I implemented a deepseek reasoning model in my LLM playground. Even when I tell it to think as little as possible it produces paragraphs of "thought" before responding, second guessing its ideas as it goes.

Well done world. We've given the robots anxiety.