[matthewfelgate]

What's a reliable benchmark for measuring "AI Distortion" in these models, allowing for consistent tracking over time and potential improvement?