I wonder if a CoT LLM would perform better if we set the temperature high during reasoning and then immediately lower it after the reasoning is finished, right before the LLM gives us the final result.

Next - Previous