AlphaEvolve began optimizing the lowest levels of hardware powering our AI stacks. It proposed a circuit design so counterintuitive yet efficient that it was integrated directly into the silicon of our next-generation TPUs.
大多数人认为AI系统的硬件设计需要人类专家精心设计，但作者认为AI本身可以设计出比人类更高效的硬件电路。这挑战了传统硬件工程领域的共识，暗示AI可能在最底层的硬件设计上超越人类专家的直觉和经验。
By customizing and co-designing silicon with hardware, networking and software, including model architecture and application requirements, we can deliver dramatically more power efficiency and absolute performance.
通常认为硬件定制化是提高性能的途径，但作者强调通过软硬件协同设计可以大幅提升效率和性能，这与单纯硬件升级的观点相悖。
Here are two products that are basic rectangular boxes with a rounded edge (the one on the left also has some unpleasant drafted walls, but that’s another article about how to become a hardware design snob). Look at the beginning and end of that rounded edge on the main surface. See how there’s a sharp shift in highlight? That’s the result of tangency.