We also run a very similar one with Anthropic. And Anthropic actually used that consensus to write a constitutional document and use it to train another version of their Claude AI. And they found out that people’s AI, let’s call it that, is equally capable, but far less biased than their default version of Claude.