I think AI is the single greatest thing humanity will ever build.
This isn't something I arrived at recently. I've been following this space for about four years. My wife and I talk about it almost every day. Where it's heading, what it means, what the world looks like in ten years. Everything before this, electricity, the internet, computing, those were stepping stones. Brilliant, world-changing stepping stones, but stepping stones.
I believed all of that for a while. But I never tried to actually understand how any of it worked.
Then I watched DeepMind's documentary on Demis Hassabis. There's a moment where they're talking about AlphaGo. A machine that taught itself to play a game humans have spent thousands of years mastering, and then played a move in Game 2 that no human would have made. Move 37. The commentators didn't know what to say. The professionals in the room didn't know what to say.
I didn't know what to say either. But I knew I wanted to understand how that was possible.
Here's the thing though. I'm a software engineer. I've been building web apps for well over a decade now. I use AI every single day. To write code, to think through problems, to learn things faster. I'm not a stranger to it.
But the internals? How models actually learn? How an agent figures out that one move is better than another? I always assumed that stuff required a serious academic background to even get started.
In my head, AI was math. Linear algebra, calculus, statistics. Layers and layers of theory before you could even touch anything real. The greatest thing humanity will ever build was happening right in front of me, and I was on the sidelines. I figured at some point I'd take a few years off and just go study it - just out of curiosity. It was always somewhere on the horizon.
Then one evening I asked Claude to build me a snake game and a reinforcement learning model that could learn to play it.
It took about a minute.
A minute. I'd been putting this off thinking it would take months of preparation, and an AI just handed me a working starting point in sixty seconds.
And it worked. Surprisingly well, actually. After just a few hundred generations, the snake was navigating the board and eating apples. I sat there watching it thinking, that's it? That's what I was so intimidated by?
Then it plateaued. The snake got decent and stayed decent. And that's where the real learning started. I began tweaking things. The reward function, the way the snake perceived the board, the network architecture. With each change I understood a little more about why these things work the way they do.
Not because I read a textbook. Because something stopped improving and I had to figure out why.
That's what this blog is.
Back to the Training Ground is a series about learning AI by building things. Not by reading papers. Not by grinding through math. By picking a project, getting my hands dirty, and writing about what I learned along the way.
I use AI to write a lot of the code in this project. A snake game, a training loop, that's all scaffolding for what we're trying to learn. What matters is what building it teaches you about how AI actually works. How reward functions shape behavior. Why some features work and others don't. The stuff that actually helps you understand how these systems think.
If you're an engineer who uses AI every day but has no idea how it actually works under the hood, this might help you. If you've been telling yourself "someday I'll learn this properly," same.
The snake game is where it all began. Read the first episode →