Does algorithm prompting help LRM accuracy?

 title: 'Figure 6: Accuracy and thinking tokens vs. problem complexity for reasoning models across puzzle environments. As complexity increases, reasoning models initially spend more tokens while accuracy declines gradually, until a critical point where reasoning collapses—performance drops sharply and reasoning effort decreases.'

The text indicates that algorithm prompting does not lead to improved performance in Large Reasoning Models (LRMs). Even when provided with a complete algorithm for solving the Tower of Hanoi puzzle, models did not show improved performance, as their accuracy collapsed at similar complexity points. This suggests that their limitations lie not just in problem-solving and solution strategy discovery, but also in consistent logical verification and execution of steps throughout their reasoning processes[1].

The findings highlight a fundamental challenge: LRM performance does not significantly benefit from algorithm prompts, as they fail to leverage explicit guidance effectively[1].