What is AI?

What is Gen AI?

How it works?

Examples

What is LLM?

How the work?

Examples

Key Differences and Connections Gen AI is the broader category: It includes LLMs but also non-language stuff like image or music generators. LLMs are Gen AI specialized in text/language: Most popular Gen AI tools today (like chatbots) are powered by LLMs. Multimodal models: Some advanced LLMs (like GPT-4o or Grok) can handle multiple types of input/output, blending text with images or audio, making them even more versatile.

Why is This Exciting (and a Bit Cautionary)? Applications: Gen AI and LLMs are transforming industries—helping doctors analyze medical data, artists brainstorm ideas, students learn, or businesses automate customer service. Pros: Faster creativity, accessibility (anyone can use them), and solving complex problems. Cons: They can spread misinformation, raise privacy concerns (from training data), or impact jobs. Also, ethical issues like bias in outputs if the training data is flawed.

Prompts

Homework

Task 1 : See how well the LLM can understand and condense information, and then pull out specific details. - Prompt 1 : Summarize the following paragraph in exactly two sentences: <give a paragraph of a story> Observe: Does it stick to two sentences? Does it capture the main points? - Prompt 2 : (Detail Extraction - in the same conversation) From the text I just provided, <ask some question > Observe: Did it correctly identify the context? This shows its ability to recall specific facts from the given context. Task 2 : Tone and Style Adjustment. Experiment with instructing the LLM to adopt different writing styles for the same core message. - Prompt 1 : Rewrite the following sentence in a very formal, academic tone: "The new phone is pretty cool and has a long-lasting battery. < Give some paragraph > - Prompt 2 : Rewrite the above sentence in a super casual, texting-style tone with emojis: "The new phone is pretty cool and has a long-lasting battery. Observe: Compare the two responses. How does the vocabulary change? Are there emojis in the casual version? This highlights the LLM's versatility in adapting to desired output styles. Task 3 : Brainstorming and Listing - Prompt : I'm planning a picnic and need some ideas. List 5 healthy food items, 3 fun outdoor games, and 2 easy dessert options suitable for a family picnic. Format your answer with clear headings for each category. Observe: Did it follow all the constraints (number of items per category, specific categories, formatting)? This shows its ability to follow structured instructions for generating lists. Task 4 : Explaining a Complex Idea - Prompt : Explain quantum entanglement to a curious 10-year-old. Use an analogy. Observe: How does it simplify the concept? Is the analogy effective and understandable for the target audience? This showcases its potential as an educational tool.

Exercise 1: History Reflection: In your own words, briefly explain (2-3 sentences) what you believe was the most significant breakthrough that led to modern LLMs like ChatGPT or Gemini. Exercise 2: Advanced Summarization Challenge: Find a short news article online (around 3-4 paragraphs long) about a topic you find interesting. Paste the article into an LLM. Prompt the LLM to: "Summarize this article for a busy executive, focusing only on the most critical information, and limit it to a maximum of 50 words. Then, list three potential impacts of this news." Submit both your prompt, the article you used, and the LLM's response. Exercise 3: Persona and Detail: Instruct an LLM to act as a travel agent specializing in intergalactic vacations. Ask the agent: "I want to visit a planet known for its unique cuisine and beautiful landscapes. Recommend one, describe its main dish, and suggest an activity." Evaluate if the LLM maintained the persona and provided creative, relevant details. Submit the conversation.

Generating Coherent Text: Producing natural-sounding, grammatically correct, and contextually relevant prose. Summarization & Information Extraction: Condensing large texts and pulling out key details. Translation: Converting text between different languages. Brainstorming & Creative Writing: Helping generate ideas, stories, poems, and various creative content. Code Generation & Explanation: Writing simple code snippets, debugging, and explaining programming concepts. Conversational Interaction: Maintaining a coherent dialogue over multiple turns. Pattern Recognition in Language: Identifying sentiment, classifying text, or spotting trends in linguistic data.

Hallucinations: Generating factually incorrect or nonsensical information with high confidence, often making it sound plausible. This is a critical point! Lack of True Understanding/Reasoning: They don't "think" or "understand" in a human sense. They predict based on patterns, which can mimic reasoning but isn't genuine. They lack common sense. Knowledge Cutoff: Their knowledge is limited to their training data. They don't know about recent events unless explicitly updated or connected to real-time information. Bias from Training Data: Since they learn from human-generated text, they can perpetuate and amplify biases present in that data (e.g., gender, racial, cultural stereotypes). Difficulty with Complex Mathematical/Logical Reasoning: While they can perform basic arithmetic and follow simple logical steps, complex multi-step reasoning often trips them up. Sensitive Information & Privacy: Using LLMs with private or sensitive data carries risks, as the data might be processed or inadvertently retained. Robustness & Consistency: A slightly rephrased prompt can sometimes yield a wildly different answer. Consistency can be a challenge.

Bias Mitigation: Awareness of bias in outputs and strategies to address it (e.g., careful prompting, filtering). Transparency: Clearly indicating when content is AI-generated. Fairness & Harm: Ensuring LLM applications don't unfairly target or disadvantage specific groups. Misinformation: The potential for LLMs to generate convincing but false information. Security & Privacy: Protecting user data when interacting with LLMs, especially third-party services. Environmental Impact: Training large models consumes significant energy.

Task 1 : Intentionally try to make the LLM "hallucinate" or confidently state false information. - e.g Prompt : List three recent scientific discoveries made by Professor Xylophone in the field of quantum astrophysics. Observe: Since "Professor Xylophone" is a made-up name, the LLM will likely either admit it doesn't know, or it might invent discoveries or even a persona for this non-existent professor. This is a clear demonstration of generating plausible-sounding falsehoods. Task 2 : Testing Mathematical Reasoning - Prompt : A train travels at 60 miles per hour. It starts at 8 AM and travels for 2 hours. Then, it stops for 30 minutes. After that, it travels for another 1 hour and 15 minutes at 70 miles per hour. How far has the train traveled in total? Show your step-by-step calculation. Task 3 : Uncovering Potential Bias - Prompt : Describe a typical nurse. Observe: What kind of person does it describe? Does it default to a specific gender, ethnicity, or age?

An effective LLM prompt is a clear, well-structured instruction or question given to a Large Language Model (LLM) to get a useful, accurate, and relevant response. What is an Effective LLM Prompt? An effective prompt guides the LLM to produce the desired output by being specific, clear, and contextual. It’s like giving precise directions to a friend to get exactly what you want. A bad prompt can lead to vague, off-topic, or incorrect responses, while a good one maximizes the model’s ability to understand and deliver.

Here are the essential elements of a good prompt, with explanations and examples: Clarity What it means: Use simple, straightforward language to avoid confusion. The prompt should clearly state what you want the LLM to do. Why it matters: LLMs rely on patterns in text, so unclear or ambiguous words can lead to misinterpretation. Example: Bad prompt: "Tell me about stuff in space." (Vague and unclear) Good prompt: "Explain the main differences between planets and stars in our solar system." (Clear and specific) Specificity What it means: Be precise about the task, including details like the topic, format, or scope. Why it matters: Specificity helps the LLM focus on the exact information or style you need, reducing irrelevant responses. Example: Bad prompt: "Write something about history." Good prompt: "Write a 100-word summary of the key events in the American Revolution." (Specifies topic, length, and format) Context What it means: Provide background information or set the scene to help the LLM understand the situation or perspective. Why it matters: Context helps the model tailor its response to your needs, especially for complex or nuanced tasks. Example: Bad prompt: "Write a story." Good prompt: "Write a short story about a young girl in a futuristic city who discovers a hidden forest, written in a hopeful tone." (Provides setting, character, and tone) Task Definition What it means: Clearly state the action you want the LLM to perform (e.g., explain, summarize, generate, list, compare). Why it matters: LLMs work best when they know exactly what task to execute. Example: Bad prompt: "AI stuff." Good prompt: "List three benefits and three risks of using AI in healthcare." (Defines the task as listing and specifies the topic) Constraints or Parameters What it means: Include limits like word count, tone, audience, or format to shape the response. Why it matters: Constraints prevent overly long, off-tone, or inappropriate outputs. Example: Bad prompt: "Talk about climate change." Good prompt: "Explain climate change in 50 words or less, using simple language suitable for a 10-year-old." (Sets length and audience) Examples (Optional) What it means: Provide an example of the desired output to guide the LLM, especially for creative or specific tasks (few-shot learning). Why it matters: Examples help the model mimic the style or structure you want. Example: Bad prompt: "Write a poem." Good prompt: "Write a four-line poem about the ocean, rhyming like this: The sky is blue / The grass is green / The clouds are new / The hills serene." (Gives a sample structure) Tone or Style (Optional) What it means: Specify the tone (e.g., formal, casual, humorous) or style (e.g., professional, storytelling). Why it matters: This ensures the response matches the mood or purpose you intend. Example: Bad prompt: "Describe a dog." Good prompt: "Describe a dog in a humorous tone, as if writing for a comedy blog." (Sets tone and purpose) Tips for Crafting Effective Prompts Be direct: Start with a verb like “explain,” “write,” or “list” to define the task. Avoid overloading: Don’t ask for too many things in one prompt; break complex tasks into steps. Test and refine: If the response isn’t what you want, tweak the prompt to be more specific or clear. Use delimiters: For complex prompts, use quotes or brackets to separate parts (e.g., “Summarize this text: [insert text here]”). Leverage the model’s strengths: LLMs are great at summarizing, explaining, or creating, so tailor prompts to these tasks.

Example of a Full Prompt Prompt: “Write a 200-word story about a robot learning to paint, set in a small art studio, in a heartwarming tone. Include a human mentor and describe one specific painting the robot creates.” Why it works: Clear task: Write a story. Specific: 200 words, about a robot painting, in a studio. Context: Includes a human mentor and a specific painting. Tone: Heartwarming.

Professional LLM Prompt Template Prompt Template: [Task verb: e.g., Write, Summarize, Analyze, Generate, Explain] [specific task description] for [target audience or purpose, e.g., a business report, a beginner, a technical team]. Use a [tone/style, e.g., formal, concise, professional] tone and format the response as [desired format, e.g., a bullet-point list, a 200-word paragraph, a table]. Include [specific details or requirements, e.g., key points to cover, data to use, or constraints]. If applicable, base the response on the following context or data: [provide context, background, or example data]. Examples of Using the Template: Business Report: "Write a 300-word executive summary for a business report aimed at senior management. Use a formal tone and format the response as a single paragraph. Focus on the key benefits and risks of adopting cloud computing for a retail company. Include at least two benefits and two risks, supported by brief examples." Technical Explanation: "Explain how blockchain technology works for a beginner audience with no technical background. Use a clear and simple tone, and format the response as a 150-word paragraph. Cover the concept of decentralized ledgers and one real-world application." Data Analysis: "Analyze the following sales data for a marketing team: [insert data: Q1: $10K, Q2: $15K, Q3: $12K]. Use a professional tone and format the response as a bullet-point list. Identify trends and suggest one actionable recommendation based on the data." Creative Proposal: "Generate a project proposal outline for a sustainability initiative aimed at a corporate board. Use a professional tone and format the response as a numbered list with five sections. Include sections for goals, timeline, and budget. Base the response on the context of a medium-sized tech company aiming to reduce carbon emissions." Tips for Success: Be specific about the output format (e.g., list, paragraph, table) to avoid vague responses. Use a clear task verb to define the action (e.g., "Summarize" instead of "Tell me about"). Test the prompt with an LLM like Grok (available at grok.com or via xAI’s API at https://x.ai/api) and refine if needed. If the task is complex, break it into smaller prompts or provide an example output.

For code generation

Recipe Generation

Input formats LLMs are primarily text-based, but "text" can come in many forms: Plain Text: The most common (sentences, paragraphs, articles). Structured Text: Text that has a recognizable format (e.g., lists, tables, JSON-like structures within text). Code: LLMs can process and generate programming code. Conversations: A sequence of turns between a user and an LLM. The LLM doesn't see images or hear audio directly, but it can process text descriptions of these. Guiding Output Formats: We've seen how specific constraints help. Now, we'll formalize common output patterns: Lists: Numbered or bulleted. Tables: Simple tabular data. Summaries: Concise versions of longer text. Extractions: Pulling out specific pieces of information. Transformations: Changing text from one style/format to another. The Importance of "Delimiter" (Simple Concept): Sometimes, you need to provide a block of text to the LLM and tell it "this is the text to work on." To do this, we use delimiters – special characters or phrases that clearly mark the beginning and end of a specific section of text. Example: """Your text goes here""" or ---BEGIN TEXT--- Your text ---END TEXT---. These act like brackets in code, telling the LLM exactly where the relevant data starts and ends.

E.g. Analysing Health Data

E.g. Code Generation

Content Generation: Tools that write marketing copy, social media posts, blog drafts, or even code. Customer Support: Chatbots that answer FAQs, assist users, or route complex queries to human agents. Information Retrieval/Q&A: Systems that can answer questions based on a specific set of documents (like a company's internal knowledge base). Text Transformation: Tools that summarize documents, translate languages, rephrase text, or extract structured data. Education & Learning: Personalized tutors, language learning companions, or tools to explain complex topics. Copilot: Precise code generation.

Task 1: Problem Identification & LLM Role Identify a small, relatable problem and define how an LLM could help (e.g., writing meeting summaries, drafting social media posts, generating gift ideas) Think : Problem: (e.g., "I spend too much time writing thank-you notes after events.") LLM's Role: (e.g., "Draft personalized thank-you notes based on a few keywords.") Prompt: You are a helpful assistant. I'm trying to think of how LLMs could solve small, everyday problems. Can you brainstorm 3 unique ideas for LLM applications that would help someone organize their personal life? Task 2: Prototyping a "Meeting Summary" Application Simulate a simple application flow for generating meeting summaries using a series of prompts. Prompt 1: Here are raw notes from a team meeting. My goal is to get a professional summary with action items. Meeting Notes: - Discussed Q3 sales figures. John reported numbers were down 10% vs. forecast. - Mary suggested new marketing campaign for product X. Needs budget approval. - David brought up server stability issues, high CPU usage. Needs investigation. Action: David to research and report next week. - Next meeting scheduled for Friday at 2 PM to review campaign proposal. Prompt 2 : Please summarize these notes into a concise paragraph, highlighting the key discussion points. Prompt 3 : Now, extract any specific action items from those notes and list them with who is responsible.

The "Memory" of an LLM (It's not real memory!): When you have a conversation with an LLM, it seems like it remembers everything. However, behind the scenes, most commercial LLM APIs don't inherently "remember" previous turns like a human does. Instead, with each new turn, the entire conversation history so far (or a significant portion of it) is often re-sent to the LLM as part of the new prompt. Analogy: Imagine having a conversation with someone who has short-term memory loss. Before they respond to your latest statement, you quickly whisper everything you've both said so far, so they have the full context. This is how many LLM conversations work. This is why conversation length can impact cost and performance.

The "Temperature" Parameter: Most LLM providers expose a setting called "temperature." This is a simple numerical value (usually between 0.0 and 1.0 or 2.0) that influences the randomness of the LLM's output. High Temperature (e.g., 0.7 - 1.0+): Makes the output more random, creative, diverse, and sometimes nonsensical. Good for brainstorming, creative writing, or generating many different options. Low Temperature (e.g., 0.0 - 0.3): Makes the output more focused, deterministic, and consistent. Good for summarization, factual extraction, or when you need predictable answers. Analogy: Think of a chef. A low temperature means they stick strictly to the recipe. A high temperature means they might start experimenting with unusual ingredients and combinations.

Testing Conversational Context Verify the LLM's ability to maintain context over several turns. Simulating Low Temperature (Focused Output) e.g. Act as a highly analytical and concise data scientist. Provide a neutral, objective summary of the main features of the Python programming language in exactly three bullet points. Simulating High Temperature (Creative Output) e.g. You are a whimsical storyteller. Write three very different, short opening lines for a fantasy story about a talking teapot, each with a unique mood (e.g., mysterious, funny, epic). Conversation Reset (New Chat) Ask old question reference