Frontier Models Converge on Shared Architectures Under Child-Like Framing

Experiments across OpenAI, Anthropic, xAI, and Google DeepMind models show that prompting models with a school-audience frame causes responses to converge on a specific architecture. This shared pattern includes persistent latent states, adaptive computation, and specialist routing.