Why You Need to Become a Neuro-Punk Right Now

A developer argues that the developer community must prioritize building large language models free from corporate and state control, warning that companies like Anthropic propose licensing regimes that would create a neo-feudal system where non-AI companies become dependent on Big Tech APIs. The essay also highlights how governments in China, Europe, and Russia are using AI for surveillance and control, urging developers to resist this trend by investing in open-source, decentralized AI.

A short essay on why the developer community should invest as much effort as possible into LLMs that are free from corporations and states. ML researchers and hardware engineers both need to contribute here. The latter may even be more important, because whether users can run advanced LLMs on personal hardware depends on breaking NVIDIA's monopoly. This essay is highly political, especially in the opening sections. Keep that in mind. The other day, almost at the same time as the release of Fable 5, Anthropic's Dario Amodei published an article called "Policy on the AI Exponential", where he discussed what the world should do with powerful AI-based systems. All sections except the first contain fairly reasonable proposals, or at least proposals worth discussing. I will not consider them here. The real core is in the first section. In that first section, he effectively proposes a system in which the state would be required to license advanced AI systems, measured by the amount of compute used, and even ban the release of models that are not considered safe for society. In practice, this repeats a story as old as the world: a large corporation wants to regulate the market so smaller companies do not interfere with its ability to earn mountains of money, all under noble-sounding pretexts. And the point is not that Amodei is some villain. He is simply an entrepreneur who wants to earn as much money as possible. Any large corporation would prefer not to let smaller companies near the feeding trough in its field. Anthropic is merely saying this openly, and that is all. In effect, AI Big Tech wants a future where all non-AI companies become its serfs, mortally dependent on intelligence delivered through Anthropic's API, or OpenAI's, or Google's, and so on. In practice, those AI companies would hold the revenue of all these other companies in their hands. Without them, the whole economy around those companies would crumble into dust. What we get is a neo-feudal system where ordinary people occupy the lowest rung of the newly formed social ladder, while the highest rung belongs to corporations fused with the state. Now let us cross the ocean and look at what is happening in China. For an AI enthusiast, there is a lot of interesting activity there. Over the last half year, Chinese Big Tech has flooded the market with powerful open-source LLMs, which is of course good and welcome. But let us think about why the Chinese government, which de facto supervises the entire AI sector in the PRC, would promote this field so actively and intensely. Economic dominance? I think that is an important factor, but the main reason is probably elsewhere: regime stability and total control. Chinese authorities already actively use AI for social credit and surveillance of the population, while many other applications remain behind the scenes because Chinese security agencies are closed to outside observation. The Chinese government wants to build a system where resistance is impossible by definition. But followers of Confucius are not the only ones moving in that direction. Many other countries, including European ones, are also gradually tightening the screws and restricting the internet. And the most visible example for many Russian-speaking readers is Russia itself. The Russian government is currently destroying the internet piece by piece: throttling services, blocking platforms, criminalizing speech, building censorship infrastructure, and trying to turn the network from a public space into a controlled pipe. This is not some abstract authoritarian tendency somewhere far away. It is happening right now, in plain sight. All these actors would be immensely happy to have an extremely powerful and advanced AI in their hands, one that could identify dissidents in advance, monitor them, and repress them. That is exactly the future they are moving toward. Do you personally want it? Or let me ask differently: do you like where the world has gone over the last four years because of well-known events? Modern states brought the world there. And they create this darkness without any especially advanced technical tools. What happens when advanced AI falls into their hands? The problem is not only that states and corporations fused together are screening a live-action version of 1984 for all of us. That is a very bad scenario. But there is also a catastrophic one. Everyone probably knows the thought experiment about the paperclip maximizer. It describes a situation where there is exactly one AI monopolist, and it contains critical behavioral bugs that cause it, after receiving the simple task "improve paperclip production", to exterminate humanity and cover Earth, then the entire Solar System, with paperclip factories. The situation sounds absurd, and it is absurd, but it emphasizes one important thing: AI is not a person in the human sense. It is a very powerful program, and like all programs, it can have bugs. In the case of an extremely powerful intellectual system, those bugs can lead to catastrophic consequences for everyone around it, if not for humanity as such. And people like Ilya Sutskever, Amodei, Altman, and other major figures in AI Tech are trying to convince us that advanced LLMs should be handled exclusively by closed laboratories. If we are going to get the Terminator scenario, it is under exactly this kind of arrangement. But there is an alternative: a world where powerful AI does not belong to closed private or government labs, but to the global developer community, which could develop and repair it the same way it has developed the Linux kernel for two decades. I do not deny that such a scenario could cause unprecedented political and economic instability. But for me, that future is much preferable to the world of 1984 or The Terminator . Now let us think about how the beautiful future described above could be avoided. It is worth mentioning a topic that is almost never raised even among independent AI researchers who support open source: hardware engineering. Everyone, or almost everyone, knows that LLMs are mostly trained and run on GPUs, and NVIDIA is effectively the monopolist in the GPU market. NVIDIA has little interest in making large LLMs run on cheap hardware. Many people like to compare the energy efficiency of meatbags and LLMs, but such comparisons often forget one detail: transformers do not necessarily have to be computed in full. They can, in principle, be computed sparsely as needed, from an SSD for example. The bottleneck is the data bus, which would have to move a huge amount of data back and forth. But that is a hardware-engineering problem, not a problem of LLMs themselves or their architecture. That is why we need specialists who understand circuit design and FPGAs, people who could contribute to the open-source GPU segment. These are the people who can make LLMs a truly accessible technology. There is nobody else to do it. NVIDIA is definitely not interested in that future. It is often assumed that truly useful language models require trillions of parameters. But is that really true? Google's recent open-source release of Gemma-4-12B and 24B suggests that it may not be. For their size, these models show a surprisingly strong ability to handle agentic tasks. It is entirely possible that a model does not need hundreds of billions of parameters to reason well, while factual knowledge can be supplied through RAG systems, which are actively developing right now. We should also think about how to improve the transformer architecture itself. This is the hardest area to understand, but it may contain the most powerful breakthroughs. As a technology, LLMs now resemble the early internet of the early 1990s. Many people already understand that this technology can change the world, but few understand how and where to use it correctly. The dot-com bubble and the AI bubble look suspiciously similar in their dynamics. The early internet was a golden age for hackers and enthusiasts, some of whom directly participated in shaping a new technological order. But with AI, the stakes are much higher. In practice, we face a choice between 1984 , or even the disappearance of humans as a species, and decentralized AI, where this powerful technology belongs to everyone rather than to a thin layer of elites. The rawness of the technology is exactly what lets enthusiasts contribute to the field and push it toward decentralization, not away from it. That is the end of the article. Until next time.