The Chinese AI tool is as good as it seems?

Deepseek, a Chinese AI-Chatbot app, which launched last week, caused chaos on US markets and raised questions about the future of AI dominance in America. The BBC is looking at how the app works.

Deepseek looks and feels like any other chatbot, even though it leans too talkative.

Just like with the Openai or Gemini Chatgpt on Google, you open the app (or website) and ask it questions about everything and does your best to answer you.

He gives long answers and will not be attracted to expression, but is directly requested.

Chatbot often begins its answer, saying that the topic is “highly subjective” – ​​is it a policy (Donald Trump is a good president of the United States?) Or soft drinks (which is more delicious, Pepsi or Cox?).

He would not even be committed to saying if he was better than the rival assistant -intelligence of Openai (AI), but he weighed the pros and cons of both -Chatgpt made absolutely the same and even used a very similar language.

Deepseek says he was trained for data until October 2023. Although the app seems to have access to current information like today, the website version does not.

This is no different from the more more Chatgpt versions and is probably a similar attempt to protect – to stop the chatbot from throwing away the real -time misinformation.

It can be quite fast in its answers, but at the moment it is moaning under the weight of so many people who rush to try it as it has become viral.

But there is an area where nothing like the US opponent is – Deepseek Censorship itself when it comes to questions about objects banned in China.

Sometimes an answer begins, which then disappears from the screen and is replaced by “Let's talk about something else.”

One obvious taboo subject is the protests since 1989 on Tiananmann Square, which ended with 200 civilians killed by the military according to the Chinese government – other estimates range from hundreds to many thousands.

But Depepeek will not answer any questions about the matter or even wider about what happened in China that day.

The USA Chatgpt developed, for comparison, does not restrain himself in his answers to Tiananmen Square.

Kayla Blomquist, a researcher at the Oxford Internet Institute and director of the Oxford Policy Laboratory China, says the Chinese government is “distributed” with the app.

“I would say there is a change, as we have seen a message in huge investments from the central government only in the last week – so this is likely to signal a change moving forward.”

Deepseek comes with the same warnings as all other chatbots in terms of accuracy and has the appearance of more established AI assistants already used by millions.

For many – especially for those who do not subscribe to the highest level services – it probably feels almost the same.

Imagine a mathematical problem where the real answer reaches up to 32 decimal signs, but the shortened version lasts up to eight.

It's not so good – but for most people it won't matter.

It may have been able to reduce costs and calculate, but we know that it is built at least partly on the giants' shoulders: use NVIDIA chips – albeit older, more expensive versions – and uses an open META code of Meta Architecture of Llama, as well as Alibaba's equivalent QWen.

“I think this absolutely provokes the idea of ​​revenue strategies that many of our leading AI companies in the US had,” said G -Ja Blomquist.

“He points out the potential methods for the development of the model, which are much less calculated and intensive resources that would potentially signal a change in the paradigm, although this is unconfirmed and remains to be seen.

“We'll see what the next few months wear.”