Habr AI author proposes building deterministic AI on a MacBook Air instead of joining the GPU race

Habr AI has published an opinion piece in which the author offers an alternative to the race for ever more expensive GPU clusters. Rather than scaling probabilistic models, he promotes a deterministic AI core written in Rust that, by his account, already runs locally on a MacBook Air M2 with 8 GB of memory.

Why the author is arguing with the market

The text opens with a direct attack on the current strategy of market leaders OpenAI, Google and Meta*. The author calls it "brute force": in his view, the fundamental limits of LLMs are not solved mathematically but masked with ever more GPUs, megawatts and investment. By his logic, the industry compensates for architectural problems with computing power instead of looking for a stricter decision-making model. The post on Habr is labeled "Opinion", so it is more a manifesto and a polemic against the market than an academic paper.

The article lists three reasons the author considers this approach a dead end. First, energy consumption: training and running large models demands ever more electricity and hardware. Second, the stochastic nature of LLMs: the system does not derive an answer from strict rules but guesses the next token. Third, hallucinations, which the author regards as an inherent property of probabilistic architecture, not a bug that extra filters can fully close.

"The future of artificial intelligence is not a question of money.

It is a question of the right mathematics."

What he proposes

Instead of the classic neural network design, the author describes a deterministic core with O(1) logic. The key idea is that the time to make a decision and validate it stays constant, and the system works not with a long probabilistic context but with intents represented as mathematical formulas. This is meant to remove the very principle of "guessing" and make the model's behavior predictable. Hence the claim of more reliable AI for scenarios where an answer must not just be generated but verified.

The text also gives concrete claims about the project. The author writes that the current core version, v0.26.0, has already been tested against cloud models, and that the solution is deployed on an ordinary MacBook Air M2. The main stack is Rust, with no Python or intermediate interpreters: according to him, high-level intents are translated directly into executable commands. He stresses that he is betting not on general "magic" of the model but on a controllable architecture.

• Deterministic O(1) logic for inference and verification
• Local execution on a MacBook Air M2 with 8 GB of RAM
• Focus on Rust and DMA instead of heavy abstraction layers
• Claimed performance of up to 5.4 million operations on a CPU
• Comparison with cloud AI systems in v0.26.0 core tests

How the protection works

A separate section of the article covers self-learning and safety. The author claims the core has a built-in "constitution of universal human values", implemented not as a set of text prohibitions but as a system of mathematical axioms inside the logic. The article states explicitly that this level of constraint is harder to bypass with ordinary prompt engineering, because unwanted actions must be cut off at the validation stage. The key thesis is simple: AI should be not only smart but also verifiable at the level of the architecture itself.

This is also where the text makes its main ideological turn: the author contrasts local, controlled, mathematically rigid AI with cloud models, which scale along with costs and risks. He also writes that the technology is already protected by a BSL license and is being prepared for patenting, and that technical details and a roadmap are published in the project's repositories. All of this is presented as groundwork not for a chatbot but for autonomous systems where an error is critical. The author's closing thought is broader still: the future of this kind of AI should begin not in a data center but on the user's device.

What this means

The Habr article matters not as proof that a new architecture has won, but as a signal of growing demand for more compact, predictable and local AI. Even if the author's claims still need independent verification, the direction is clear: the market is tired of the idea that every AI problem can be solved only with one more rack of GPUs. Against the backdrop of rising compute prices and interest in edge scenarios, texts like this already set an alternative agenda.

*Meta has been designated an extremist organization and is banned in Russia.