I thought we could have the model play a little game where it first generates a puzzle to solve, then we ask another instance of the model to try to solve the problem that it generated.

So, yeah, I'll ask it, generate, say, a 5x5 nonogram, where the final answer is the letter M.

We'll see what it comes up with.

All right, we see it just gave us a little puzzle.

We'll go ahead and copy this, open up another window of O1, ask it, solve the following puzzle, and let's say, visualize the answer in some pretty way, why not?

This puzzle doesn't look like it's too hard, but the way that a nonogram works is for each list of numbers, and the numbers tell you how many squares are filled in, and if the squares are consecutive, then you'll see a 2.

For two consecutive squares, if there's a space between them, you'll see a 1, 1.

And so you're supposed to just try to figure out which squares do I indeed have to fill in, and it looks like the model got this right.

It illustrated a nice little letter M.

But I think one of the things that is nice about examples like this is it's similar to Sudoku or a crossword, for example, where you have to make a guess and then see if that's a right guess or a wrong guess, and then backtrack if you get it wrong.

And so any type of task where you have to search through a space where you have different pieces pointing in different directions, but there are mutual dependencies, so you might get a bit of information that tells you that these two pieces contradict each other.

A model like O1 is really good at trying to refine the search space here.