Perplexity AI Launches Self-Hosted DeepSeek R1 Model Amid Concerns Over Data Privacy

On Monday, Perplexity AI, which is backed by Jeff Bezos, announced the release of a self-hosted version of its DeepSeek R1 model. This move is expected to support deep web research for users.

What Happened: Perplexity AI's CEO, Aravind Srinivas, took to X, previously known as Twitter, to share the exciting news. He stated that the DeepSeek model is now available for research purposes.

According to Srinivas, DeepSeek R1 is claimed to be the world's most powerful reasoning model due to its advanced capabilities that aid daily research tasks. Users can look forward to increased limits throughout the day, along with periodic updates.

However, the announcement has sparked numerous misconceptions, with some users expressing fears that the data processed by DeepSeek might be accessible to the Chinese government or that the search results could be subject to censorship.

In response to these concerns, Srinivas reassured users, saying, "All DeepSeek usage in Perplexity is through models hosted in data centers in the USA and Europe. DeepSeek is *open-source*. None of your data goes to China." This statement aims to clarify the location and management of user data.

To further alleviate concerns, Srinivas shared a screenshot of an unfiltered response generated by the DeepSeek R1 model, demonstrating that the model remains uncensored.

Why It Matters: The privacy policy of DeepSeek has previously been scrutinized, with critics like Ross Gerber voicing distrust regarding the handling of data, especially given the current tensions surrounding AI developments in China. Additionally, the White House has placed DeepSeek under review for national security reasons.

During an interview, David Sacks, who is recognized as a key figure in White House discussions on AI and cryptocurrency, mentioned that there is significant evidence suggesting that DeepSeek has extracted knowledge from OpenAI's models. This implication is of concern to OpenAI and its leadership.

Srinivas further elaborated on the issue by stating, "There's a lot of misconception that China 'just cloned' the outputs of OpenAI. This is far from true and reflects incomplete understanding of how these models are trained in the first place." He emphasized that DeepSeek R1 has made advancements in Reinforcement Learning (RL) fine-tuning, showcasing its unique contributions to AI technology.

Reportedly, DeepSeek has been recognized for its capabilities and is said to surpass other significant AI models such as OpenAI's ChatGPT. Accessing DeepSeek R1 via the API starts at $0.14 per million tokens, which is approximately 750,000 words. In contrast, OpenAI's comparable model, referred to as "o1," charges $7.50 per million tokens.

Overall, this launch is significant in the competitive landscape of AI, particularly in addressing the current concerns surrounding data privacy and performance standards.

Perplexity, DeepSeek, AI