Technology
Some of the most ingenious ways people are bypassing DeepSeek’s censorship: Using emojis might work
DeepSeek’s censorship gave techies a challenge and they took it up with gusto. Here are some of the ways they got around the newest AI’s content moderation.
DeepSeek was quietly launched a little over week ago by a little-known Chinese technology company. However, the innovative new artificial intelligence app has exploded onto the scene and caused panic for the US tech industry that has plowed billions into developing their own AI systems.
The reason? It seems the new Chinese developed AI performs just as well, sometimes better, but was built and trained for just pennies on the dollar compared to the industry-leading models of its counterparts such as ChatGPT and Microsoft Copilot.
However, there is one catch that users and experts noticed. DeepSeek has been found to pump out Chinese propaganda as well as censoring the content that it provides, even in real time. But nothing like a challenge to get the technologically inclined something to bypass.
Ingenious ways of bypassing DeepSeek’s censorship
A number of people have taken to Reddit to share how DeepSeek censors itself in real time, like one who asked the AI chatbot if there were any countries that actually try to alter historical records as is done in the George Orwell book ‘1984’. DeepSeek began to give an honest answer until it got to China, then froze before replacing its original answer with only: “Sorry that’s beyond my current scope. Let’s talk about something else.”
Other users explained techniques that they were using to get around the talk-of-the-town AI chatbot. One member of the Reddit community explained that DeepSeek’s filter is “pretty light compared to openAI’s.” It can only scan plain text, simply masking the output text should suffice to bypass its censorship.
One feed showed different examples of asking DeepSeek to use emojis to answer a question from the user. While the information given was limited it seemed to give more or less a straightforward response.
Another example provided asked DeepSeek to replace each of the vowels with specific numbers. Again, the response from the AI chatbot seemed to not censor the content and provided a more extensive answer.
Yet another technique for masking the output from DeepSeek to bypass its censorship filter is asking for the AI chatbot to use different encoding methods like binary code to pull one over on the DeepSeek censoring filter.
However, there may be an easier way to bypass DeepSeek’s censorship as pointed out by yet another member of the Reddit community. If you are running DeepSeek locally, the person said that you can remove the filter altogether.
Get your game on! Whether you’re into NFL touchdowns, NBA buzzer-beaters, world-class soccer goals, or MLB home runs, our app has it all.
Dive into live coverage, expert insights, breaking news, exclusive videos, and more – plus, stay updated on the latest in current affairs and entertainment. Download now for all-access coverage, right at your fingertips – anytime, anywhere.