Jack Clark, former OpenAI and current Anthropic co-founder, mentioned bittensor's subnet covenant in a newsletter discussing distributed AI training. Fundamentally, distributed learning is a political technology that will transform the politics of frontier computing. Today, the AI frontier is essentially defined by five companies, and in the next few years, perhaps ten, who can commit enough computing resources to train a competitive model in six months. Currently, these companies are all American, and with the recent relaxation of export controls on Chinese companies, they may eventually include Chinese companies. However, frontier learning is not being driven by academia, government, independent organizations, or non-technical industry actors. Distributed learning offers a way for these groups and other stakeholders to pool their computing resources to shift this dynamic, making its development crucial to monitor. Even if distributed learning doesn't quite reach the frontier, its impact will grow as the gap narrows. "Distributed learning can still be a very important part of AI. As long as distributed networks continue to be linked with open weights, it can lead to the emergence of larger open models that follow the frontier." importai.substack.com/p/import...