Wouldn’t the same outcome be achieved much more simply by giving LLMs a two choices (colors, numbers, whatever), asking “pick one” and assessing the results in the same way?