Trivially, because any two of rarely produce the same "next token"! An ensemble of LLMs trained identically would generate the same next token(s) forever. But

Trivially, because any two of rarely produce the same "next token"! An ensemble of LLMs trained identically would generate the same next token(s) forever. But we don't - we generate different sequences. We are not LLMs.