From Sports Stats to AI Safety: The Ranking Renaissance

In 1952, two statisticians proposed a simple model for comparing sports teams. In 1960, it became the foundation of chess ratings. Today, that same math powers everything from ChatGPT's alignment to Chatbot Arena leaderboards.

But you don't need to be OpenAI to use it.

I'll show how Bradley-Terry and preference-based ranking can solve everyday problems: comparing A/B test variants, ranking search results, evaluating ML models, prioritizing features, and more.