Join daily and weekly newsletters to obtain the latest updates and exclusive content to cover the leading artificial intelligence in the industry. Learn more
Last week, Chinese start starting Dibsic Shock waves were sent in the artificial intelligence community Open source version, Deepseek-R1. The form is used to learn pure reinforcement (RL) to match Openai’s O1 on a set of criteriaChallenge the old idea that only wide training with strong chips can lead to high performance.
However, with the quadruple version, many also began to think about the effects of the Chinese model, including the ability to transfer Deepseek to user data to China.
Fears began with the company privacy policy. Shortly after, the SnownBallLed case, with the technical employee Stephen Heidel, indirectly indicates that Americans like to “give up their data” of the Chinese Communist Party to get free things.

These allegations are important from a safety point of view, but the fact is that Deepseek can only store data on Chinese servers when using models through a company -like service similar to the company.
If an open source model is hosted locally or organized via graphics processing units in the United States, the data does not go to China.
Fears about Dibsic Privacy Policy
There privacy policyAnd that was also not available for a few hours, Deepseek notes that the company collects information in different ways, including when users register or use its services. This means everything starting with account preparation information – names, email messages, numbers and passwords – to use data such as calling text or audio and files that have been uploaded, comments and the broader chat record to the company.

But this is not all. The policy also states that the information collected in safe servers will be stored in the People’s Republic of China and can be shared with law enforcement agencies, public authorities and others for reasons such as helping government requests or requests.
The latter is important because data protection laws in China allow the government to seize data from any server in the country at the lowest pretext.
With such a set of information about Chinese servers, countless things can be turned on, including individuals and institutions, leakage of sensitive business data, and even electronic surveillance campaigns.
hunting
Although policy can easily raise safety and privacy warnings (as it already has for many), it is important to note that it only applies to private Deepseek services – applications, web sites and programs – using the R1 model in the cloud.
If you register on Deepseek chat on Deepseek or use Deepseek AI on Android or iOS, there is a good opportunity to send your device data, personal information and demands so far to China.
The company did not share its position on this issue, but given that the iOS Deepseek application was It goes to #1, even before ChatGPTIt is fair to say that many people may have already registered for the assistant to test his abilities – and share their data at the level of this process.
Android app for the service has been recorded more than a million downloads.
Deepsek-R1 open source
As for the Core Deepseek-R1 model, there is no doubt about data transfer.
R1 is fully open source, which means that the teams can operate locally for the targeted state of use through open source implementation tools such as Boys. This ensures that the model is effectively doing its work while maintaining the data settled on the device itself. according to Mustak eraThe former founder and CEO of AI, the R1-Distill-Swen-32B can run smoothly on new Mac devices with 16 GB of VRAM.
Alternatively, the teams can also use the GPU sets of third-party orchestics for training and control the form and publish them-without data transfer risk. One of these is Excess laboratoriesAllowing users to rent the graphics processing unit to host the R1. The company also allows inferiority through the guaranteed application programming interface.

However, if one looks only to chat with Deepseek-R1 to solve the specific thinking problem, the best way to go now is with Confusion. The company has just added R1 to the model, allowing users to conduct deep online research with a series of thinking.
According to Aravind Srinivas, CEO of Perplexity, the company enabled this state of use to its customers by hosting the model in the data center servers in the United States and Europe.

Long short story: Your data is safe as long as it will go to a local host version of Deepseek-R1, whether it is on your device or GPU somewhere in the West.
https://venturebeat.com/wp-content/uploads/2022/02/GettyImages-1076658494.jpg?w=1024?w=1200&strip=all
Source link