Internal State Control is a General Property of LLMs

Researchers have identified that large language models (LLMs) possess an inherent ability to control their internal states, a property that influences their output generation and behavior.