Describe the challenges of fine-tuning a Large Language Model on a domain-specific dataset without overfitting.