The co -founder and Chief Executive Officer from Google Deepmind, Demis Hassabis, offers during the mobile world congress (MWC), the largest annual meeting of the telecommunications industry, on February 26, 2024 in Barcelona in Barcelona.
Barrena Pau | AFP | Getty pictures
Paris – Deepseek's Ki model “is probably the best work” from China, Demis Hassabis, said Google Deepmind CEO on Sunday, but added that the company did not show any new scientific progress.
Last month, Chinas Deepseek published a research paper woman on which the global markets folded after claiming that his AI model was a fraction of the costs for leading AI Nvidia Chips.
The announcement of Deepseek triggered an aggressive sales sale and triggered a considerable debate about whether large technology companies spend too much for the AI infrastructure.
Hassabis praised Deepseek's model as “an impressive piece of work”.
“I think it's probably the best work I saw from China,” said Hassabis at a Google eventual event in Paris in front of the AI-Action summit, which is organized by the city.
The Deepmind -CEO said that the AI model shows that Deepseek can do “extremely good engineering” and that it “changes things at a geopolitical level”.
From a technological point of view, however, Hassabis said it was not a big change.
“Despite the hype, there is no actual new scientific progress … it uses known techniques [in AI]”He said and added that the hype around Deepseek was” a little exaggerated “.
The Deepmind -CEO said that the Gemini 2.0 flash models of the company that Google published this week are more efficient than the deepmind model.
Deepseek's claims about its low costs and the chips they use were questioned by experts who consider the development costs for the models of the Chinese company to be higher.
Agi five years away
The AI world has been discussing for years when the arrival of artificial general intelligence or AGI will take place. Agi generally refers to AI, which is smarter than humans.
Hassabis said that the AI industry was “on the way towards Agi”, which he describes as “a system that shows all cognitive skills that people have”.
“I think we are close now, you know, maybe we are only 5 years or something that is away from a system like that, which would be quite extraordinary,” said Hassabis.
“And I think.”
Hassabis' comments reflect on those of others in the industry who have proposed that Agi could be closer to reality.
Sam Altman, CEO of Openaai, said this year that he is “confident that we know how we can build AGI, how we traditionally understood it”.
Nevertheless, many in the industry have also marked several risks associated with AGI in connection with AGI. One of the greatest concern is that people lose control of the systems they created, a view of prominent AI scientists Max Tegmark and Yoshua Bengio, which recently shared their concerns with CNBC about this form of AI.
Comments are closed.