Transmuting prompts into weights

Hanna Mazzawi, Benoit Dherin, Michael Munn +3

A growing body of research has demonstrated that the behavior of large language models can be effectively controlled at inference time by directly modifying their internal states,…