And hundreds of language communities, each governing its own civic context, its own wiki and norms and talk pages, and interoperating — is exactly how the movement has worked for 25 years. The concentration — like mainframes, I guess, of our age, that you describe — is real. But also, frontier models cost billions to train. And smaller models, commercially speaking, are much easier to train together. We are already seeing, like, Thinking Machines Lab with the Tinker tool, Nemotron, many others, enabling local communities to decentralize the post-training, fine-tuning, steerability and so on. And they don’t cost a lot of electricity. And they don’t cost a lot of water for cooling. And that is, I think, a far more preferable way for communities to develop.