Did DeepSeek Use Google Gemini to Train Its Latest AI Model?
Discover why experts suspect DeepSeek’s new AI model was trained on Google Gemini data and what it means for AI development.
Matilda
Did DeepSeek Use Google Gemini to Train Its Latest AI Model?
DeepSeek’s recent release of its updated R1-0528 AI reasoning model has sparked widespread curiosity and debate among AI enthusiasts and developers. Many are asking: Did DeepSeek use Google’s Gemini AI outputs to train this new model? How does this affect the competitive AI landscape? The updated R1-0528 model excels at solving complex math and coding problems, but DeepSeek has kept its training data sources under wraps. However, speculation is mounting that DeepSeek’s team leveraged Google Gemini’s powerful AI outputs as part of its training process. This raises important questions about AI data usage, model training ethics, and the impact on innovation. Several AI researchers, including Melbourne-based developer Sam Paech, have pointed to strong linguistic similarities between DeepSeek’s R1-0528 and Google’s Gemini 2.5 Pro outputs. Paech noted that the choice of words and expressions in DeepSeek’s model “echoes” those favored by Gemini, suggesting a possible link. Similarly, the crea…
