Lecture 1 of 3

What is an LLM?

Demystify the technology

Overview

Large Language Models like ChatGPT have captured the world's attention, but what actually are they? This lecture cuts through the hype and jargon to give you a clear understanding of the technology—where it came from, how it works, and what it can (and can't) do.

Hays shares his perspective as someone who uses AI 8-10 hours a day in his work as a software engineer, including live demonstrations and practical examples from his own experience.

What This Lecture Covers

The 70-Year "Overnight Success"

From Turing asking "Can machines think?" in 1956, through the AI winters of failed symbolic approaches, to the breakthroughs of ImageNet (2012), the transformer architecture (2017), and ChatGPT's explosive arrival (2022)—the fastest technology adoption in human history.

How Neural Networks Work

Using your brain as the model: neurons, connections, and learning through pattern recognition. We walk through how these networks get trained—from recognizing edges in images to understanding the patterns in all the text ever written.

LLMs: Fancy Autocomplete?

Yes, but with a crucial twist. The "attention" mechanism lets these models remember context across thousands of words. They're prediction machines trained on essentially every piece of digital text that exists, wandering through "semantic space" to generate responses.

Live Demonstrations

ChatGPT, Claude, and Gemini in action—from asking about current events to selecting wine from a restaurant menu to generating images of historical figures who never met. See both the impressive capabilities and the limitations firsthand.

Practical Tips for Using AI

Why vague prompts get vague answers. How to give context that actually helps. What these tools are good at (creative tasks, summarization, learning, research) and where they fall short (math, reasoning, current events).

Key Concepts Explained

Hallucinations: Why LLMs confidently give wrong answers—they're built to predict, not to know
Knowledge cutoff: These models are frozen in time; they don't learn from your conversations
Context windows: How the conversation builds up and why that matters
Tools and agents: How LLMs can now search the web and take actions
Semantic space: Words mapped to thousands of dimensions—patterns, not facts

The challenge at the end: Try it this week. Go to ChatGPT, Claude, or Gemini. Be specific with your prompts. Verify what it tells you. See what it can do for your own learning and work.

No Prerequisites

This lecture assumes no technical background. We start from the basics and build up understanding together, using analogies and clear explanations rather than math or jargon.

Back to series overview