LLM for Everyone: Representing the Underrepresented in Large Language Models