https://anonymous.4open.science/api/repo/more_agent_is_all_you_need/file/img/algorithm_2.pdf?v=1c8ea7a90afd32108b7f754321bfdaa3a46edae4 https://anonymous.4open.

https://anonymous.4open.science/api/repo/more_agent_is_all_you_need/file/img/algorithm_2.pdf?v=1c8ea7a90afd32108b7f754321bfdaa3a46edae4 https://anonymous.4open.science/api/repo/more_agent_is_all_y... How can you have majority voting based on text produced by LLMs, won’t every answer string be essentially distinct with extremely high probability if it’s over a couple dozens bits of output (which any useful LLM output would be, right?)?