Demis Hassabis, CEO of Google DeepMind, acknowledges that the AI model of DeepSeek is the best work he has seen from China, describing it as impressive and showcasing exceptional engineering prowess, which is crucial in geopolitics. However, he believes that DeepSeek’s model does not introduce new scientific advancements but relies on existing AI techniques, and the excitement surrounding DeepSeek’s model may be exaggerated.
In contrast, Gemini 2.0 Flash recently unveiled by Google is more efficient than DeepSeek. Additionally, Hassabis mentions that DeepSeek’s minimal training costs using NVIDIA chips may lead to misunderstandings, as it might only report the costs of the final training stage, excluding initial experiments and preliminary testing that contribute to overall development expenses.
TLDR: Demis Hassabis praises DeepSeek’s AI model from China but notes it lacks new scientific breakthroughs compared to Gemini 2.0 Flash by Google. He also points out potential misunderstandings in DeepSeek’s low training costs, which may not consider all development expenses.
Leave a Comment