Stargate can lead to Amnesty International models smaller and faster

When Openai, The White House, Oracle, Softbank and MGX announced that it will invest in the Stargate project – a new company to lead the development of artificial intelligence in the United States – questions about the effects of infrastructure facilities of the data center, energy and obstetrics, Amnesty International, Organization Amnesty International Development, followed immediately.

“Like a space race with Soviets in the 1960s,” said Cuba Stollraski, Vice President of IDC Research and Global Research Pioneer to lead the infrastructure of the providers and service providers in a new report.

“America is now in the Amnesty International Race with the Chinese for the rest of 2020 by starting the largest project for Amnesty International in history,” he said, as the country’s gross domestic product will be affected by the goods and services of artificial intelligence sold. The report, Stargate Project launches a new era for the development of artificial intelligence, after that advertisement On January 21.

However, new large language models such as Chat GPT 3.0 were more than five years. It was challenged by the capacity of the GPU memory specified in the rate and the level of accounts required for LLM an exerciseIt has been proven costly. Consequently, the race to achieve efficiency in the types of artificial intelligence that achieves or transcends human cognitive capabilities means that the large Amnesty International is likely to adjust its course, according to Stolarski.

“Arms Race” for GPUS

Through the preliminary efforts to build the ambitious infrastructure of Stargate data – $ 500 billion in servers market investments over the next four years – already in Abelin, Texas, Openai said that the first $ 100 billion will secure the American leadership in artificial intelligence.

However, “development on large language models will require sufficient commercial use to justify investments” in the largest project of Amnesty International’s infrastructure in history, according to Stolarski.

Besides LLMS, competing server techniques are used through a variety of cases of use, as he indicated in a report.

He emphasized that although the Stargate scale was not identified from the infrastructure point of view, the demand for GPU from Stargate may put a large strain on the graphics processing unit, which just began to mitigate during the year 2024 after 2023 ”

“Although the offer will continue to increase every year, we have noticed that the demand is still outperforming the supply in the very hot market.”

Also, while President Donald Trump’s involvement “can generally indicate the removal of barriers” needed to implement a project on this scale, Stolki said there are many uncertainty -related policies.

He wrote: “The interesting development is how some of the project elements that may be necessary for its long -term success, such as clean energy, will apparently conflict with the direction of the new administration.”

Finally, it is also possible that GPU farms are useful investments, and it is not clear if it is possible to achieve artificial general intelligence, or AGI, “with the power of raw computing and current LLM methods.”

“With comparative financial support, quantum computing may provide, for example, much more revolutionary breakthroughs than LLMS,” said Stolarski.

Dibsic and narrower models

In the face of the expansion of the LLM approach as a wise path to work, the Chinese company Deepseek claimed that it is able to train the AI ​​model similar to a decrease 11 times in GPU.

“I think the thing that Deepseek hit from the nerve is this claim that they can do this in a small part of the cost, but not from apples with apples, because when the basic models in the United States are actually used to build Deepseek, Stolski said. Health care news In a follow -up conversation last week, the company’s foreign announcement was followed.

“The improvement levels that we will get may continue to improve,” he said.

Openai participates with Tech giants such as ARM, Microsoft, NVIDIA and Oracle on the development of Stargate technology. Oracle promised to move to resolving the challenges of health care data before, as is the case with her aspiration for a national health record database and the recent cyber security initiative.

“There are all these ways they are trying to work on in the background at the end of the background to try to make these things more efficient,” Stollurcke said.

The cost should decrease the spread of artificial intelligence through cases of use – along with the model scale.

“The smaller models that are caught more and allocated to specific cases of use will agree,” said in Stargate’s position.

“These small linguistic models (unlike large language models) do not require huge infrastructure environments. Sporadic models, tight models and low accuracy models – a lot of research is currently being conducted to reduce infrastructure needs to develop the artificial intelligence model in a significantly while keeping accuracy rates .

For example, Stolarski explained that IBM is building smaller models to generate the revenues of its customers.

If we can get more efficient [fewer GPUs]Then the demand will increase. “

“I think everyone who follows this market expects to get these efficiency gains, which is a timing issue, and I think so far that the knee reaction seems to be declining.”

Andrea Fox is the first editor of the Healthcare IT News.
Email: Afox@himss.org
Healthcare is news is Himss Media Publication.