LakeHopper: Knowledge-Aware Adaptation of Column Type Annotators across Data Lakes
The episode discusses 'LakeHopper,' a method for adapting AI models to annotate column types across different data lakes. The process is an iterative, four-stage system that uses Large Language Models (LLMs) as diagnostic tools to identify knowledge gaps. This allows for highly efficient training by focusing only on the most critical 'weak samples' rather than retraining on massive datasets.