A bomb -like news fell on the agenda. According to Reuters, OpenAI pushed the button to get rid of his dependence on Nvidia. The company is moving towards removing its own artificial intelligence chip. Let’s see what’s going on together without further ado.

According to Reuters’ quotes OpenAIDesigned by the artificial intelligence chip can be mass production next year. Sources said the company has completed the company’s chip design within a few months and plans to send it to TSMC for production. Thus, the OpenAI will use its own chips instead of Nvidia to train and run artificial intelligence models. While the production of TSMC will be produced by 3 NM process, the chip will be equipped with high -band width memory and comprehensive networks.

The OpenAI chip will use it on a limited scale during the launch. Sources, however, said that the chipset will mostly be used to run artificial intelligence models. The future versions of the chip will have more advanced processors and abilities.

OpenAI’s chip design team is the head of Google’s former TPU engineer Richard Ho. It is stated that the work has accelerated thoroughly in the last months and the team has increased from 20 people to 40 people. Last year, Reuters shared a report claiming that OpenAI was working with Broadcom to develop a special chip. Again, in another report of a few months ago, it was mentioned that the company gave interesting job advertisements and turned to chip design. When we combine the information mentioned in the old reports, we see that the puzzle was completed before.

Firms operating in the field of artificial intelligence made billions of dollars of chip purchases to create artificial intelligence infrastructure and strengthen hungry models. But initiatives such as Deepseek opened up to discuss whether to get thousands of chips to equal and provide power.

Google has announced today that Gemini’s application contains the newest Gemini 2.0 experimental models.

One of the most remarkable innovations for users is that the Gemini 2.0 Flash Thinking model can now be tried free of charge by everyone. The company states that this model is particularly trained in a way that divides the commands into steps to strengthen their reasoning capabilities and provide better responses.

According to Google, Gemini 2.0 Flash Thinking, users’ why the model reacts in a certain way, Which It allows them to see how it is based on assumptions and how it creates the line of logic ”. Of course, these explanations do not mean that artificial intelligence really “think” or “reason”. However, it can be said that it is an important step in terms of providing transparency to understand how the models are concluded. In this way, when the Geminin sometimes provides false or fabricated information, it will be a little easier to understand why it is caused.

In addition, there is a great innovation for Gemini Advanced subscribers: Gemini 2.0 Pro is now accessible. Google says that this model has been developed especially in complex tasks, to ensure that it gives more accurate results in coding and mathematics. The expression “better reality ında in the company’s statement seems to actually mean ür producing less fabricated information ..

Both models are currently available on Gemini mobile application and Gemini web interface. Google also announced that Gemini 2.0 Pro will soon be available to Google Workspace Business and Enterprise customers. These developments stand out as an important part of Google’s artificial intelligence strategy, which aims to make Gemini more functional for both individual users and businesses.

Behind Deepseek’s success technological infrastructure And Financial strategies It causes controversy in the USA.

In a report published on Friday, it was claimed that the company has purchased advanced Nvidia chips by overcoming the US export restrictions on artificial intelligence chips. It is claimed that the company supplied these chips through third -party intermediaries in Singapore.

It is known that Deepseek uses much less resources compared to billions of dollars that giants such as OpenAI and Google spend to improve chatgpt and Gemini models. This suggests US analysts about whether China is ahead of the USA in development. However, Deepseek’s success is not the same in every field. In a test between 11 major artificial intelligence platforms, Deepseek ranked 10th in 10th by answering only 17 %of the questions correctly. Moreover:

• 30 %of his answers contained false claims,
• 53 %of them provided uncertain or useful information about the news.

This shows that the model is not yet to compete directly with Google Gemini or OpenAI chatgpt.

In -depth review and investigations from the USA

The US administration is investigating whether this new company has used intermediaries in Southeast Asia to overcome restrictions on artificial intelligence chips. FBI and the White House examined whether Singapore had a role in providing Nvidia chips to Deepseek.

In particular, the percentage of income of Nvidia from Singapore increased from 9 %to 22 %in the last two years. attracts. This increases speculation that Singapore has become an important center in the supply of chips to Chinese AI companies. Nvidia said in a statement that his partners had to act in accordance with the law and said he did not believe that Deepseek violated the laws. On the other hand, Howard Lutnick, elected by Trump to the Ministry of Commerce, said that Deepseek is not against the competing with US AI companies, but that this should not be done using the US’s own tools.

The company admitted that it uses 2,048 Nvidia H800 GPU chips to train the V3 model. However, it is estimated that the stronger -R1 is trained using advanced Nvidia GPUs, which are prohibited in sale in China. All these developments show that the US can bring tighter controls to China in artificial intelligence technologies and global AI competition will be further escalated.