KARLA Trains LLMs to Emit Special Tokens That Trigger Knowledge Base Queries

KARLA trains models to generate special tokens mid-generation that trigger live knowledge base lookups, enabling factual updates via KB edits rather than retraining. The approach improves factual grounding in short and long-form generation and lets smaller models match larger models on factual accuracy.