Google DeepMind CEO praises DeepSeek but says Gemini 2.0 models better than Chinese chatbot

Google

Google DeepMind CEO Demis Hassabis has praised China’s DeepSeek AI model but says it does not bring new scientific advances. He also claimed that Google’s Gemini 2.0 models are more efficient.

Hassabis made these comments at a Google event in Paris before the AI Action Summit. He called DeepSeek’s model “probably the best work” from China and said it shows strong engineering skills. However, he added that the model does not introduce new AI techniques.

“There’s no actual new scientific advance … it’s using known techniques,” Hassabis said. He also said that DeepSeek is overhyped.

DeepSeek gained attention after it claimed to train its AI model at a much lower cost than other AI companies. The company also said it used less advanced Nvidia chips. These claims led to a drop in stock prices and raised concerns about the high costs of AI infrastructure.

Despite DeepSeek’s bold claims, experts question whether its costs were really as low as the company suggested.

AGI may be five years away

Hassabis also spoke about artificial general intelligence (AGI). AGI refers to AI that is as smart as or smarter than humans. He believes that AGI could arrive within five years.

“I think we’re close now, maybe only five years away,” he said. “Society needs to get ready for that and make sure we benefit while reducing risks.”

Other AI leaders, including OpenAI CEO Sam Altman, have also said AGI could be coming soon. However, some experts worry that humans may lose control of powerful AI systems. Scientists Max Tegmark and Yoshua Bengio have warned about the risks of AGI.