Daniil and David Liberman

China’s post 1990s generation of AI talent is something to behold. One thing that has caught my attention is the age of the people building China’s frontier AI. Kimi’s founder was born in 1992. Many of the teams behind MiniMax, Seedance, Zhipu, and other leading AI companies are dominated by engineers born in the 1990s and early 2000s. This isn’t limited to AI. This year’s Fields Medals also went to two Chinese mathematicians from the same generation. I don’t think these are isolated events. They may be early signals of a much deeper trend: what we’re witnessing is not just an AI boom, but the emergence of a new generation of scientific and engineering talent in China. One possible explanation is education. After the Cultural Revolution ended, China’s education system was gradually rebuilt. The generation that became today’s parents of many 90s and 00s children benefited from that recovery, and many placed an extraordinary emphasis on education. The result may be the first generation of Chinese researchers and entrepreneurs raised in families where higher education had become both accessible and deeply valued. If this hypothesis is right, AI is only the beginning. We may see the same generation produce breakthroughs not only in frontier models, but also in quantum computing, biology, mathematics, computer science, and many other fields over the coming decades.

Great conversation with @MRRydon from Aethir. As a DeAI industry, we should not compete with each other, we should fight the centralization. It doesn't matter which protocol gets more recognition. Ultimately, with more decentralization, we all win

🤑 Exciting update: the Gonka Bug Bounty Program on HackerOne has increased rewards for reported critical-severity vulnerabilities to $15K–$25K, up from the previous range of $5K–$10K. Read the program details here: medium.com/product-scienc… Analyze the code on GitHub: github.com/gonka-ai/ Report vulnerabilities through HackerOne: gonka.ai/docs/report-vu…

@brian_armstrong Let’s cut the spend even further. @gonka_ai serves GLM and Kimi at 10x lower cost

How to keep AI spend flat while token usage grows exponentially: Not with friction and spend alerts. With better defaults, routing, and caching. Better Defaults (not Usage Caps) – Engineers can choose any model they want, but defaults matter. We’re experimenting with defaulting to open weight models like GLM 5.2 and Kimi 2.7 through our LLM gateway, while still encouraging engineers to choose the right model for the task. 91% of our employees were never hitting their usage caps, so instead of lowering caps and driving up alerts, we're moving to cheaper defaults. Note that code reviews use a diversity of models, so they can check each other's work. Better Routing – In our custom harnesses, we preprocess prompts and route to the best model for the job, considering cache hits and model pricing. For instance, you may want a frontier model for planning, but not for execution where they can be overkill. Ultimately, humans shouldn't be choosing models - AI can automate this task. Better Caching – Cache misses are the easiest way to drive your cost up. All of our requests are cache aware, so we’re reusing a warm cache wherever possible. For example, our cache hit rate went from 5% → 60% in LibreChat once properly implemented. Keep Context Lean – Start fresh sessions when switching tasks. Scope file context narrowly. Disconnect unused tools. Don't just compact. The goal isn't fewer tokens used, it's fewer tokens wasted. Better Visibility – Our engineers can use as many tokens as they want, from whatever model they want, but we’ve made usage visible – and the more you spend on AI, the more impact we expect. The goal isn't to suppress usage. It's to build the infrastructure that makes exponential growth sustainable. Putting this into practice has cut our AI spend nearly in half, while our token usage continues to grow.

Satya Nadella suggested in his latest essay that the future belongs to those who build both human capital and token capital When you pay for AI tokens or subscriptions to OpenAI or Anthropic, you don't accumulate any lasting capital - they do. The providers use that revenue to build data centers and train their next proprietary models But decentralized AI introduces a different dynamic When you use @gonka_ai , you can both consume AI compute today and accumulate tokens that represent future AI compute in the decentralized network

Migrate your agents to @gonka_ai before centralized dependencies become effectively irreversible proxy.gonka.gg

Access to neural networks based on your passport has ceased to be a dystopian fantasy A few days ago, Anthropic released Fable 5. Essentially, it's their flagship Mythos 5, but censored (which was already a concern). On top of that, almost immediately, the U.S. government demanded that access to both models be banned for all foreign nationals. And it doesn't matter where you are, even if you work at Anthropic itself There are around 300 million U.S. citizens, while the world's population is 8.3 billion. That means roughly 8 billion people will officially be cut off from the most powerful AI model available today Even Andrej Karpathy, a true AI industry celebrity who recently joined Anthropic, got caught in the crossfire. He simply doesn't have an American passport. As a result, the company temporarily shut these models off for all customers while the dispute with the authorities is being sorted out. We're getting used to a new reality First, AI has officially become a strategic resource. On par with uranium, chips, and weapons. Second, access is granted based on citizenship. Your talent, needs, or money no longer matter Third, we now have two closed camps: the U.S. and its allies versus China and its allies. Everyone else will have to make do with whatever they are allowed to use Today, it is not so important who exactly built the best model. What matters is the precedent itself: access to intelligence for 95% of the planet can be cut off with a single letter from a regulator. Which means the world desperately needs independent infrastructure that cannot be switched off at the snap of a finger

Join us for the two most important days in Japan's Web3 industry this year at @WebX_Asia. Don't miss our talk and then come chat with the Gonka community at the booth x.com/WebX_Asia/stat…

"... building a world in which everyone can flourish requires shared responsibility and courage. No one can single-handedly bear the weight of the challenges the world is facing, just as no one is so weak that they cannot play their part, for “power is made perfect in weakness” (2 Cor 12:9). All are given their own section of the wall..." - Pope Leo XIV’s Magnifica Humanitas

AI deflation is starting to hit. @nebiusai is rapidly increasing its GPU prices: H100: $2.95/hr → $3.85/hr B200: $5.50/hr → $7.15/hr This is what the capital shift into AI looks like. As intelligence becomes abundant, most goods, labor, and even fiat currencies may fall in relative value against AI infrastructure assets: compute, energy, chips, data centers. AI infrastructure is becoming monetary infrastructure.

@PeterMcCormack Exactly, the feeling is incredible. This is the moment when AI stops being a chatbot and becomes part of the company’s nervous system. But it also shows dependency

I’m addicted to this. I keep expanding the scope of what I’m building. I’ve built into the system AI Agents for design and security. The designer keeps working on UX improvements and the security agent is providing a 24/7 review of the systems, external security updates, patches etc… When I’m done I’d like to get an actual developer and security expert to review it with an honest assessment.

We are officially past the point where decentralized AI inference is only a philosophical argument Over the last 24 hours, Gonka processed hundreds of thousands of inferences and billions of AI tokens across the network. Across a decentralized network of real hardware built by dozens of participants - companies and individuals This is still early and rough around the edge, people are still hitting limits, but when they do, it’s another milestone for the network. But this is exactly how new infrastructure starts. Perfection is not important, usability is. It is already real, being used, and improving faster each week The demand for inference is already coming from millions of developers, agents, builders, and users around the world. The question is whether this demand will keep flowing into a few centralized providers or whether it will drift towards an open, verifiable, decentralized alternative If you want decentralized AI to exist, use it. Try the inference, build with it, help other people connect, launch a swarm of agents, and try to bump into the limit to push it further. Every request, every test, every fix, every GPU, every new user makes the network stronger Bitcoin started this way. This is how open infrastructure wins

Thank you, @PeterMcCormack, for the opportunity to discuss this important topic and share our thoughts with your audience regarding what might happen without the decentralization of AI (including the darkest scenarios). Watch and beware

Our conversation with Alexander Sokolovskiy, seen by 1M+ Russian viewers, is now in English It’s about more than AI. It’s about intelligence becoming infrastructure. The key question: Who owns it? If AI is controlled by a few corporations and states, the future stays permissioned Open models, distributed compute, shared intelligence Let’s build that future youtu.be/2-3CFlNk180

@p_volnov launched a great product as part of YC 🔥 Congrats Peter, you rock ⚡️

‘The world is changing. AI—the biggest frontier models—can fit on a tiny flash card. They are essentially copyable, so every human on Earth can have them. We can actually bring abundance to billions of people around the world if we manage to make these technologies universally accessible.’ @DaLiberman at LA Hacks 2026. youtube.com/watch?v=NtZrvn…

DeAI is not built on technology alone. It also depends on transparency, trust, and responsibility to the community around it Periods of strain are when principles are most severely tested. They rarely collapse in a single moment. More often, they erode gradually, through small concessions, accumulated compromises, and decisions made for immediate convenience If DeAI is meant to differ from centralized corporate systems, that difference has to be visible in practice, not only in language

Proud to be early supporters of @Pron Congratulations, @antlevandowski!

Anthropic has not released its most powerful model to the public and has given access only to a closed group of partners - Apple, Meta, Microsoft, Nvidia This is not an exception. It is a pattern Not long ago, public leaders in the industry were saying that AI would be accessible to everyone, that it would expand human potential. In reality, we are returning to a familiar structure: a closed club, selective access, control This is not about safety, no matter how it is framed. It reflects a deeper logic of elites - dividing the world into insiders and outsiders and controlling access to key resources In the past, those resources were oil, energy, finance, water. Now intelligence and compute have been added to that list And the outcome is always the same: not abundance, but competition for access, gatekeeping, constant tension and conflict AI is only accelerating this dynamic. For the first time, this is not just about resources, but about a system that shapes thinking, decisions, and the structure of society itself If this layer remains centralized, we do not get a future of abundance, but a more advanced form of dependency So the question is no longer about technology. It is about architecture Either AI remains in the hands of a few players, and the world is ultimately divided between those who have access and those who do not Or an alternative layer emerges - decentralized compute, open-weight models, distributed infrastructure This is the only way to break the cycle in which resources inevitably concentrate in the hands of a few Otherwise, history will repeat itself. And we will find ourselves back in the Thucydides Trap