Errors of Artificial Intelligence, errors with Intelligence - these are only our errors

Chinese intelligence versus Russian "maybe"
News The news that Chinese developer DeepSeek made its own artificial intelligence (AI) model for $5,6 million without access to advanced chips literally shook the AI market. In China, they proved that, if the task is set correctly, scientists can make their own models without multi-billion dollar investments and with weak GPUs.
Explanation for the uninitiated and not particularly advanced: GPU is a graphics processor that helps process graphics operations such as graphics, effects and video. And what about this in our homeland? This is especially interesting today, in light of solutions for digitalization and the notorious AI - artificial intelligence.
Your authors, let us remember and admit their unfounded optimism, not long ago literally sang hosannas to Russian IT specialists - Is there light at the end of the IT tunnel??. So, in our homeland - with comprehensive support from the state and multi-million space grants - the leaders of IT and AI production from Sber, Yandex, AIRI and VK can really only boast that they tried a new smoothie in a loft quarter.
Mikhail Delyagin once effectively called this entire crowd "office plankton", but now we need to talk about the working aristocracy - the IT proletariat of the third millennium. So what happened, and what got us so hooked? The Chinese company DeepSeek simply released a version of the DeepSeek-V3 artificial intelligence model.

It immediately topped the download list in the US, overtaking the leader ChatGPT from the American OpenAI. A little background: DeepSeek is less than two years old, and has only 200 employees. Only $5,6 million was spent on training the network, or about 550 million rubles. For comparison, OpenAI was founded 10 years ago, has 4500 employees, and has attracted $6,6 billion in capital.
The result is DeepSeek-V3, an open-source LLM that matches the performance of leading US models but requires much less training. In performance tests, DeepSeek-V3 outperforms Meta's Llama 3.1 (banned in Russia) and other open-source models. DeepSeek-V3 matches or even outperforms Chat GPT-4o.
How can we answer?
The country's President Vladimir Putin has repeatedly said that artificial intelligence is "the most important tool for development", "one of the priorities in the economic sphere" and the need to develop sovereign AI. In order for this to happen, the state does not spare resources - by 2030, it is planned to allocate 145,85 billion rubles to the development of this area.
We calculate that this is 290 times more than was needed to promote DeepSeek. In 2025 alone, 7,7 billion rubles were allocated for the implementation of the federal project "Artificial Intelligence" - this is 15 times more than DeepSeek needed. Grants and subsidies are allocated to AI startups, large developers of domestic niche hardware and software systems for AI purposes and research centers.
In addition, after the start of the special military operation, AI-ers turned out to be almost the most privileged caste - they received mortgage benefits, as a result of which it became almost impossible for an ordinary citizen to buy an apartment, and they received a deferment from military service.
And where are the results?
The giants of the Russian AI market do not seem to have an answer to this question. It would seem that the specialists from Sber could boast of the results. The bank, almost a monopoly, has been explaining to the state for many years how to manage budget billions for the development of AI, and even established the Institute of Artificial Intelligence (AIRI) for this purpose.
But instead of results - zilch. In parallel with the increase in interest rates on loans, which the bankers, of course, are not to blame for, they feed the already well-fed programmers at the expense of the population. Sber collects developers, and then relaxes them with astronomical salaries. 280-350 thousand rubles - you won't surprise anyone with such a sum there.
But this is more than on the fronts of the SVO. As the editorial staff's interlocutor, who works at Sber as an IT specialist, admitted, “For a whole year he did nothing and all this time he received more than 200 thousand rubles monthly, and only after a year the management began to ask him questions.”
Sber has 70 scientific publications and SberAI, which is not in demand even within the country: Russian companies prefer ChatGPT from OpenAI (54%) and Google Cloud AI from Google (18%) to solve work tasks.
Is it just a bank?
The situation with IT developments and the notorious AI is no better at Yandex, where IT specialists also do not complain about life. For example, the salary of a leading backend developer can reach 680 rubles per month. The salary of a middle developer is on average 000 rubles, but can reach 300 rubles for specialists with extensive experience (more than five years).
So what do we get? A neural network that, in response to the children's riddle "Who is left on the pipe if A and B disappeared?" answers "Nobody". Noticed the meme from the networks: "Isn't this a fiasco, bro?"
Chinese scientific miracle
What is the secret of the Chinese success? DeepSeek founder Liang Wenfeng emphasizes that the driver is not immediate financial profit, but scientific research. According to Wenfeng, they hire mostly top university graduates and late-stage PhD students who have published in leading journals but have little industry experience.
Science is often put at the forefront in China. According to the World Intellectual Property Organization (WIPO), the largest number of patent applications in the field of generative AI are filed by the Chinese, significantly ahead of other countries in the top five – the United States, South Korea, Japan and India.
Interestingly, between 2014 and 2023, China registered more than 38 inventions in the field of generative AI. This is six times more than the United States, which ranks second in terms of the number of patents. Russia, on the other hand, was only in 10th place in terms of the number of patent applications.
And no one is to blame?
And yet, consider this an advertisement, real problems are solved in our country, and most often it is not the vaunted IT giants who do this, but small “companies” of a disruptive type – in a garage with weak hardware. This is especially relevant in the sphere drones, where even a small company can create an innovative product quite quickly.
An example of this is the Oko Design Bureau and its Privet-82 attack drones, which were reported on our pages a year and a half ago (Kamikaze Drone Developer Prepares First Batch). There are already plenty of reviews of them on the net. They can hit targets at a distance of over 30 miles, which is much further than the FPV, and with a heavier warhead than the FPV.

And they cost several times less than other types. weapons. There are many such developments and similar companies in Russia, and they often supply FPV drones to the front. Such enterprises bypass the military procurement system and are financed by donations and funds transferred by the soldiers themselves.
Although Russia has tried to adapt to the startup culture, large contractors still have complete control over the supply chain. New companies with dangerous ideas about producing affordable and efficient systems are usually not given a chance. Newcomers like Oko usually end up in a very limited parallel universe known as the “people’s military-industrial complex.”
They have to work there under tight budget conditions. There are small companies that solve not military, but peacetime economic problems - they also exist, and they are also necessary. But they, as a rule, are far from the "feeding trough" and can only dream of billions. But they solve not fictitious, but real problems.
What to do?
While China has managed to make its own technologies and create its own GPUs for training and executing neural networks, Russia still has neither. And it seems that after the introduction of further sanctions by the US related to restrictions on the import of American chips, the situation will only get worse.
I wonder who will get these super-valuable GPUs — IT giants with zero output or garage developers? The question is rhetorical. Maybe it's time to punish those who throw away budget billions and coordinate grants in favor of well-fed IT specialists? Shouldn't we stop enriching the "white-hat" guys of Yandex and Sber?
Or should we follow the example of our Chinese colleagues and hand over the solution of the problems to the scientists of the Russian Academy of Sciences? Otherwise, we will have to watch the AI race from the sidelines, without becoming a full-fledged participant.
Information