A new study offers a simple pre-prompt that coaxes AI models to reveal multiple possible answers instead of just one—restoring the diversity lost through alignment training.

November 13, 2025

A new study offers a simple pre-prompt that coaxes AI models to reveal multiple possible answers instead of just one—restoring the diversity lost through alignment training.





