Chinese artificial intelligence company that does not fall from the agenda these days Deepseek, Accusing in the United States and against an investigation in Italy across.
David Sacks, one of the White House official, to develop Chinese initiative Deepseek’s own artificial intelligence technologies OpenAI models (At chatgpt He claimed that there were strong evidence that he had benefited from the outputs used. Italy opened an investigation and demanded information about the data collected by Deepseek. The country, who wants to learn how this data is processed, is concerned that the personal information and other data of its citizens will go to China.
You may be interested in
Deepseek, Although they are under a US embargo on GPU Here He says that the “R1” model, which can be tested, gives better results than OpenAI’s “01” model. According to the company, the R1, AIME, MATH-500 and SWE-Bench leaves O1 behind the verified comparisons, and in many tests reveal similar performances. R1, which has a parameter of 671 billion, There are also special versions ranging from 1.5 billion parameters to 70 billion parameters. The full capacity version of the smallest model can be operated on a laptop, the full capacity of the R1 It is stated that it is much more “cheap ında than OpenAI in both education and full capacity..
You may be interested in
Meanwhile, Deepseek released the Janus Pro series image models on Github and Huging Face yesterday. At this stage, one and seven billion parameters, two different versions of Janus Pro, the big one, according to the company’s claim In some tests, Stable Diffusion and Dall-E can leave behind. It is not known exactly which resolution of models that seem capable of both visual analysis and visual production from the articles (some examples at 768 x 768 pixels). With the seven billion parameter Janus Pro model, realistic photographs can be produced, while the system is not yet perfect.