.
π₯ More on How Models Learn
Hands-On Activity β 20 Min
π Activity 1: Bad Prompt vs. Great Prompt
We will execute a live A/B test. Take a complex topic from your current syllabus.
Step 1 (The Bad): Enter a basic search prompt. Record the generic result.
Tell me about X.
Step 2 (The CREATE): Build a 5-sentence prompt using the Context-Role-Explanation-Action-Tone framework.
I am a second-year student at DU. Act as a strict academic advisor. Explain the core concept of [Topic] and create a bulleted outline. Keep the tone formally academic.
Step 3: Present the difference in depth and utility to the class.
π Activity 2: Introduction to NotebookLM
Open Google's NotebookLM (a free tool by Google Labs). Upload a PDF of an academic paper you are currently studying. Do NOT ask it to write an essay. Instead, use these prompts:
Create a comprehensive study guide from this document.
What are the weaknesses in the author's methodology?
Generate a 5-question multiple choice quiz to test my understanding.
π Activity 3: Design a Study Session
In your groups of 5, design a 30-minute AI-assisted study session for your faculty's subject:
Which AI tools would you use and for what specific tasks?
What would you still do without AI? Why?
What verification steps would you include to check AI accuracy?
Present your study session plan to the class (2 mins per group)
π§°
Your Takeaway: The Personal Prompt Toolkit
Save 3 customized "mega-prompts" tailored to your specific major (e.g., one for summarizing dense readings, one for brainstorming essay structures, one for debating your thesis). Keep these in a digital note app to use throughout your degree.
π Sample Prompt Templates to Get You Started
The Summarizer: "Summarize [topic] in [X] bullet points for a [year]-year [subject] student at DU"
The Critic: "I believe [argument]. Challenge this position with 3 counterarguments"
The Simplifier: "Explain [complex concept] as if I'm hearing it for the first time. Use a Bangla cultural analogy"
The Comparer: "Create a table comparing [A] vs [B] across these dimensions: [list]"
The Quiz Master: "Generate 5 practice questions on [topic] at university exam level with answers"
The Draft Helper: "I need to write [type of document] about [topic]. Give me an outline with key points for each section"