The co-founder and chief executive officer of Google Deepmind, Demis Hassabis, gives a conference at the mobile World Congress (MWC), the largest annual gathering of the telecommunications industry in Barcelona on February 26, 2024.
Pau Barrena | AFP | Getty images
Paris – The Deepseek AI model “is probably the best job” in China, Demis Hassabis, the CEO of Google Deepmind, said on Sunday, but added that the company had shown no new scientific advances.
Last month, Deepseek de China published a research document that rocked the world markets after affirming that his AI model had been trained in a fraction of the cost of the main players of the AI and on Nvidia fleas.
Deepseek’s announcement sparked a Aggressive stock sale And sparked a considerable debate on the question of whether large technological companies spend too much in AI infrastructure.
Hassabis praised the Deepseek model as “an impressive work”.
“I think this is probably the best job I have seen out of China,” said Hassabis during an event animated by Google in Paris before the AI Summit AI which is hosted by the city.
Deepmind CEO said that the AI model shows that Deepseek can do “a very good engineering” and that it “changes things on a geopolitical scale”.
However, from a technological point of view, Hassabis said it was not a big change.
“Despite media threshing, there is no new real scientific advance … It uses known techniques [in AI]”He said, adding that the media threw around Deepseek was” exaggerated a little “.
Deepmind CEO said the company Gemini 2.0 flash modelsThat Google this week has published to everyone, is more effective than the Deepmind model.
Deepseek’s claims around its low cost and the tokens he uses Questioned by expertsWho think that the cost of developing the models of the Chinese company is higher.
Acted at five
The world of AI has been debating for years when the arrival of general artificial intelligence, or acted, will occur. AGE makes large reference to AI which is smarter than humans.
Hassabis said the AI industry was “on the Way of Act”, which he describes as “a system that presents all the cognitive capacities that humans have”.
“I think we are close now, you know, maybe we are only, you know, maybe 5 years or something of a system like the one that would be quite extraordinary,” said Hassabis.
“And I think that society must prepare for this and what implications that will have. And, you know, make sure that we take advantage of it and all of society take advantage of it, but we also attenuate some of the risks, too.”
Hassabis’ comments reflect those of others in the industry who suggested that AGI could be closer to reality.
Openai CEO Sam Altman said this year that he was “confident that we know how to build act as we traditionally understood”.
However, many in the industry have also reported several risks associated with AC. One of the biggest concerns is that humans will lose control of the systems they have created, a vision shared by eminent AI scientists Max Tegmark and Yoshua Bengio, who recently shared their concerns With CNBC on this form of AI.