Ask HN: How Does DeepSeek "Thinks"?

There is a useful feature in DeepSeek that isn't present in other commercial LLMs. It displays its internal "thinking" process. I wonder what technological aspect makes this possible. Do several LLMs communicate with each other before providing a solution? Are there different roles within these LLMs, such as some proposing solutions, others contradicting or offering alternative viewpoints, or reminding of overlooked aspects?