Jaap Arriens | Nurphoto via Getty Images
Large linguistic models like the ones he developed Microsoft-Open-based company should become comfortable this year due to rapid progress towards the agent of artificial intelligence of the new generation and faster open code rivals, according to top technical executives.
Last week Chinese company AI Deepseek posted R1, The open code reasoning model that claims that OPENAI OPENAI Model O1 both at the expense and on performance. Open-Source refers to software whose original code is freely available on the open web for possible changes and redistribution.
This week, growing awareness of the new Deepseek model led to strong downs in stocks Nvidia Other technological giants, as investors were afraid of a possible reduction in consumption on powerful graphic units for processing needed to train and start advanced AI loads.
Nvidia Lost close to $ 600 billion in market capitalization On Monday-Finding a one-day fall for any company in American history.
In the meantime, executives and scientists in leading AI laboratories all talk about moving from large language models to so -called “AI agents” that can conduct actions on your behalf.
LLMS are the basic technology behind today’s generative AI apps. However, experts believe that they will push towards agency AI systems – which include LLM technology – this year will disrupt the value of these models.
Commodity LLMS
LLM is expected to become more goods in the near future because technology is becoming more advanced, and the costs involved in training and their leading continue to fall.
Thomas Wolf, co -founder and chief scientific director of Hugging Face, suggested that LLMS integrate into intelligent systems associated with his own databases.
“I think people are moving out of this craziness around the model, realizing that, thanks to the open car … Many of these models will be free and freely available,” Wolf told CNBC. Facial hug is a popular warehouse for open code AI projects.
“In the Internet Revolution, we are moving from construction websites as the main business to the creation of the companies ejected by the Internet, Airbnb from AI, the AI railway,” he added. “It’s not a model. They relate to the system and how to make a model useful for tasks.”
The arrival of the Deepseek model R1 last week some technological executives view as and Further indicators of LLM are becoming more comfortable.
R1 includes some key aspects that differentiate it from other models on the market. For example, the model includes the “mixed precision” frame that uses a combination of 32-bit 32-bit points (FP32) full precision number (FP32) and numbers with 8-bit point (FP8) low precision.
The latter is faster to process, but it can be less precise. However, instead of relies on one or the other, the Deepseek uses the FP8 for most budgets and exceeds the FP32 for certain tasks where a higher degree of accuracy is required.
“As Ai becomes more effective and affordable, we will see that its use is increasing, turning it into goods that we simply cannot get enough,” ” Microsoft CEO of Satya Nadella wrote On the Social Media Platform X, in an obvious reference to Deepseek.
Meanwhile, Matt Calkins, the CEO of US S
“In my opinion, we will see the comfort of AI. Many companies will achieve competitive AI, and the lack of differentiation will be bad for the first movements of the big consumption,” Calkins said by email.
What are agents AI?
In the last two years, much of the focus has been on LLM products, from generative AI chatbot such as Openai’s Chatt to tools that can produce pictures and video content.
Generative applications based on LLM, such as Chatgpt, rely on a certain amount of user posture to continuously seek them for outputs, while, in contrast, the agents are designed to be oriented to the action.
If, for example, you wanted to book a doctor’s meeting, you can ask Chatgpt for information about the closest practices of GP – much like you would ask the web using Google’s search engine.
In the future, technologically executors predict that the next generation AI will include agents who can find out where your closest general doctor is, provide available time for your doctor and reserve a meeting – all directly inside one platform. It is expected to transform the way we communicate with the web.
“During the decade since the internet invented, the standard website interface and the menu has really not changed much,” said Bobby Blumofa, the CNBC CEO of Akamai.
“We somehow live in this tyrant that we are tied to the screen and related to websites and menus that you have to remember how to move.”
Transition from a model to agents
Last week, Arthur Mensch, the Mistral CEO, said he thought “focus should be switched to systems” that integrate both linguistic models and contextual business data.
“This is the place where it changes,” Mensch told CNBC Arjun Kharpal at the annual gathering of the World Economic Forum in Davos, Switzerland.
“It also means that the industry that accepts it will distil its expertise in these systems – and this is where the value of AI -A will actually push into these industries.”
Mistral, who is supported MicrosoftHe quickly became one of the most interesting AI companies in Europe. Founded only two years ago, whether The last investor estimate of about $ 6 billion In June 2024.
The trend of agents has already gained significant attraction. Last week, Openai started his own agent, called the operator. Openai describes this as “an agent who can go to the web to do tasks for you,” adding that he is dressed for interaction with “buttons, menus and textual fields that people use daily.”
Microsoft, who has invested billions of dollars in Openi, I took out AI agents in November.
Anthropically, an Amazon-The AI startup that founded former research executives, last year Introducing “Computer Using”, Which is the ability that allows your AI agents to use computers like people to complete complex tasks.
Dario AModei, ANTHROPIC Executive Director, said last week Andrew Ross Ross Ross Ross Ross Ross Ross Ross Ross Ross Ross’s in Davos that the use of a computer should be considered as a predecessor of the “wider vision” that integrates these agents with the most commonly used applications.
“This year we plan on 2025 to build something we call” a virtual associate, “said Amodei, describing him as” an agent acting on your computer. “
“You can give him tasks. You can tell him, write this feature for this application, and she will write a code, will put together a code, will check the code, talk to his associates on Slacka, or on Google documents or on another platform,” explained the boss of Anthropic.
Anthropically, he plans to gradually build parts of this vision over the next year – and at the same time, ensuring that he is “safe and controlled,” AModei added.
Paul O’Sullivan, Salesforce CEO for UK and Ireland, provides that verbal communication will become more important in the new AI world. Salesforce was probably early at the agent AI game, after launching his own agent system before Microsoft, Openi or Anthropic in September.
“We’ll talk more,” he told CNBC in a recent interview. “The screens will decrease, our eyes will get worse. But they are likely to replace them with mobile devices that are more powerful than laptops.”
