The thing that lets LLMs select the next token is probabilistic. This proposed a deterministic procedure Problem is, we sometimes want LLMs to be probabilistic

The thing that lets LLMs select the next token is probabilistic. This proposed a deterministic procedure Problem is, we sometimes want LLMs to be probabilistic. We want to be able to try again if the first answer was deemed unsuccessful