9 AI 101: Prompt Eng 3.26

Here is the updated breakdown with the examples formatted as specific quotes to show exactly how you would word the prompts:

Definition: A high-level overview comparing vague, "bad" prompts to structured, professional ones.
Example: "You are a senior B2B copywriter. Write a two-sentence LinkedIn ad for our project management SAS. The audience is ops managers. The tone is confident but not salesy."

What is Prompt Engineering (01:44)

Definition: Programming in natural language by defining tasks, roles, and constraints rather than using code.
Example: "You are a tutor. Help me improve the thesis and first paragraph of this 500-word history essay on the causes of World War I. Keep my voice."

How to Prompt Faster (04:07)

Definition: Using voice dictation to provide massive amounts of detail and context quickly, overcoming the "laziness" of typing.
Example: "Hey, I want to write an email to my team about the new project. Include these three bullet points: point one is apples, point two is bananas, point three is pears."

How LLMs "Think" (07:05)

Definition: LLMs are text prediction models that calculate the most likely next word based on the input and provided context.
Example: "Based on our previous conversation about the meeting notes, summarize the main conflicts we discussed."

Steering vs. Commanding (10:40)

Definition: Commanding is a flat instruction; steering provides specific direction on length, style, and focus.
Example: "You are an executive assistant. Summarize the meeting transcript in four bullet points. Focus on decisions and action items. No filler."

Be Specific & Set the Scene (11:53)

Definition: Using a structure of Role, Audience, Tone, and Format to ground the model’s response.
Example: "You are a customer support lead. Reply to this complaint from a paying user whose export failed. Acknowledge the frustration, apologize briefly, and offer a concrete next step. Keep it under 150 words."

Few-Shot Prompting (14:32)

Definition: Providing 1–3 examples of input and output pairs so the model can infer and replicate a specific pattern.
Example: "Feedback: 'App crashed on PDF upload.' Title: 'Upload: iPhone app crashes on PDFs.' Feedback: 'Login is broken on Safari.' Title: 'Login: Safari authentication failure.' Feedback: 'The keyboard does not disappear.' Title:"

Chain of Thought (18:10)

Definition: Asking the model to "reason step-by-step" to reduce errors in logic, math, or complex planning.
Example: "A store sells pens for $2 and notebooks for $5. Sarah buys 3 pens and 2 notebooks with a 10% off coupon. Reason through this step-by-step before giving the final price."

Structured Output (20:05)

Definition: Requesting the model to provide output in machine-readable formats like JSON, tables, or XML.
Example: "Compare Trello and ClickUp for a team of 10. Output this with valid JSON only in this shape: {'tool': '', 'features': [], 'limitations': []}. Do not give me any other text."

Constraints & Negatives (22:04)

Definition: Telling the model exactly what not to do, which can be more effective than simply saying what to do.
Example: "Write a short intro for our onboarding doc. Start directly with what the user will do. Do NOT start with 'Welcome' or generic greetings."

Iterative Refinement (24:19)

Definition: Treating prompting as a conversation where you refine the output based on initial results.
Example: "That's a bit too salesy. Cut it down to two sentences and make it more factual."

Interview Style Prompting (25:22)

Definition: Asking the model to interview you to gather the necessary context before it attempts a complex task.
Example: "I need a LinkedIn post about lessons learned switching to a 4-day work week. Before you write it, interview me and ask one question at a time about the audience, metrics, and tone."

Advanced Techniques & Parameters (29:13)

Definition: Adjusting System Prompts for persistent behavior and Temperature for randomness.
Example: (System Instruction): "You are a direct, technical assistant. Always provide code solutions in Python and never use polite filler like 'I hope this helps'."

Common Mistakes (35:04)

Definition: Avoiding errors like being too vague, overloading one prompt, or assuming cross-session memory.
Example: "First, give me a five-heading outline for this blog post. Once I approve that, we will move on to writing the first section."

9 AI 101