Which of the following approaches is best suited for optimizing prompts to ensure that a language model generates responses that are both concise and contextually relevant?

Increasing the model's attention span.

Incorporating keywords related to the desired context.

Iteratively refining the prompt based on generated outputs.

Increasing the number of training epochs.