DATABRICKS HAS A TRICK THAT LETS AI MODELS IMPROVE THEMSELVES

Summary from the TLDR AI newsletter

Databricks has developed Test-time Adaptive Optimization (TAO), a method that leverages reinforcement learning and synthetic data to improve AI models without the need for clean labeled data.