摘要：“I think DeepSeek has had some real innovations,” Microsoft Chairman and CEO Satya Nadella told analysts when asked about the tren

TMTPOST -- Microsoft Corporation proactively embrace DeepSeek while cautioning Chinese artificial intelligence (AI) upstart could be involved in intellectual property (IP) theft case.

Credit：China Daily

“I think DeepSeek has had some real innovations,” Microsoft Chairman and CEO Satya Nadella told analysts when asked about the trend of AI scaling at lower costs. Nadella believes the development of AI is no different than the regular compute cycle, and DeepSeek is good for big tech company like Microsoft since it can help reduce AI cost, thus unlock more demand and expand deployment of the technology.

Based on Moore's Law and the AI scaling laws, all the software optimizations on inference result in 10 times on improvements per cycle, and the big beneficiaries of any software cycle that is the customers, Nadella said at an earnings call on Wednesday. “When token prices fall, inference computing prices fall, that means people can consume more and there'll be more apps written,” said the chief executive. Accordingly, the type of optimizations of powerful models like DeepSeek “means AI will be much more ubiquitous, and so therefore for a hyperscaler like us, a PC platform provider like us, this is all good news as far as I'm concerned,” said Nadella.

Nadella at the conference said DeepSeek's R1 reasoning model launched Wednesday via the model catalog on Azure AI Foundry and GitHub. In a blog post on its website, Microsoft announced DeepSeek R1 is available in the model catalog on Azure AI Foundry and GitHub, and has undergone rigorous red teaming and safety evaluations, including automated assessments of model behavior and extensive security reviews to mitigate potential risks. Customers will soon be able to use distilled flavors of the DeepSeek R1 model to run locally on their Copilot+ PCs, Microsoft said.

Bloomberg earlier Wednesday reported OpenAI along with its partner and largest shareholder Microsoft are probing if DeepSeek-linked group obtained data output from OpenAI’s technology in an unauthorized manner.

It was reported security researchers working for Microsoft found in the fall of 2024 that DeepSeek may have exfiltrated a large amount of data using OpenAI’s application programming interface (API). Microsoft then notified OpenAI of the suspicious activity. Such activity could violate OpenAI’s terms of service or could indicate the group acted to get around OpenAI’s restrictions on how much data they could obtain, the report cited people familiar with the matter.

Microsoft didn’t respond to the report. OpenAI later confirmed it is investigating into DeepSeek for the possible violation of its terms of use.

OpenAI is investigating whether DeepSeek improperly improperly obtained data from its model to launch a super popular AI assistant, a spokesperson told The Hill. The person referenced a machine learning technique called distillation, which allows developers to enable a smaller model to deliver similar performance on a specific task by leverage outputs of a larger one.

Distillation does not expose a model’s inner workings and can be used by developers to improve their applications, the spokesperson noted. But OpenAI prohibits from using “output to develop models that compete with” it, according to its terms of use. The company also doesn’t allow anyone’s automatical or programmatical extraction of data or output.

“We know that groups in the PRC [People’s Republic of China] are actively working to use methods, including what’s known as distillation, to try to replicate advanced U.S. AI models," the OpenAI spokesperson said. "We are aware of and reviewing indications that DeepSeek may have inappropriately distilled our models, and will share information as we know more."

In an interview with Fox News Tuesday, the White House AI and crypto “czar” David Sacks raised concerns about DeepSeek, noting “substantial evidence” that DeepSeek relied on the output of OpenAI’s models to help develop its own technology.

Asked about whether DeepSeek stole intellectual property from the U.S., Sacks said it is "possible." For Sacks, DeepSeek took advantage of the distillation process that allows student AI models interrogate parent models, mimic their logic, and "suck" their knowledge from them.

来源：钛媒体