> The first version is basically telling the CPU: "Hey, I'm going to read this memory in order, you can prefetch the next cache line." While the second is more like: "Surprise! We're jumping to a random memory location! Again! And again!"

> It's amazing how many performance problems boil down to "stop making the CPU cache sad" 😆​​​​​​​​​​​​​​​​

See? The AI gets it