This works because a problem could be broken down to a prompt which rarely hallucinates. Most real world prompts can't be reduced to something so consistent an

This works because a problem could be broken down to a prompt which rarely hallucinates. Most real world prompts can't be reduced to something so consistent and reliable. Their key finding was that the number of votes grows linearly with number of prompts you are trying to chain. However the issue is that the number of votes you need will grow exponentially with hallucination rate.