Interesting Tic-Tac-Toe Observation

I was persuing YouTube as I do 8 hours a day and came across the latest drop by Mark Rober. Where he showcases the well known tic tac toe layout that results in a win (unless theres your opponent plays the center).

  X |   |
-------------
    |   |
-------------
  X |   | X

And remembered an interesting layout I had managed to find when passing time at my internship last November. Which often beats people who rely on that technique, and adds a little more spice to a ‘one and done’ game

Background

Back in November of 2025, I had decided to pass the time programming what I liked instead of going insane sitting alone in the room designated for Interns. So I came across the Hartnell Exploring AI Book, and decided to dive into my muse subject of computer learning (no matter what you say, llm is not it fam).

There it showed off a simple Tic-Tac-Toe program, which implemented a simple reinforcement learning algorithm. I picked that up and programmed my own version using pygame and ofc python. Data was stored in json files to keep memory, and the gui was drawn using stackoverflow and an llm, suppose my refusal to learn pygame played a part.

Finding

I had left it running for about 2 weeks straight, after which id switch it up by teaming novice X against a trained O and vice versa.

What came out of that machine was able to beat a human player about 1/8 times.

It had been able to find a method that worked against the ‘unbeatable’ method that did not involve always going center!

Big Reveal

This design is what will beat normal users, and more often than not, the ‘power users’. Since it lets you control the center without giving away the trick.

    |   |
-------------
    | X |
-------------
  X |   | X

Like the known layout, its not necessary to create the pattern in this direction, one may use it on the side :)