If an LLM is inherently an aggregation of voices and data generation processes, «we» is exactly correct. On capabilities side, wouldn't surprise me if it helps with diversity. Speculative further advances: dynamic in-CoT role allocation, may capture some multi-agent benefits.