LLMs & Learning

CSE-6040

Welcome

Large Language Models (LLMs) have become part of everyday work for students and knowledge workers alike - but using them well is a skill, not a given.

This module looks past the hype to help you understand what LLMs actually are, how they behave, and how to use them responsibly and effectively in your own learning.

LLMs should be an addition to your tool bag, not a substitution.

Presentation

llms-4-learning.pdf

What You'll Explore

The Fundamentals - What LLMs are under the hood: transformer architectures, attention mechanisms, and the simple predict-sample-repeat loop that powers everything from chatbots to coding agents.
Why Output Varies - Why the same prompt can produce different results, why "hallucinations" happen, and why prompt quality is the one knob you truly control.
Models vs. Harnesses - The difference between the underlying model (the engine) and the interface built around it (the vehicle); and why that distinction matters when choosing tools.
Benefits and Risks - A balanced look at what AI-generated solutions get you, and where they can quietly introduce risk: overly complex answers, homogenized thinking, bias, sycophancy, and outdated knowledge.
Practical Skills - How to prompt with precision, audit AI output critically, and match the right tool to the right task.

Guiding Philosophy

This is meant to teach you to think with LLMs, not instead of them. As the material puts it: LLM technology is not magic - it's understandable. Building that understanding is the first step toward using these tools wisely, rather than being used by them.

A Word of Caution

We're still early in understanding human-AI interaction. Productivity gains are real, but so is the risk of eroding critical thinking. Good prompting requires real domain expertise; and without that expertise, it's hard to know whether an AI's answer is actually right. Keep that tension in mind as you work through this material.

Sample Prompts for Learning

Context	Prompt
Critique a working solution	"Audit my solution and suggest improvements given constraints `<xyz>`."
Find more efficient methods	"Audit my solution and suggest more efficient approaches using only `<xyz>` (e.g., pandas, SQL, numpy)."
Generate practice problems	"Evaluate these problems & solutions, then generate similar practice questions restricted to `<xyz>`."
Identify underlying themes	"What is the general theme here? What should I focus on?"
Learn a package / create study guide	"Faithfully respond from provided sources. Include details, code examples."
-	"Create a report combining our chat + official docs with detailed examples."
-	"Generate high-level overview - dive into specific API mechanics with code."

Tip: Use NotebookLM (aka Gemini Notebook)'s Mind Map and Data Table features for structured learning.