Chinese artificial intelligence Startup Deepseek may have created waves rocking global technological actions on Monday, while American technological domination in the field has become questioned. Shock waves in Silicon Valley were not without reason.
The Chinese model, which is an alternative to Openai chatgpt and Google Gemini, unlike its American counterparts are fully based on open source technology and low -end chips. Developed at a cost of only $ 5.6 million, it avoids the need for high -end equipment limited by American export controls. In other words, Deepseek is available at a cost, which represents 2% of what users would spend on the O1 model of Openai.
However, despite these advantages, it is unlikely that it will extend its scope in India, as other Chinese technological products has done so in the past. Reason: trust factor. Since the border clashes of 2020 with China, the government has restricted Chinese technology because they do not fall under sources of category of confidence. Consequently, Indian companies developing applications would continue to depend on technologies of American AI and the processing units of graphics manufactured by Nvidia.
Technological policy experts said that, taking into account the history of prohibitions in the midst of security implications with Chinese applications such as Tiktok and equipment by companies such as Huawei and ZTE, Deepseek could face a similar examination in the country.
Officials have said that the Chinese generative AI model is being evaluated and that any decision -making decision would come in the Voluntary Ethics Code, which are underway.
“It is premature to talk about any type of safety implications with the use of open source models like Deepseek. However, companies seeking to take advantage of the same thing should make their reasonable diligence since there have been concerns in the past on Chinese applications, “said Dhruv Garg, partner of the Governance and Political Politics Project and Politics (IGAP) and a technological lawyer.
“The security threat of using models like Deepseek can be avoided if companies using these models deploy solutions on their locally servers,” said Ankush Sabharwal, founder and CEO of Corover and Bharatgpt.
Experts point out that there may be a transfer of personal data by users using these models as happens in the case of Chatgpt. This means that if national companies opt for the use of Deepseek, they should ensure that they do not use ready -to -use models but only APIs to create their own models with data hosted on their own servers.
Kamesh Shekar, senior dialogue program director, said: “Deepseek offers an affordable open-source alternative, in particular in the midst of tightened American export controls. However, its increase also highlights increasing dependence on Chinese models, raising critical questions about self-selfdependence and global digital sovereignty ”.
According to Shekar, the flow of cross -border data which is examined by the government with the deployment of rules to implement the data protection law, can pose restrictions on certain types of data to reach China or another country which would serve as braking in the use of Chinese technology.
Jameela Sahiba, senior program manager in dialogue, said: “This is also a call for action for developing countries like India to optimize its calculation and training resources, improve efficiency And promote innovation in the development of local models, meet regional needs with locally with locally locally. AI tailor -made solutions ”.
According to Sahiba, the launch of the Stargate project of $ 500 billion in the United States and the last order of control of exports on graphic processing units (GPU) open up the entire debate on the geopolitical implications of AI. For India, the progress of native AI capacities will not only strengthen its technological sovereignty, but will also allow it to engage more effectively in geopolitical dialogues on fair access to AI progress and infrastructure.
While India also plans to build its own calculation infrastructure and have major language models (LLM), these are the beginnings and proposals in this regard must still be finalized.