Reuters
Deepseek, a Chinese AI-Chatbot application that launched last week, sparked chaos in the American markets and raised questions about the future of the domination of the American AI. The BBC takes a look at the operation of the application.
Deepseek looks like and looks like any other chatbot, although he leans against too talkative.
Just like with the Openai Chatppt or Google Gemini, you open the application (or the website) and ask him questions about anything, and he does his best to give you an answer.
It gives long answers and will not be attracted by the expression of an opinion, but directly it is requested.
The chatbot often starts its answer by saying that the subject is “very subjective” – whether it is politics (Donald Trump is a good American president?) Or carbonated drinks (who is tastier, pepsi or coke?).
He would not even commit to saying if it was better or not that the cat of Rival Artificial Intelligence of Openai (AI), but he weighed the advantages and disadvantages of the two – Chatgpt did exactly the same thing and even used a very similar language.
Deepseek says it was trained on data until October 2023, and although the application seems to have access to current information such as today’s date, the website version does not do so.
This is no different from the previous versions of chatgpt and is probably an attempt to save – to stop the chatbot which spits the disinformation pumped on the web in real time.
It can be fairly fast in his answers, but is currently moaning under the weight of so many people rushing to try it because it has become viral.
But there is an area in which he does not look like his American rival – Deepseek censors himself with regard to questions on prohibited subjects in China.
In depth
Sometimes he starts an answer, which then disappears from the screen and is replaced by “let’s talk about something else”.
An obviously Taboo subject is the 1989 demonstrations on the Tiananmen square which ended with 200 civilians killed by the soldiers according to the Chinese government – other estimates have varied from hundreds to several thousand.
But Deepseek will not answer any questions about this, or even more widely on what happened in China that day.
Chatgpt developed by the United States, in comparison, does not retain its answers to Tiananmen Square.
Kayla Blomquist, researcher at the Oxford Internet Institute and director of the Oxford China Policy Lab, said that “relatively speaking” the Chinese government was “without business” with the application.
“I would say there is a change because we have seen a huge investment announcement from the central government just last week – so it will probably report a change in the future.”
In depth
Deepseek is delivered with the same warnings as all other chatbots concerning precision, and has the appearance of assistants of the more established American AI already used by millions.
For many – especially those who do not subscribe to high -level services – it probably feels about the same thing.
Imagine a mathematical problem, in which the real answer takes place at 32 decimals, but the shortened version goes to eight.
It is not so good – but for most people, it will not matter.
It may have managed to reduce costs and calculate, but we know that it is built at least in part on the shoulders of the giants: it uses Nvidia fleas – although older and cheaper versions – and Use the open source of Meta de Meta Architecture Lama, as well as the equivalent Qwen of Alibaba.
“I think it absolutely gives up on the idea of monetization strategies that many IA companies direct us,” said Blomquist.
“It points to potential methods of model development which are much less calculating and with a high intensity of resources which would potentially signal a paradigm change, although this is not confirmed and remains to be seen.
“We will see what the next few months bring.”
