DeepSeek-V3 calls itself ChatGPT? The truth or points to "AI pollution"

On December 30th, "DeepSeek-V3 surpassed all open-source models to date." This is the conclusion reached after testing DeepSeek-V3 by Artificial Analysis, a foreign independent evaluation agency. More importantly, the model used NVIDIA H800 GPU to train DeepSeek-V3 in just two months, at a cost of only about 5.58 million US dollars. However, during use, the reporter found that DeepSeek-V3 actually claimed to be ChatGPT. For a time, there were doubts about "whether DeepSeek-V3 is training with ChatGPT output". TechCrunch speculated that the deep search may have used a public dataset containing the text generated by GPT-4 through ChatGPT. " If DeepSeek-V3 was trained on this data, the model may have memorized some of the outputs of GPT-4 and is now regurgitating them verbatim. "However, DeepSeek-V3 is not the first model to misidentify itself, and Google's Gemini and others sometimes claim to be competing models. Gemini, for example, claims to be Baidu's ERNIE Bot chatbot at a Mandarin prompt. The reason for this may be that AI companies obtain a lot of training data on the Internet, but today's Internet is already flooded with all kinds of data produced by AI.