Lecture 11 – Prompts, Message Types & How LLMs Actually Generate Text

Lecture 11 goes deeper into how you actually talk to an LLM — and what happens inside the model when it responds. This is a foundational lecture for anyone who's going to write prompts or build LLM-powered apps. What we cover in this lecture: What a prompt is — any text input (instruction or question) given to an AI model The 3 message types every LLM API uses: System message — the instruction/rule set by the developer (e.g., "You are a medical expert, only answer medical questions") that defines the model's role and boundaries User/Human message — the actual question or input from the person using the app (mandatory — no input, no output) AI/Assistant message — the model's generated response Why the system message is optional but powerful — and can even be made dynamic/conditional based on user input (template-based system messages) Guardrails and jailbreaks — why system messages should stay private, and how developers prevent a model from leaking its own instructions Tokens — how LLMs break text into tokens rather than processing raw words How LLMs actually generate answers: taking the entire input in parallel, but generating output sequentially, one token at a time, appending each new token back into the input A simplified look at the transformer architecture — encoder (understanding relationships between words) and decoder (generating the next word) Vocabulary and probability — how a model picks the next token from its full vocabulary (often 100,000+ tokens) based on calculated probabilities Before next class: Install Ollama to start running open-source models locally. Who this is for: Anyone following the Gen AI module of the course — Lecture 10 (What is Gen AI & LLM) is the direct prerequisite for this one. Topics/Tags: Prompt Engineering Basics, System Message vs User Message, LLM Tokens Explained, How LLMs Generate Text, Transformer Architecture Basics, Next Token Prediction, GenAI for Beginners, Learn GenAI 2026, LLM Guardrails, Ollama Setup #GenAI #PromptEngineering #LLM #Tokens #Transformers #AITutorial #LearnAI