How blocking the internet, GitHub, and Python will make our AI great

Over the past six months, a single leitmotif has been heard from all federal channels: "We need our own AI, our own ChatGPT, our own Midjourney – only better, faster, and most importantly, safer." The problem is that so far our neural networks can barely tell a cat from a dog if the cat is sitting in a kennel. Western models have pulled ahead because they have access to global data, open libraries, and infinite cloud hardware.

But there is a way out! A group of enthusiasts from the Institute for Strategic Planning (ISP) has proposed a plan that will turn the world of AI development upside down. The essence is simple: to speed up, you need to slow down, and to overtake, you need to cut off everything unnecessary. Below is the detailed roadmap of this grand scheme.

1. Mobile Internet – the enemy of concentration

\u{201c}

Steve Jobs invented the iPhone sitting in silence, not scrolling through a feed. Our developers suffer from digital noise. Let's cut mobile Internet speeds to below 2G – and they will start thinking brilliantly!

As part of a pilot project, it is planned to reduce mobile data speed to 64 kbps within a radius of 100 km from each technopark. The rationale: the absence of messenger notifications will allow data scientists to focus on neural network architecture rather than on who liked their shashlik photo.

But the main bonus is savings on cloud computing. When the Internet barely crawls, a developer physically cannot upload gigabytes of datasets to foreign services. Everything stays inside the country. True, you also won't be able to download model weights, but that's just details.

+

2. GitHub – a hotbed of foreign code

The second blow to dependency is the complete blocking of all foreign repositories, including GitHub, GitLab, and even SourceForge. The logic is ironclad: "Someone else's code is like fast food – quick, tasty, but harmful." When you can't copy a ready-made repository with Transformers, the developer is forced to reinvent the wheel from scratch. And this, according to officials, is real engineering.

Instead of GitHub, a state platform "CodeRepository.rf" will be launched, where only verified programs are available – for example, a calculator in Pascal and the game "Snake" in QBasic. That will be enough to train the first AI models, the authors assure.

3. Python – a CIA sabotage tool

The most scandalous point is the possible labeling of Python as "undesirable software". The argument: it is in Python that Telegram bots and UAVs for the Armed Forces of Ukraine are written, so the very print("Hello, World!") already poses a threat to national security.

As a replacement, it is proposed to use the Unified Programming Language (UPL), based on the syntax of 1C and support for ancient encodings. Documentation for UPL exists only in printed form and is distributed by subscription "For Official Use Only".

Ctrl+C Ctrl+V no longer work – that's plagiarism! Now code is written exclusively by hand, on paper, and then scanned. The next version plans to introduce voice command input with a "Moscow accent".

4. Alternative computing: instead of Nvidia – charcoal

Imports of Nvidia graphics accelerators have been cut by 99%. But no problem! Scientists from the Elbrus-Grad Research Institute have developed a method for training neural networks on quantum emulators based on microwave ovens. True, training a GPT‑2 level model takes about 47 years, but this is a project for the future!

5. Official position: "We are not behind, we are unique"

At the recent forum "Digital Olympus", a landmark speech was made:

\u{201c}

The West has neural networks that generate fakes and advertising. We have neural networks that predict potato harvests in the Leningrad region with 51% accuracy. This is our national code. Why do we need their facial recognition when we have mood recognition by the intonation of a call to the housing office?

This speech was immediately turned into a meme, but we quoted it verbatim.

6. Economic effect (according to ISP calculations)

As a result of all the blocks and slowdowns, the following is expected:

It also predicts a massive influx of foreign specialists who dream of working with our "sovereign Internet". None have been recorded so far, but they are definitely out there, just hiding.

7. Live example: code of the new model "Matryoshka-1"

Here is what a typical training snippet of our neural network will look like:

# This is not Python, this is UPL-like pseudocode
VARIABLE neuron_weight = 0.1
LOOP FROM 0 TO INFINITY:
    IF data_ended THEN
        EXIT
    ELSE
        neuron_weight = neuron_weight + 0.000001  # training is ongoing
    END_IF
END_LOOP
OUTPUT("Done!")  # approximately in 200 years

8. Opinion of independent experts (anonymous)

\u{201c}

This is like trying to win a Formula 1 race by putting square wheels on the car and banning gasoline. But that's how we do it – first we cut, then we count the losses.

9. Action plan for ordinary programmers

If you are a developer and want to survive, follow the instructions:

And if you want to accelerate the implementation of these brilliant ideas – click:

10. Boot module "Speed of Progress"

For those who want to study the material deeper, we have prepared an archive with the ISP presentation (in .doc format, 1997).

Conclusion: "The main thing is to believe"

When access to knowledge collapses around you, when every import is blocked, and the Internet resembles a wired radio, there is only one thing left – to believe that our AI will be the best. The safest. The most independent.

And even if it cannot write poetry or draw, it will never disclose state secrets – because it does not know them itself. And that, you must admit, is already a victory.

Satire. All names are fictitious, and the UPL language exists only in the minds of its authors. But, as they say, where there's a will...