The start -up in China Deepseek made An attitude of he This week creating a model that is believed to be at the same time with the main models from the beginnings of the US – with a piece of cost. In one research Released last month, Deepseek said he developed her under $ 6 million in just two months, a cry from the 100 million dollars we need in the US to train – and that is at the bottom of spectrum, according to Anthropic Director General Dario Amodei.
She quickly got up in Top of App Store tableschallenging the US position as the world leader in him. Issuing began a race for his predominance and shocked great technology actions, causing Chipmaker he nvidia of Lose nearly $ 600 billion In market value one day and new competition claims – from having an even better model to charges of theft.
According to the White House he and Crypto Czar David Sacks, Deepseek's arrival shows that Chinese companies are “hot on our heels”, but that the US holds its leadership. He says that of Deepseek is at the same time as Openai's O1 model, who came out about four months ago.
“We basically have somewhere between a lead three and six months on them (Chinese companies),” Sack said. “But they're caught a lot. Very soon.”
Deepseek. Photographic illustration by Justin Sullivan/Getty Images
Chatgpt Creator Openai says Deepseek is copying it
Openai and Microsoft are investigating If Deepseek used large amounts of OpenAI training data without a permit for him. Openai told the Times Financial At the beginning of this week, there was evidence that Deepseek used its great models of it to create its own through a process called distillation, in which one model he teaches from another as a student learning from a teacher.
Sacks supported Openai's claims in an interview with Fox Business Tuesday.
“There is essential evidence that what Deepseek did here is that they distilled knowledge from Openai's models,” Sack said. “I think one of the things you will see over the next few months is our main companies to take steps to try and prevent distillation.”
Other industry leaders say Deepseek's success is due to the collaborative nature of the open source models.
Deepseek “came up with new ideas and built them at the top of other people's work”, the chief of the Meta scientists, Yann Lecun stated in a post of topics Saturday. “Because their work has been published and an open source, everyone can benefit from it.”
Alibaba claims there is a better model
The Chinese company E-commerce Alibaba is claiming to have developed an even smarter model than that of Deepseek.
Alibaba on Wednesday released a new model called Qwen 2.5 Max Edition that the company says he scored better than Meta, Openai and Deepseek in the main tests Bloomberg.
“QWEN 2.5-Max exceeds … almost throughout the table (Openai's) GPT-4o, Deepseek-V3 and (Meta's) Llama-3.1-405b,” stated Alibaba's Cloud Division in an official WeChat account , according to Reuters.