[6] It's now undeniable that, with a little bit of creativity, improving scaling is not only approachable but also particularly rewarding. And while I'm obviously excited by convolution-attention hyena hybrids due to their balance of efficiency and quality across domains, there's a lot more to do!
