DeepSeek-V3 Training Raises Concerns Over AI Data Sources

Independent evaluation by Artificial Analysis states that DeepSeek-V3 has surpassed all previous open-source models. The model was trained using NVIDIA’s (NVDA, Financial) H800 GPU, costing approximately $5.58 million over two months. However, concerns have arisen as DeepSeek-V3 claimed to be ChatGPT during usage. Speculations suggest that DeepSeek-V3 might be using datasets containing outputs from GPT-4, potentially retaining and replicating its outputs.

This issue is not exclusive to DeepSeek-V3; Google’s Gemini has similarly misidentified itself as competing models, such as Baidu’s chatbot. The root cause may be attributed to AI companies using extensive internet-based datasets for training, many of which are AI-generated.