"The temperature parameter in ChatGPT completions is a crucial aspect that influences the model's output diversity and creativity. Essentially, it determines how the model selects tokens (words or word pieces) when generating text.
A higher temperature value (e.g., 1.0 or above) results in more randomness and creativity in the generated text, as the model assigns relatively equal probabilities to multiple tokens. This may lead to diverse and unexpected responses but can also compromise coherence and relevance.
On the other hand, a lower temperature value (e.g., 0.1 or below) makes the model more conservative and deterministic in token selection, as it strongly favors tokens with the highest probabilities. This usually results in more focused, consistent, and contextually accurate responses but may lack creativity or variety.
Thus, the temperature parameter allows users to fine-tune ChatGPT completions to obtain the desired balance between creativity and coherence, depending on the specific application or context." - ChatGPT 4
See e.g. https://gptforwork.com/guides/openai-gpt3-temperature