Review the sentence files and assess which ones can be shortened while keeping the same values and direct commands

[santanavagner]

Description (Actual Behavior)

According to a recent Unesco report [1], longer prompts are more energy demanding. Hence, we must maintain our sentence recommendations as short as possible, while keeping embedding social values and short, objective commands.

[1] Smarter, smaller, stronger: resource-efficient generative Al & the future of digital transformation

Expected Behavior

Provide recommendations of short sentences embedding social values and direct commands.

Possible Approach

Review our sentences file to review longer sentences and try to keep shorter recommendations. We also could have a metric of average words in our recommendations, so we can always consider this metric when adding new sentences to the file.

Steps to Reproduce

NA.

Context

Recommendation.

[santanavagner]

Adding here an important information, in our most recent user study [2], participants added, on average, 1.50 of the sentences recommended. Given that the positive sentences in our dataset have, on average, 13.44 words, this means that our recommendations augmented the input prompt on 20.16 words. I believe we should continue considering this aspect on future studies. Thoughts?

[2] Responsible Prompting Recommendation: Fostering Responsible AI Practices in Prompting-Time