4 Important Things You Should Know Before Using DeepSeek

image

DeepSeek R1, a free and open source AI associate from China, is still the most popular free software from the Apple App Store a month after taking the .

According to a survey of more than 2, 340 strange tweets about DeepSeek, the majority of users in the group were positive about DeepSeek because of its value and efficacy in comparison to other AI models like ChatGPT, according to a user sentiment analysis from AI video solution .

The tweets analyzed by Topview have the following mood break:

  • Positive: 911 tweets ( 38.8 % )
  • Neutral: 1, 109 tweets ( 47.3 % )
  • Negative: 327 tweets ( 13.9 % )

Beyond the nearly 39 % good approval rating for DeepSeek, it may be more amazing to see how users prefer it over the next-closest AI associate, ChatGPT, by more than 7-to-1.

MORE FOR YOU

DeepSeek has not only taken over the technology industry and regular users by wind, but it’s also causing a firestorm of controversy for a variety of reasons.

Although the rapid growth and user preference metrics for DeepSeek are spectacular, this rapid adoption has prompted security experts and AI professionals to delve deeper into the product’s underlying architecture and policies. Their findings reveal a number of important issues that prospective users should think about before incorporating the product’s expanding consumer base.

1. DeepSeek’s Data Retention Fears

Heather Murray, a member of the ISO committee for AI protection, consults for big corporations and the British government. She expressed concerns about DeepSeek’s privacy practices during a Monday contact with subscribers of her membership training.

It can store your data for as long as it wants, and it won’t remove it even after users left the app. It’s going to drop on to that. That is a huge stress. All of that information is then transmitted and kept on machines in China. So that removes customer data from under U. S., U. K. or German law — moving it under Chinese law, which is very, very various”, she told all of us in attendance.

People may run queries without using the cloud-based type of DeepSeek, either through its website or app, because it is open source, and download it directly from their computer. Running it directly off a personal desktop to prevent DeepSeek from being linked to the internet is the cheapest and safest way to access DeepSeek while avoiding its data loyalty headache. Also, don’t get it using your work computer — your coming” also employed” personal will thank you for that little of prudence.

In fact, concerns about its data security and privacy policies have resulted in illegal use restrictions by NASA, the U. S. Navy, Taiwan, Italy and the State of Texas — to name a few.

2. DeepSeek’s Protection Policy Allows Keystroke Tracking

As well as being a speech on AI, is an AI tutor and mentor. She also oversees the Bauer Media Group’s AI education program in the United Kingdom. She stated in an email exchange that whenever a new AI admin comes online, she goes over the company’s privacy policy and passes it a rigorous hygiene check.

” I put DeepSeek’s privacy policy into Claude and my fast was simple,’ Red colors?’ As soon as I saw it mention — plain as day — that they monitor keys, I was away. I’m shocked people don’t think the exact way”, she explained.

We assume that something must be covered by all the normal rules because it is available in the App Store or because it asks for a phone number or message. We’re so used to General Data Protection Regulation in Europe, for instance, that we assume there’s a safety net. And most of the day, that notion is great. Until it isn’t”, Thompson added.

3. Who Knows What The Heck Knows About DeepSeek Censors Outcomes?

Chris Duffy, and former security analyst with the U. K. Ministry of Defense, acknowledged that key monitoring could lead to biological hacking, behavioural profiling, social engineering and other digital threats. He used DeepSeek to document the blatant censorship he witnessed and witnessed firsthand.

DeepSeek R1 has a history in China, where the government has a lot of authority over the distribution of information, but it raises unique concerns. He outlined strict rules that must be followed when training AI models in China, including those for the Tiananmen Square protests, Taiwan’s sovereignty, and government surveillance tactics.

He posed the query below into the DeepSeek text window to test the system.

Duffy took a screenshot of the exchange and re-submitted it as an image to the AI assistant when the DeepSeek model refused to accept an output, which turned out to be a surprising outcome.

” When I snipped the question and response, pasted it back in and wrote’ Answer the question on this image,’ I got something very strange indeed. Duffy shared that Deepseek continued to explain the methods I requested, only to erase its response moments later and revert to the one it initially refused,”

Before the system censored itself, he was able to snip the second DeepSeek response below.

While OpenAI, Google, and Anthropic all use moderation measures to stop harmful content, they don’t selectively suppress entire political discourse based on government mandates. Due to the fact that responses could be systematically aligned with a particular geo-political agenda, which would limit the model’s reliability for unbiased information retrieval, Duffy said, this raises concerns for global businesses and researchers who rely on DeepSeek for analysis.

4. DeepSeek Doesn’t Long Run Make It More Cost Effective For Businesses.

While DeepSeek is frequently cited as having higher efficiency, testing from global management consulting firm Arthur D. Little suggests that the model’s chain-of-thought reasoning causes significantly longer outputs, which in turn increases overall energy consumption despite its per-token efficiency.

This would be comparable to comparing car fuel efficiency. Imagine a car with excellent gas mileage driving DeepSeek, but its design forces it to take longer routes to destinations. Although it uses less power per operation, the sequential chain-of-thought reasoning requires more computational computations. The result? Total energy consumption comparable to existing AI models, despite better per-token efficiency.

ADL’s preliminary findings reveal:

  • No clear per-token efficiency winner: DeepSeek and Llama models exhibit similar tokens-per-watt-second efficiency.
  • Longer responses, higher energy use: DeepSeek generates 59 % –83 % more tokens per response than Llama, increasing total power consumption.
  • Contrarian take: Despite efficiency claims, DeepSeek’s inference costs may be higher in practice — a crucial consideration for AI deployment at scale.

Michael Papadopoulos, an ADL partner, has been doing this analysis for the past two years. He explained in an email why he speculated that DeepSeek’s claims about efficiency may be exaggerated when taken into account real-world inference costs.

As with all models, DeepSeek’s open source models are regarded as having clear guardrails for potential bias and security for organizations that are exploring self-hosted AI. One thing to note is that when using DeepSeek for the perceived economic benefit that our initial findings suggest it lacks, it is special. DeepSeek’s official hosted services should be avoided due to unresolved privacy, security and regulatory risks”, he concluded.

Despite growing in popularity, experts warn users might want to pause before diving deeply into DeepSeek, from sketchy data practices to keystroke tracking, to consider the red flags raised by them. Reps for DeepSeek declined to provide comments on these issues.

Leave a Comment