I recently co-authored a research paper on The Effect of Sampling Temperature on Problem Solving in Large Language Models.
It discusses how changes to the sampling temperature of an LLM from 0.0 to 1.0 have no statistically significant effect on its performance in MCQA problem-solving tasks.
A pre-print version of the paper is now available on arXiv at: https://arxiv.org/abs/2402.05201