what the fuck! The head of DeepMind publicly criticized DeepSeek, angrily denouncing its claim of ultra-low training costs as completely false advertising! Last night, the CEO of DeepMind fired directly at the recently popular Chinese AI team DeepSeek during an interview. Although he admits that this is currently the strongest team that has emerged in China, he bluntly states that the ultra-low cost data widely promoted by the other party is seriously exaggerated and even highly misleading! He immediately peeled off the so-called low-cost underwear. He pointed out sharply that DeepSeek's publicly disclosed expenses were only the cost of the final formal training, completely concealing the exorbitant amount of real money invested in the massive trial and error, exploration, and low-level testing stages in the early stage. This account is extremely dishonest. What's even more ruthless is that he directly revealed that DeepSeek heavily relied on existing advanced Western models for distillation and fine-tuning during the development process, and these hidden costs of borrowing were not even taken into account. He firmly stated that this team has not invented any revolutionary new technology that can be fatal, only pieced together the existing technology stack in the industry and applied it quite well, which cannot be considered a miracle of breaking the efficiency curve. In the end, this AI giant even went straight to the point: if we really want to strictly compare the input-output ratio, their Gemini model actually completely outperforms DeepSeek in efficiency. It's just that as an established giant, they usually keep a low profile and disdain to boast about these data everywhere!