Self consistency - involves classifying and selecting the best answer of the multiple answers for a given prompt.
Given the undeterministic-nature of LLMs, this can be quite useful.
To generate multiple responses we can use the n parameter in the GPT API
However, generating mutliple variants affects token consumption and increases costs.