The big US labs are naively hoping that compute power will always be their moat. I strongly suspect they are dead wrong, and newer architectures will require va

The big US labs are naively hoping that compute power will always be their moat. I strongly suspect they are dead wrong, and newer architectures will require vastly less training data and outperform the brute force approach of US labs.