#llama-3

Steering Emotional Responses in Large Language Models

Researchers have developed a method to steer the emotional output of the Llama-3-8B model by manipulating activation axes across transformer layers.