Perplexity lets you try DeepSeek R1 without the security risk, but it’s still censored

Chinese language startup DeepSeek AI and its open-source language fashions took over the information cycle this week. Moreover being corresponding to fashions like Anthropic’s Claude and OpenAI’s o1, the fashions have raised a number of considerations about information privateness, safety, and Chinese language-government-enforced censorship inside their coaching. 

AI search platform Perplexity and AI assistant You.com have discovered a means round that, albeit with some limitations. 

On Monday, Perplexity posted on X that it now hosts DeepSeek R1. The free plan provides customers three Professional-level queries per day, which you might use with R1, however you may want the $20 per thirty days Professional plan to entry it greater than that. 

In one other publish, the corporate confirmed that it hosts DeepSeek “in US/EU information facilities – your information by no means leaves Western servers,” assuring customers that their information could be protected if utilizing the open-source fashions on Perplexity. 

“None of your information goes to China,” Perplexity CEO Aravind Srinivas reiterated in a LinkedIn publish. 

DeepSeek’s AI assistant, powered by each its V3 and R1 fashions, is accessible through browser or app — however these require communication with the corporate’s China-based servers, which creates a safety danger. Customers who obtain R1 and run it regionally on their gadgets will keep away from that concern, however nonetheless run into censorship of sure subjects decided by the Chinese language authorities, because it’s inbuilt by default. 

As a part of providing R1, Perplexity claimed it eliminated not less than a few of the censorship constructed into the mannequin. Srinivas posted a screenshot on X of question outcomes that acknowledge the president of Taiwan. 

Nevertheless, once I requested R1 about Tiananmen Sq. utilizing Perplexity, the mannequin refused to reply.

After I requested R1 whether it is educated to not reply sure questions decided by the Chinese language authorities, it responded that it is designed to “deal with factual info” and “keep away from political commentary,” and that its coaching “emphasizes neutrality in world affairs” and “cultural sensitivity.”

“We now have eliminated the censorship weights on the mannequin, so it should not behave this fashion,” mentioned a Perplexity consultant responding to ZDNET’s request for remark, including that they had been trying into the difficulty. 

You.com gives each V3 and R1, equally solely by means of its Professional tier, which is $15 per thirty days (discounted from the standard $20) and with none free queries. Along with entry to all of the fashions You.com gives, the Professional plan comes with file uploads of as much as 25MB per question, a 64k most context window, and entry to analysis and customized brokers. 

Bryan McCann, You.com cofounder and CTO, defined in an e mail to ZDNET that customers can entry R1 and V3 through the platform in 3 ways, all of which use “an unmodified, open supply model of the DeepSeek fashions hosted completely inside the USA to make sure consumer privateness.” 

“The primary, default means is to make use of these fashions throughout the context of our proprietary belief layer. This provides the fashions entry to public net sources, a bias in direction of citing these sources, and an inclination to respect these sources whereas producing responses,” McCann continued. “The second means is for customers to show off entry to public net sources inside their supply controls or by utilizing the fashions as a part of Customized Brokers. This selection permits customers to discover the fashions’ distinctive capabilities and habits when not grounded within the public net. The third means is for customers to check the bounds of those fashions as a part of a Customized Agent by including their very own directions, recordsdata, and sources.”

McCann famous that You.com in contrast DeepSeek fashions’ responses primarily based on whether or not it had entry to net sources. “We seen that the fashions’ responses differed on a number of political subjects, generally refusing to reply on sure points when public net sources weren’t included,” he explains. “When our belief layer was enabled, encouraging quotation of public net sources, the fashions’ responses revered these sources, seemingly overriding prior political biases.”