Is China’s AI tool as good as it seems?

Deepseek, a Chinese AI-Catbot app, which was launched last week, has sparked chaos in the American market and raised questions about the future of AI’s dominance in America. BBC meets a look at how the app works.

Deepseek looks like another Chatbot, although it tends to be very excessive.

As with Chatgpt from Openai or Google’s Gemini, you open the application (or web site) and ask him questions about anything, and he is doing his best to give you a response.

It gives long answers and will not be drawn on the expression of opinion, but this is requested directly.

Chatbot often begins its response by saying that the topic is “very personal” – whether this is the policy (is Donald Trump a good American president?) Or soft drinks (which are more delicious, pipse or coke?).

You will not even comply with saying whether it is better than the competing artificial intelligence (AI) competing with Openai, but it weighs both positives and negatives – Chatgpt did the same thing completely, and even use a very similar language.

Deepseek says he has been trained in data until October 2023, and while it seems that the application can access the current information like today’s history, the site version does not.

This is not different from previous versions of Chatgpt and may be a similar attempt to protect – to stop Chatbot that comes out of the wrong information that was pumped on the web in the actual time.

It can be fast in his responses, but he is currently moving under the weight of many people who rush to try it as a viral went.

But a single field is not similar to its competitor in the United States – Deepseek Danbors himself when it comes to questions about the topics banned in China.

Sometimes a response begins, which then disappears from the screen and is replaced by “Let’s talk about something else.”

It is clear that one of the forbidden topics is the 1989 protests in the Tiananmen Square, which ended with 200 civilians killed by the army, according to the Chinese government – other estimates ranged from hundred to several thousand.

But Dibsic will not answer any questions on this topic, or even more broadly about what happened in China that day.

Chatgpt, which was developed in the United States, in comparison, does not retract its answers to Tiananmen.

“The Chinese government was” launching “with the application,” says Kayla Bloomakist, a researcher at the Oxford Institute of the Internet and China Oxford Laboratory Director, says.

“I would like to say that there is a shift because we saw a declaration in huge investments from the central government only last week – so it may indicate a change forward.”

Deepseek comes with the same caveats as any other Chatbots keys in accuracy, and it has more appearance and appearance of American artificial intelligence aides already used by millions.

For many – especially those who do not share services at a higher level – it may be so largely.

Imagine a mathematical problem, where the real answer is run to 32 decimal places, but the short version reaches eight.

It is not good – but for most people, this will not be important.

It might be that, as it was able to reduce costs and account, but we know that it was partially designed at least on the shoulders of the giants: it uses NVIDIA chips – although the oldest and cheapest versions – and Meta uses open source architecture Lama, as well as QWEN equivalent to Ali Baba.

“I think this completely challenges the idea of ​​income liquefaction strategies faced by many AI companies,” said Blomquist.

It refers to the possible methods of developing models that are less accountable and intense for resources that are likely to indicate a transformation in the model, although this is uncertain and still should be seen.

“We will see what the next two months bring.”