Perplexity AI has incorporated the Chinese DeepSeek AI model into its offerings for Pro level users, enabling them to utilize DeepSeek for research. This announcement faced public backlash, with concerns about potential data access by the Chinese government and censorship of search results.
Perplexity AI, based in USA, provides a unique AI search engine that utilizes web content and large language models. Its Pro Search version allows unlimited file uploads, image generation, and access to various AI models, including OpenAI's o1 and Anthropic’s Claude 3.5. DeepSeek R1 is now included as an option for Pro users.
Aravind Srinivas, Cofounder and CEO of Perplexity, addressed the controversy on LinkedIn, clarifying that DeepSeek operates through models hosted in the USA and Europe, is open-source, and does not send user data to China. He also reassured users on X (formerly Twitter) by sharing an uncensored response from DeepSeek R1 to demonstrate that the model is not censored.
While anyone can download the DeepSeek AI model for local use, it is reported to be censored, particularly regarding sensitive topics like the Tiananmen Square massacre. Testing by The Register confirmed that the model is designed to avoid politically sensitive discussions. However, the CEO emphasized that a self-hosted version does not communicate with external servers, ensuring data remains local.
