Google AI head also believes DeepSeek relied on Western models for learning
February 12, 20255:06 pm is0 comments
Google DeepMind boss Demis Hassabis said that China’s DeepSeek AI didn’t spend less than US$6 million to develop the AI system. He says the claim is exaggerated and “a little bit misleading.” Hassabis Run Google’s ai Division.””https://www.bnnbloomberg.ca/business/technology/2025/02/10/google-ai-chief-says-deepseeks-cost-claims-are-exaggerated/””>Bloomberg claims DeepSeek’s claim of spending less than $6 million is misleading, and that the company “seems only to have reported the final training round which is a small fraction of the total costs.”
According to reports, DeepSeek spent approximately US$5.6 (roughly C$8,000,000) on computing costs to train its model to work with older Nvidia processors. The U.S. authorities want to know if DeepSeek circumvented the ban on chips by purchasing new chips via Singapore.
OpenAI and Microsoft also investigate whether a group linked to DeepSeek distills OpenAI. Hassabis concurs that distillation is a method of transferring knowledge from a large language to a smaller model. He toldBloomberg,that DeepSeek “seems to have relied on Western models for distillation”without giving more details.
Alphabet on the other had reportedly planned to invest US$75 billion in its cloud computing division, which will improve services such as Google’s AI model Gemini and Search, among other products. Source: BNN Bloomberg.
MobileSyrup could earn a commission on purchases made through our links. This helps fund the free journalism we provide on our website. These links have no influence on our editorial content. Support us here
