Comment by kaba0

The implementation details don't matter. LLMs not being able to properly reason though is a fundamental limitation and no amount of re-running will help.