While we use AI to draft emails, it's already rewriting the source code of reality

I recently came across a thought: most people still treat AI as a smart replacement for Google or a cool generator for pictures, music, and video. And while society keeps thinking in those terms, the world has quietly but fundamentally flipped upside down.

Just look at the facts from recent weeks: a guy with no biology background uses ChatGPT to create a personalized mRNA vaccine for his dog, and the tumor shrinks by 75%. AI models are finding zero-day vulnerabilities in operating systems that top engineers missed for decades. And defense giant Palantir publishes a manifesto openly stating: AI weapons are inevitable — the only question is who builds them first and whose values they will defend.

This isn't Google or Wikipedia. This is something far more massive - a leviathan that hasn't yet shown its full size.

It just so happens I've been watching this unfold for a while. Last year flew by at sub-light speed: crypto, stablecoins, Real World Asset (RWA) tokenization, and the explosive growth of AI. There was so much information that I built a no-code pipeline (n8n + GPT-4o + Telegram) that scoured the news, analyzed it, and delivered a ready-made digest. Everything worked perfectly - but the quality of the AI analysis wasn't quite at the level I'd actually enjoy reading. By that point I was a bit burned out on tech, so I took a month-long break.

It didn't last long. Cloud LLMs started producing noticeably better output, especially on paid tiers. And one sunny winter day, walking through the woods in Zurich with my daughter, an unsettling thought hit me: what if corporations just cut off access to these LLMs one day? Maybe I should download one to my own machine. After all, a file containing the weights of a trained neural network is a treasure trove of knowledge - like having Google and Wikipedia running entirely on your local hardware. That thought was the white rabbit. It led me down the Local AI rabbit hole, and I've been wandering there ever since.

My first attempts to run a model on a standard CPU were agonizingly slow - just like that supercomputer in The Hitchhiker's Guide to the Galaxy that spent 7.5 million years arriving at the answer 42. Apparently it was also running an LLM with a hundred billion parameters on a rack of overheating DDR3.

Exploring local AI pulled me into a lot of different questions, but here I want to focus on one: the governance and ethics of AI - whether cloud-based or local.

Back in 2013, defending my Master's thesis on hybrid AI systems in medicine, I was asked: "Is it even ethical to trust a machine with a diagnosis?" At the time it sounded like a philosophical puzzle. Today it's a deeply practical question that cuts to the foundations of society.

Just look at what's happening right now. OpenClaw - a popular open-source AI agent with full system access - turned out to be riddled with critical vulnerabilities: key leaks, remote code execution, data exfiltration. Not malice - just no one thought through the consequences at the design stage. Anthropic intentionally restricted access to its new Mythos model because it finds vulnerabilities that engineers hadn't caught in decades - and the company itself admits it doesn't fully know how to handle that. This isn't alarmism. It's a statement of where we actually are.

The conversation about AI is no longer just about technology. It's a conversation about power. Whoever owns the stack - the hardware, the data, and the models - will set the rules. That's exactly why I completed the AI Governance program authorized by Oxford University's Saïd Business School. Not for a certificate - to understand the architecture of this new world.

The main takeaway: ethics committees alone won't save us. We need solid legal frameworks with Meaningful Human Control and protection against the monopolization of compute - otherwise we risk waking up in digital neo-feudalism.

I'm a technocrat and I love innovation. Technology is neutral: in the right hands it solves problems that once seemed impossible. But that's precisely why governance matters - so this colossal power serves humanity, not the other way around.

P.S. For those who also see the value in technological sovereignty and have gone down the Local AI rabbit hole - I've started publishing my own quantized models (Gemma, Qwen, etc.) on Hugging Face. I use imatrix quantization - a more precise method than standard 4-bit compression, which better preserves model quality at smaller sizes. Just getting started, but test them, critique them, use them: huggingface.co/Krasnopjorovs

P.P.S. Drowning in noise was the other option. Instead: a local personal intelligence pipeline — news from dozens of sources, embedded into a vector database, synthesized by a RAG-powered AI agent into a single daily briefing. 5–10 minutes and you know what's actually happening in the world, not just what's trending. MLP (Minimal Lovable Product) live at artjoms.ai (and yes, naming a site after yourself is the height of modesty).