Chinese startup DeepSeek AI and its open-source language designs took over the news cycle this week. Besides being comparable to models like Anthropic's Claude and OpenAI's o1, the models have raised several concerns about data privacy, security, and Chinese-government-enforced censorship within their training.


AI search platform Perplexity and AI assistant You.com have actually found a way around that, albeit with some restrictions.


Also: I checked DeepSeek's R1 and V3 coding skills - and we're not all doomed (yet)


On Monday, Perplexity published on X that it now hosts DeepSeek R1. The totally free strategy gives users 3 Pro-level queries per day, which you might use with R1, but you'll require the $20 per month Pro plan to access it more than that.


DeepSeek R1 is now offered on Perplexity to support deep web research study. There's a brand-new Pro Search reasoning mode selector, along with OpenAI o1, with transparent chain of believed into model's thinking. We're increasing the number of day-to-day uses for both free and paid as include more ... pic.twitter.com/KIJWpPPJVN


In another post, the company validated that it hosts DeepSeek "in US/EU data centers - your information never ever leaves Western servers," assuring users that their information would be safe if using the open-source models on Perplexity.


"None of your information goes to China," Perplexity CEO Aravind Srinivas repeated in a LinkedIn post.


Also: Apple researchers expose the secret sauce behind DeepSeek AI


DeepSeek's AI assistant, powered by both its V3 and R1 models, is available through browser or app-- but those need communication with the company's China-based servers, which creates a security danger. Users who download R1 and run it locally on their devices will avoid that issue, however still encounter censorship of certain subjects identified by the Chinese federal government, as it's integrated in by default.


As part of providing R1, Perplexity declared it eliminated at least some of the censorship built into the model. Srinivas published a screenshot on X of query results that acknowledge the president of Taiwan.


However, when I asked R1 about Tiananmen Square using Perplexity, the design declined to respond to.


When I asked R1 if it is trained not to respond to specific questions figured out by the Chinese government, it reacted that it's created to "concentrate on factual details" and "avoid political commentary," which its training "stresses neutrality in global affairs" and "cultural level of sensitivity."


"We have actually removed the censorship weights on the design, so it should not act this way," said a Perplexity representative responding to ZDNET's demand for remark, adding that they were looking into the issue.


Also: What to know about DeepSeek AI, from expense claims to data personal privacy


You.com uses both V3 and R1, similarly just through its Pro tier, which is $15 each month (marked down from the normal $20) and without any totally free inquiries. In addition to access to all the designs You.com provides, the Pro plan comes with file uploads of up to 25MB per question, a 64k maximum context window, and access to research study and custom representatives.


Bryan McCann, You.com cofounder and CTO, described in an email to ZDNET that users can access R1 and V3 via the platform in 3 methods, all of which utilize "an unmodified, open source version of the DeepSeek models hosted entirely within the United States to ensure user personal privacy."


"The first, default way is to utilize these designs within the context of our exclusive trust layer. This gives the models access to public web sources, a predisposition towards mentioning those sources, and a disposition to appreciate those sources while creating responses," McCann continued. "The second way is for users to turn off access to public web sources within their source controls or by utilizing the designs as part of Custom Agents. This option enables users to explore the models' unique capabilities and behavior when not grounded in the general public web. The third way is for users to test the limitations of these models as part of a Custom Agent by adding their own instructions, files, and sources."


Also: The very best open-source AI models: All your free-to-use options discussed


McCann kept in mind that You.com compared DeepSeek designs' reactions based upon whether it had access to web sources. "We observed that the models' actions varied on a number of political topics, in some cases declining to answer on certain issues when public web sources were not consisted of," he discusses.