Now, here's the second thing I did, which is similar-but-different. It's called the 8 queens problem. The rules are simple: place 8 queens on a chess board such that none of them can attack each other. Simple to say, hard to execute. There are 92 solutions to this problem, and so I wrote a script to test if THRML/TSU would get stuck on a single solution. Here's the trick: there are billions of possible board states, but the 92 solutions all have the same energy level: 0. So how would a TSU handle that? In reality, there are rarely going to be multiple solutions that are TOTALLY equal in loss/energy. So, I fired up this script that baked in the rules (with no starting conditions) and watched as the TSU quickly iterated through, discovering the solutions. It took over 400 iterations to get all 92 solutions due to the massive amount of duplicates it generated. But it did ultimately generate 92 unique solutions. There are a few things that are remarkable about this whole process: FIRST, ChatGPT has never seen the THRML library. Yet it was able to read the docs, look at the examples, and code it up ONE SHOT. The 8 queens puzzle required no refinements or fixes. It just... worked... SECOND, since the 8 queens is a known problem space with a known solution set, it was an ideal benchmark to test THRML against (maybe @beffjezos will send me a TSU!) it will be interesting to see how the real hardware fares. But considering it makes use of natural noise, I figure it will probably perform similarly. THIRD, there's probably something mathematical and interesting going on around the fact that it took about 400 iterations to discover all 92 solutions. But I am not smart enough to figure that out. FOURTH, it will be far more interesting, I think, to point TSUs at totally novel problems where we do NOT have a complete solution set

Next, I wanted to riff on where I think this is going. The first, and I think most obvious direction, is protein folding and protein design. With proteins, you're trying to find the lowest global energy state of a complex molecule. You can brute force this with old methods like FOLDING@HOME and you can also train ginormous models like AlphaFold to use neural networks to approximate the shapes (albeit very accurately and quickly). OR You can use thermodynamic computing to rapidly design/discover the optimal shape for any given protein. Not only that, it can be done on the SAME hardware that solved sudoku and the 8 queens problem. And that, I think, underscores the key difference. The TSU is a general purpose accelerator. It's not onlike how a TPU or GPU accelerates a particular kind of math. The TSU accelerates other kinds of math, and it's all based on noise. True measurements of entropy. Not pseudo-random number generators. AND because the way the pbits interact, you can approximate a true Boltzmann distribution (imagine billiard balls ricocheting off each other, but without the simulation) To put this into another metaphor, this is like solving sudoku by simply gluing some 9 sided dice in place, and shaking the board with a bunch of other 9 sided dice until they settle into the correct values. Even doing this would be a chance of like one in billions or trillions of working. But because of the physics of the TSU, the correct configuration is the default result. It's still blowing my mind.