Could it be this simple—that this is always a sign that an AI has been trained on the output of another? So DeepSeek was trained on ChatGPT, and Kimi was trained on Claude, and Claude was trained in Chinese on DeepSeek?
Some LLMs trained on pre-2022 data and SFT’d with no identity/name in the SFT data will, when questioned, claim to be Siri or Alexa.
It only follows that models trained on more modern data will do similar things—and invert the “HHH” SFT data into latching onto any available “AI assistant” persona.
Could it be this simple—that this is always a sign that an AI has been trained on the output of another? So DeepSeek was trained on ChatGPT, and Kimi was trained on Claude, and Claude was trained in Chinese on DeepSeek?
Extremely unlikely.
Some LLMs trained on pre-2022 data and SFT’d with no identity/name in the SFT data will, when questioned, claim to be Siri or Alexa.
It only follows that models trained on more modern data will do similar things—and invert the “HHH” SFT data into latching onto any available “AI assistant” persona.