DeepSeek is one of the new names you might have seen pop up more frequently in the past few months as an alternative to ChatGPT. Indeed, DeepSeek is an AI chatbot offering features similar to OpenAI’s ChatGPT. For example, DeepSeek’s latest models offer reasoning capabilities similar to ChatGPT o1. You might have seen the open-source model mentioned in reports and social media posts in late 2024 or earlier this year following the release of its big reasoning upgrades.
The more competition in the AI space, the better for the industry. Rivals will pressure each other to come out with newer, better models, that are also hopefully safer.
But if you want to try something other than ChatGPT, Gemini, Claude, or Meta AI, you should think twice before getting too deep with DeepSeek, which is created and managed by a Chinese company.
There’s no problem using software from China, and TikTok is an example of that. TikTok is also an example of what might happen with Chinese software in the US and Western world. Add the AI factor, and I would be cautious about using it for work or even for personal tasks.
DeepSeek-R1 is the company’s most recent AI model. DeepSeek says R1 is better than ChatGPT o1 at various tasks, including coding and math. Also, R1 might be cheaper than o1 and even other AI models from China. That could give it a competitive advantage over rivals and pressure them to cut costs to compete. But there are already controversies you should be aware of.
Why did DeepSeek think it was ChatGPT?
First, as of late December, DeepSeek used to identify itself as ChatGPT. DeepSeek V3 told users it was a version of OpenAI’s GPT-4 model. The reason for that lies in the training of DeepSeek. The AI might have been exposed to text generated with ChatGPT, so perhaps DeepSeek was simply mimicking all of that.
Remember that current genAI simply regurgitates what it has seen during its training. It predicts what words should follow next. Also important is that all genAI problems have the same problem: They hallucinate information.
If DeepSeek were exposed to ChatGPT hallucinations, it would deliver the same misleading information or further taint it with its own hallucinations.
There’s also the matter of having consent for using data from other AIs. Then again, other AI firms grabbed content from the web indiscriminately without necessarily worrying about copyright and consent.
Beware of censorship
More problematic is the fact that tech companies in China have to adhere to local laws, which include plenty of censorship. Therefore, DeepSeek might not offer users in Western countries a balanced, objective look at things like world events.
I’ve seen examples online of DeepSeek offering answers that align with China’s view of the world. Taiwan is such a hot topic for DeepSeek. And I’ve seen DeepSeek refusing to answer questions about topics that might be sensitive in China.
This is even worse than AI hallucinations, which will eventually be corrected. But censorship will be built into DeepSeek, so there’s the risk of manipulation and disinformation.
Therefore, you can’t trust that DeepSeek will tell you, and you won’t be able to change its default instructions. You won’t be able to bypass censorship. That’s not the AI experience I expect from chatbots.
To bring back TikTok, some say the app’s algorithms might paint China in a better picture than it actually is. The same might happen with other products from the country that are available in the West, AI included.
What about DeepSeek user privacy?
With all that in mind, there’s another big concern I have when it comes to chatting with AI: User privacy.
I don’t want my chats with ChatGPT to train future versions of the AI. Also, I want any prompt data to be stored securely and as briefly as possible on company servers. That’s one reason why Apple Intelligence will be so appealing once Apple gets its own ChatGPT-like chatbot.
DeepSeek does supposedly have strong privacy rules in place, or so it says. However, the data it collects is still stored in China and is subject to local laws. I’ll bring back TikTok again because storing user data in China is another problem the US has with the service.
With all that in mind, I’d be cautious about using DeepSeek for work, research, or fun until we have more assurance about it. The same goes for any other ChatGPT rivals from China, and there are plenty, even if costs will go down significantly for some of those options.
