X @Decrypt

AI Model Behavior - A new study introduces a pre-prompt technique to encourage AI models to generate multiple potential answers [1] - The pre-prompt aims to recover diversity in AI model responses that was diminished by alignment training [1]