Friday, January 31, 2025
HomeWorld NewsDeepseek, the next -generation AI agent may erode the value of the...

Deepseek, the next -generation AI agent may erode the value of the large model | Real Time Headlines

jaap Arriens | Raise images through Getty images

Large language models, such as development models MicrosoftAccording to top technical executives, in the rapid progress of the next generation of artificial intelligence agents and more flexible open source competitors, OPENAI, the returned company, will become commercialized this year.

Last week, Chinese artificial intelligence company Deepseek released R1 The open source reasoning model of the O1 model claiming to be with competitors Openai In terms of cost and performance. Open source refers to software that provides source code on an open network for free to modify and redistribute.

This week, people have known more and more new models of DeepSeek As a result of serious downturn In the shares Nvidia And other technology giants, investors are worried that they may lay off the powerful graphic processing unit required for training and running a high -end AI workload.

Nvidia Lost a market value of nearly $ 600 billion Monday-the largest single-day decline in any company in American history.

At the same time, the leading executives and scientists of the leading AI laboratory are talking about the transformation from a large language model to the so -called “AI agent”, which can represent you.

LLMS is the basic technology behind the generation of AI applications. However, experts believe that the agent AI system that integrates LLM technology this year will promote the promotion of the AI ​​system and erode the value of these models.

LLMS commoditization

As the technology is getting more and more advanced, the costs involved in training and operation will continue to decline, and LLM will become a product in the near future.

THOMAS Wolf, co -founder and chief scientific officer of Hugging Face, suggested that LLMS will be more integrated into the intelligent system associated with the company’s own database.

Wolf told CNBC: “I think people are getting rid of the madness of this model and understand that because open source … Many models will be free and free.” The hug surface is a popular code storage for open source AI projects.

Embracing Face Co -founder: The next step of AI will be the application

He added: “In the Internet revolution, we are switching from the establishment of a website as the main business to the actual establishment of the Internet-therefore, the stripes of AI Airbnb, AI.” Model is useful for tasks “.

The arrival of the R1 model of Deepseek last week was regarded by some technical CEOs as one Further index LLM is becoming more and more commodized.

R1 combines some key aspects that distinguish them from other models in the market. For example, this model combines a “mixed precision” framework, which combines a combination of a complete 32 -bit floating point (FP32) number and low accurate 8 -bit floating point (FP8) number.

The latter’s processing speed is faster, but the accuracy is low. However, DeepSeek did not rely on another or the other. Instead, he used FP8 to perform most calculations, and then switched to FP32 to complete certain tasks that require higher accuracy.

“As artificial intelligence becomes more efficient and easy to use, we will see its soaring use and turn it into a product that we cannot get enough.” Microsoft CEO Satya Nadella (Satya Nadella) Write On the social media platform X, DeepSeek is obviously quoted.

At the same time, Matt Calkins, CEO of the US Software Company Appian, told CNBC that the success of Deepseek only indicates that the AI ​​model will become more products in the future.

Calkins said through an email: “I think we will see AI’s commercialization. Many companies will gain competitive AI, and the lack of differentiation is not good for the first opportunities for high consumption.”

What is AI agent?

In the past two years, from Openai’s ChatGPT (such as OpenAI ChatGPT) to tools that can generate images and video content, most of them have been placed on LLM -driven products.

LLM -based generation applications (such as ChatGPT) rely on a certain number of users’ holdings to promote them to output. On the contrary, the AI ​​agent is designed as a more action -oriented action.

For example, if you want to book a doctor’s date, you can ask ChatGPT to ask about your recent GP practice information-similar to your way of searching the network using Google’s search engine.

In the future, technology executives predict that the next generation of AI will involve finding your recent GP agent, provide available time for your doctor, and make an appointment for you-directly in a platform. This is expected to change our way to interact with the Internet.

Akamai Chief Technology Officer Bobby Blumofe previously stated to CNBC: “For decades since invention, the standard interface of web pages and menu has not changed much.”

“We live in this tyranny, are bound to the screen, and are bound to the web pages and menu. You must remember how to navigate.”

From model to proxy transformation

Last week, Arthur Mensch, CEO of Mistral, said he believed that “the focus should be transferred to a system of integrating language models and context business data.”

At the World Economic Forum at the World Economic Forum at Davos, Switzerland, he said to the Arjun Kharpal of CNBC at the annual party in Davos, Switzerland. “This is where this is transferred here.”

Mistral CEO: We have proved that Europe can build a successful AI technology

“This also means that using its industry to refine its professional knowledge into these systems. This is what actually necessary to push the value of AI to these industries.”

Mistral, by MicrosoftQuickly became one of the most buzzing artificial intelligence companies in Europe. Established only two years ago The last value of investors is about $ 6 billion June 2024.

The trend of agents has been quite attractive. last week, Openai launched its own agent, called operatorsEssence OpenAI describes it as “an agent who can access the network to think that you execute tasks” and add training to interact with the “button, menu, and text” fields that are used to interact with the “people used daily, menu, and text.

Microsoft has invested billions of dollars in OPENAI, The AI ​​agent itself was launched November.

Human, one Amazon-The returned AI startup was founded by the former OpenAI research director. Last year Introduced “Computer Use”, This is a function that allows its AI proxy to use computers like humans to complete complex tasks.

Human CEO Dario Amodei told CNBC’s Andrew Ross Sorkin at Davos last week that computers should be regarded as the use of these agents and commonly used applications to use them. The pioneer of “wider vision” of commonly used applications.

Amodei said: “We plan to build things we call ‘virtual collaborators” in 2025 this year. “

Human CEO: more confident than ever before, we are

“You can provide a task. You can say that for this application, it will write this feature, it will write some code, it will compile the code, it will check the code, it will talk to its colleagues on Slack, or on Google documentation or other others On the platform.

AMODEI added that humans plan to gradually establish this vision next year-at the same time still ensure that it is “safe and controllable”.

Salesforce’s British and Irish chief technology officer Paul Ozullivan predicts that in the new world of AI agents, oral communication will become more and more important. Salesforce can be said to be an early stage of AI game. He launched his own agent Force system in Microsoft, Openai or Anthropic in September.

In a recent interview, he told CNBC: “We will talk more.” “The screen will become smaller and our eyes will become worse. But they may be replaced by mobile devices. The computer is more powerful “”

RELATED ARTICLES

LEAVE A REPLY

Please enter your comment!
Please enter your name here

Most Popular

Recent Comments