"...found that the increasing response length phenomenon is not due to the emergence of self-reflection, but a consequence of RL optimizing well-designed rule-b

"...found that the increasing response length phenomenon is not due to the emergence of self-reflection, but a consequence of RL optimizing well-designed rule-based reward functions." What is the difference?