What Is Reasoning

This blog post explores the concept of reasoning traces in AI systems, particularly focusing on closed-weight and open-weight models. It discusses how reasoning is often hidden from users, the mechanics of how models communicate their 'thinking' processes, and the implications of reasoning effort baked into model prompts. The author, Armin Ronacher, provides insights into the behavior of these models, how they phrase responses, and the potential for misleading users if reasoning channels are not properly managed.