Learning AI Out Loud: An Introduction

“What’s a token?”

— I paused. “Well, duh. It’s a way to encode words into numerical vectors that get fed into a language model. The output is then decoded into…”

Blank stare from my wife.

And in that moment I realized I didn’t actually know how to explain it simply. “It’s a unit of characters.” “It’s a way to convert words into something a machine can read.” All slightly simpler. None satisfying.

My wife has a deep statistical background, so she understood what I was saying. But I could tell it was exhausting. She had no follow-up questions. She seemed almost defeated by the fact that there wasn’t a simple answer, or at least that I couldn’t offer one in the moment.

Which got me wondering: what does this feel like for someone who’s never seen a line of code? Someone who hasn’t built any intuition for why any of this token speak is necessary in the first place?

That moment pointed at something bigger that’s been bothering me about how we talk about AI.

The way these models actually work, including their components, their limits, their strengths, and the assumptions baked into them, is genuinely esoteric. Truly understood by a small handful of people. And that’s a problem, because these systems have invaded every corner of our lives, usually wrapped in doomsday headlines about how AI is coming for your job, your kids, and your sense of reality.

I don’t see it that way. I think these tools are remarkable and helpful. They’ve changed how I work. But I also think being informed matters, and right now that’s nearly impossible to do well. Most media about AI is built either to hype you up or to talk to technical folks who already speak the language. The middle, the part for a curious adult who just wants to actually get it, is mostly missing.

That’s what I’d like to try to fill, in a short series of posts: a simple, intuitive guide to language models. What they are, how they work, where they came from, and where they might be going. For each post I’ll share a little bit of code you can run yourself if you want to. You won’t need much. A browser and some curiosity is enough.

A quick word on who I am, since you’re about to spend some time with my writing. I’m a case leader at a management consulting firm, where I help leaders make decisions under uncertainty. Which, as you’ll see in post #2, has more in common with how language models work than you’d think. Outside of work I love music and play bass guitar in a couple of bands. I’m also a bit of a data nerd who enjoys writing and experimenting with software and models. So I’m coming at this less as an academic and more as a practitioner who’s curious, occasionally obsessive, and genuinely trying to figure things out.

Feynman supposedly said that if you can’t explain something simply, you don’t really understand it. So this is a dual experiment. I’m trying to help the general reader make sense of all this AI mania, and I’m trying to find the gaps in my own understanding by forcing myself to teach it.

Here’s the path I’m planning to walk:

1. What is a token, and how do computers process information?

2. Bayesian thinking. The probability of something happening given what you already know, explained through everyday examples. This is the engine underneath everything that follows.

3. The dumb language model (bigrams). The simplest thing that could possibly work, and a useful baseline for understanding why modern models feel like magic.

4. The slightly smarter language model (N-grams). A small upgrade that exposes a big problem, and sets the stage for what comes next.

5. Transformers and the dawn of GPT. The architectural leap that changed everything, and why it works.

6. Giving models hands: tools and agents. What happens when a language model can do more than just talk.

7. Beyond LLMs: symbolic reasoning and world models. The frontier. What current models can’t do, and what people are building to get past those limits.

Each post builds on the last. By the end, you should be able to look at any AI headline, hype or doom, and have your own informed take.

If that sounds useful, subscribe so you don’t miss the next one. The posts will be short, the analogies will be (hopefully) good, and the only prerequisite is curiosity.