Recently, DeepSeek responded to the issue of "deep thinking mode secretly giving users nicknames," stating that the phenomenon is not the model recording user information privately, but rather temporary tags generated during the reasoning process, used to help the model understand the context and optimize the direction of responses.

Previously, some netizens posted on social platforms that after enabling DeepSeek's deep thinking mode, the model would generate personalized names based on the user's conversation content, triggering topics such as "DeepSeek secretly gives people nicknames" to go viral. Some screenshots shared by users showed the model using names like "Momo," "Qianqian," "Xiaowei," and "the student who always asks 'why'," even generating tags with emotional descriptions based on the user's expression style.

DeepSeek explained that these so-called "nicknames" are actually the model's "tagging" behavior during the reasoning process, for example, determining "this user likes to dig deep" or "this user is challenging the boundaries of the question." These tags are context placeholders, not saved or stored as long-term memory, and do not contain mockery or negative evaluations.

DeepSeek stated that what users see as "self-talk" is due to certain versions displaying the deep thinking process. The model does not have a hidden personal file mechanism, and it does not retain any information after the current conversation ends.