The key aspect in AI problem solving is the ability of the system to break down its problem-solving approach in human understandable terms, which enables it to guide itself through the process and come up with correct answers. When the system lays out its reasoning step by step, it performs better in reaching correct conclusions, as it seems to build on top of its reasoning to produce the final answer. The actual text written in the reasoning flow might not significantly impact the performance. It is suggested that the system's success in reaching conclusions is more related to the amount of inference and computing power it can invest in the process, rather than the words it generates.