“The capabilities of large, leading-edge AI models are being replicated in smaller, more efficient models within 6 to 12 months of their release.”