The company's Openai, located behind Chatgpt, proves that Chinese emerging companies, Deepseek, create a competing artificial intelligence model using that technology and promote concerns about the growing industry's intellectual property theft. I say I'm doing it.
Openai believes that DeepSeek, which was established by Math Whiz Liang Wenfeng, used a process called “distillation.”
This is common in AI development, but Openai states that Deepseek may have broken the rules by creating a unique AI system using a method.
“The problem is when someone shoots our technology and built a unique product using it,” he told Financial Times on Wednesday.
Americans have long been concerned about intellectual property theft from China, especially in industries such as artificial intelligence, biotechnology, and advanced manufacturing.
This fear is due to a repetitive claim that the entity supported by Chinese companies and states engages in cyber spy, the stolen of the corporate secret, and forced technology transfer, and gains competitiveness in the global market.
According to the report, the U.S. Navy officially prohibits the use of DeepSeek for fear of abusing confidential data.
According to FT, Openai does not share the details of evidence that it is allowed to support the claim.
However, the Terms of Use clearly states that the user is not allowed to copy the technology, or that it is not allowed to build an AI model that uses its output.
Despite being built on a much less budget, DeepSeek's recent AI model, called R1, has performed as well as the major AI models developed in the United States.
This surprised many people in the high -tech industry and questioned how Deepseek achieved such results.
With the advent of DeepSeek, a large -scale sale in Wall Street was promoted on Monday.
NVIDIA, the main supplier of AI hardware, has fallen by 17 % and has lost $ 589 billion.
Investors feared that companies like DeepSeek could no longer need expensive AI hardware.
However, NVIDIA's shares bounced off 9 % due to the recovery of high -tech stocks on Tuesday.
Last year, Openai and their partner Microsoft surveyed and blocked the account they suspect.
The sources told FT that these accounts used Openai tools in a way that could violate the rules.
Suspicion account survey It was first reported by Bloomberg News.
This post requires comments from Openai and Deepseek.
Recent high -tech leders have increased the possibility that Deep Shek has benefited from the theft of intellectual property.
David Sacks, an advisor to AI and cryptocurrency President Trump, suggested that DeepSeek might have stealed Openai technology.
“AI has a method of learning from another model by copying knowledge from a model. Sacks told Fox News.
But he did not provide specific evidence.
DeepSeek uses only 2,048 NVIDIA H800 graphic cards, spending $ 5.6 million, training AI models, and has 671 billion parameters.
This is much less than those that companies that like Openai and Google are spending on the same model.
Some AI experts say that DeepSeek models seem to be trained using Openai GPT-4 and violate Openai's service conditions.
Experts say that it is common for Chinese and US AI companies in the United States to improve their own systems using the results of established AI models.
“Many emerging companies and researchers are training their own models using answers from commercial AI models such as ChatGpt,” said Ritwik Gupta, a doctorate candidate at the University of California Berkeley. I talked.
“I don't need to collect human feedback myself, so I save time and money.”
Openai's challenge is to protect the technology while accessing AI.
The company states that it will actively defend intellectual property.
“We know that companies, including Chinese companies, are always trying to copy AI models,” Openai stated in the statement to FT.
Openai also stated that it is working closely with the US government to prevent enemies from stealing advanced AI technology.
Ironically, Openai itself faces a lawsuit that is allegedly using copyrighted materials without permission.
New York Times and some publishers have accused Openai and accusing them to train ChatGpt for their jobs without appropriate approval.
Openai and NEWS CORP last year are Openai Access to Current and Archive Contents, such as The Post, The Wall Street Journal, The Times of London. Multiple permit I attacked the year's contract.
Openai has established the same content partnership as other publishers, such as Condénast, Le Monde, and Prisa Media.
