Into the Unknown: Self-Learning Large Language Models