I hope for a future where intelligence isn't measured by the sophistication of discourse, but by the ability to understand reality, anticipate consequences, and make better decisions. A future where AI doesn't replace human autonomy, but amplifies it. I want to see systems capable of observing markets, science, energy, economics, and behavior as parts of one complex reality — not to build a machine that "knows everything," but to build tools that help us see better, err less, and learn faster. I hope for technology that is more local, more private, more auditable, and less dependent on structures that concentrate power. I hope AI models stop being just interfaces that answer questions and become scientific instruments: capable of formulating hypotheses, testing predictions, measuring their own errors, and updating their conclusions when reality disagrees. I also hope that technological progress reduces scarcity. Cheaper energy. More accessible computing. More precise medicine. Faster science. Smaller companies capable of building what once required giant corporations. Individuals with access to capabilities that today belong only to large institutions. To me, the interesting future isn't the one where machines become more like us. It's the one where we manage to build better tools to understand the world — without giving up freedom, responsibility, and the ability to decide our own path. Maybe that's why I like the idea of a horizon so much. The horizon is never the final destination. It's just the limit of what we can see today.

In Silicon Valley, the New Rules Are Brutal — and No One Is at the Wheel but the Owners Spoiler conclusion: Silicon Valley has rewritten its rules. Work, governance, and the place of people — three consensus points that have collapsed — now have new answers, and the logic is one and the same: speed over well-being, self-regulation over oversight, machine over human. 1. From "work hard, play hard" to "just work hard" The emblematic slogan of entrepreneurial culture is dead. In its place, a purer and harsher ethic has taken hold. In the words of investor Kulveer Taggar: "It used to be 'work hard, play hard.' Now all that's left is 'work hard.'" The cases are explicit. Corgi, backed by Y Combinator, requires in its job postings a "willingness to work 7 days a week." Khosla Ventures asks for 12-hour days, "24/7" availability, and three years at that pace. The term "996" — 9 a.m. to 9 p.m., six days a week — born in the Chinese tech industry, now describes AI startups in San Francisco. The crucial point: the intensity is framed as a choice, not oppression. Aune Choudhary, founder of Octolane, at 24, is blunt: "996 used to be a flex; now it's just the baseline." The engine is the temporal compression of the AI race — cycles measured in weeks, compute costs burning, the funding window narrowing. In this logic, rest, family, and health become "acceptable sacrifices." Researcher Nathan Lambert warns that this environment has "a human cost," and burnout is certain. First new rule: high intensity is not the exception, it's the entry condition. Those who question it are out. 2. Governance where everyone grades their own exam At the end of September 2026, six of the most influential AI companies — OpenAI, Anthropic, Google, Meta, xAI, and Nvidia — signed a voluntary agreement in Washington. Trump called it "almost a constitution," with "binding moral force." The document provides for internal controls, external audits, and oversight committees. The problem isn't the content; it's the nature: it's voluntary. There is no new regulatory body. There are no enforceable sanctions. Critics are sharp: the CEOs "grade their own exam and escape meaningful safeguards." More revealing is the profile of the signatories: they are exactly the ones building the most powerful systems. Players and referees at the same time. Amodei and OpenAI have already advocated for tough regulation — but when real restriction arrived, what materialized was a promise. The question was never "whether AI needs guardrails," but "who builds the guardrails, who inspects them, and what happens when someone ignores them." Second new rule: responsibility is handed to the governed themselves. Voluntary is the new mandatory. 3. The human: do not cause inconvenience Jensen Huang offered, in June 2026, an influential analogy: the automobile, it was once said, would kill children. The world didn't ban the car — it built sidewalks, crosswalks, and laws. Children left the street; drivers learned to yield. Conclusion: AI needs the same adaptation — not resistance, redesign. The analogy is powerful because it is partially true. But it fails at a decisive point: traffic laws took decades, and in the gap between technology and governance, real people died. Children were run over before the sidewalks. Huang implicitly accepts that cost — and does not accommodate the question: during the gap, who pays? Who participates in writing the rules? When the CEO of a trillion-dollar company says "society needs to adapt," the weight is different: he is a stakeholder. In daily life, the norm seeps in subtly: startups already display rows of shoes at the entrance, with employees invited to take them off to create a "second home." On the surface, cultural fusion; underneath, the same logic — work absorbs all the space of life, and people adjust to the place, not the other way around. Third new rule: adaptation is one-directional. The system doesn't change; you do. 4. What it all means together The three rules reveal a clear outline. At work, human time has been repriced: intensity used to be traded for benefits and flexibility; now the terms have been altered unilaterally — only intensity remains. In governance, responsibility has been redistributed: those who build define "safety," and the public is invited to trust. In culture, adaptation has been redefined: it's not technology that adjusts to humans, it's humans who adjust to the pace of technology. The underlying structure is the same: decision-making power concentrates upward, costs disperse downward. Those who define the rules of work are investors and founders; those of safety, the signatory companies; "social adaptation," those who profit most from it. Those who bear the fatigue, risk, and uncertainty are employees, users, and everyone asked to be patient while the rules mature. Silicon Valley has never lacked the capacity for self-narration. It narrated "changing the world," it narrated "don't be evil"; today it narrates the "new rules." But the legitimacy of a rule never comes from who narrates it — it comes from whether those regulated participated in writing it. In the current arrangement, the answer is not reassuring. Sources: Business Insider, KOTRA, and Sohu (996 culture); Financial Times, Swissinfo, and CNBC TV18 (voluntary agreement); AIforBusiness (Huang's analogy); Times of India (shoeless work).

The hardest blows are the quietest ones. The US built a wall of export controls. China knocked the door down with an MIT license. Washington calculated big: block Nvidia's advanced chips and China will never reach the AI frontier. DeepSeek responded by dropping the weights of a 763-billion-parameter model on Hugging Face under an MIT license — you can modify it, you can sell it, you owe nothing to anyone. That doesn't bleed Beijing. It bleeds the San Francisco unicorns that raised hundreds of billions promising to monopolize the infrastructure of human intelligence. The shot backfired in a way only the Americans themselves could engineer. Jensen Huang admitted publicly: the controls tanked Nvidia's share of the Chinese AI data-center market from "near monopoly to zero." Sounds like a victory? It isn't. ByteDance, Alibaba, and Tencent rushed to order more than US$3.5 billion worth of RTX PRO 5500s — over 3 million units. Washington bans the powerful versions; the Chinese buy the "compliant" ones and keep stacking computing capacity. Nvidia warns it can only deliver 5% to 10% of the orders. In other words: export controls hurt American revenue and, as a bonus, fattened Huawei Ascend with Washington's own hands. But DeepSeek's blow is more brutal: buying chips is spending money; open-sourcing an MIT model is ripping out the opponent's pricing power at the root. Silicon Valley was built on the word "scarcity." OpenAI valued at US$852 billion; Anthropic jumped from US$380 billion to US$965 billion in four months. Based on what? On the narrative of "we have the smartest model, the world comes to rent it." Alphabet issued US$80 billion in stock to burn infrastructure, and the five biggest hyperscalers pledged US$969 billion in AI. All of that requires one premise: AI is expensive, scarce, a thing for the few. DeepSeek gives away a 763-billion-parameter model for free and tells the world: frontier intelligence doesn't have to be rented. Download it, deploy it, modify it. A developer in Berlin, a startup in São Paulo, an institute in Tokyo — all run near-frontier inference without paying an "intelligence tax" to Uncle Sam. The numbers confirm it: by the end of 2025, Chinese open-source models accounted for 17% of downloads on Hugging Face, surpassing the US's 15.8%. Among the world's ten largest open-weight models, seven are Chinese. It isn't generosity. It's elegant economic warfare: don't fight by burning money, but turn the opponent's product into air. When the marginal cost of intelligence trends toward zero, what do OpenAI and company sustain trillion-dollar valuations on? MERICS's analysis is surgical: China's strategy is "broad diffusion, cheap tokens" — occupying the global market with open code and low prices, even without short-term profit. Washington still repeats that "the chip is our leverage." The lever is still there, but DeepSeek moved the fulcrum. You block your lithography; I open my weights. You manufacture scarcity with export controls; I manufacture abundance with an MIT license. When a 763-billion-parameter model flows like tap water through the world's servers, Silicon Valley's script — monopolize the infrastructure, charge the global intelligence toll — was torn up from the last page. The hardest blows are the quietest ones. No stage with drums, no audience in Congress. Just a repository on GitHub, a line of MIT license — and a world that no longer has to beg San Francisco for computing capacity.

The Silicon Valley Nightmare: How a Chinese Company Blew Up AI's Trillion-Dollar Fortress with "Free AI" While Wall Street was toasting the billion-dollar valuations of GPT-5.6 and Claude Opus 5, Hangzhou-based DeepSeek knocked over the first domino. This isn't a routine upgrade — it's a planned economic war. 763 billion parameters, MIT license, $0.1 per task. V4.1-Flash-1017 tells Silicon Valley: your competitive moat is a layer of paper. A Precision Strike on the "Pocketbook" DeepSeek didn't try to build the smartest AI on the planet. It calculated precisely what companies pay for — and delivered exactly that. On official benchmarks, V4.1-Flash-1017 dominates the commercial landscape of agentic coding and tool execution. On DeepSWE v1.1, it scored 77.0 points, against 74.0 for Claude Opus 5 and 73.0 for GPT-5.6 Sol. On Automation-Bench, its 60.5 leaves Opus 5's 50.3 at a lower level. Cost is the final blow: less than $0.2 per task, dropping to $0.1 at low reasoning. OpenAI's and Anthropic's flagships operate between $2 and $10. An order-of-magnitude difference that transcends cost-effectiveness and enters the realm of extermination by dumping — an advantage no competitor can match. The Surgical Strategy That Redefined the Game DeepSeek didn't try to win on every board. It identified where companies actually pay — code fixes, automated execution, fast agents — and concentrated everything there. This demolished the closed giants' narrative. Silicon Valley sold Wall Street the creed that cutting-edge intelligence requires computing power and expensive subscriptions. DeepSeek, with an MIT license and 552 billion activated parameters (763 billion total), proves the opposite: agility and utility beat abstract academic metrics. With the minimum price of intelligence flattened to zero, billion-dollar revenue models based on API fees become precarious. The Genie Out of the Bottle This crisis can't be solved with acquisition or prohibition. The weights of V4.1-Flash-1017 are under the MIT license: any cloud, any lab with 8 H200 GPUs, can deploy it. Pandora's box has already been opened. DeepSeek even optimized compatibility with frameworks like DeepSeek Harness and Tencent CodeBuddy — putting down roots in the open-source ecosystem. Silicon Valley faces a prisoner's dilemma: entering the price war destroys the high-margin model; keeping prices high hands customers to the Chinese model that offers more for a fraction of the cost. DeepSeek didn't just win the tests that matter — it did something more fatal: it made the opponents' victory economically worthless. When the best model and the cheapest model are the same thing, selling "intelligence" is no longer a business. And DeepSeek wrote the final chapter of that story.

Bitcoin Doesn't Follow the S&P 500 — It Shouts First (Every Rule Has Exceptions) The market that never sleeps is sending the message, and the stock exchanges only confirm it when they open. There's a comfortable myth circulating among investors: that Bitcoin is a baby of the S&P 500, that it obeys the Nasdaq, that it only moves after Wall Street sets the tone. It's a lazy — and wrong — reading. The truth is more uncomfortable: at various moments, Bitcoin moves first. And traditional markets merely confirm the move hours later, when they finally open. The reason is structural, not circumstantial. Crypto operates 24 hours a day, seven days a week. There's no overnight close, no weekend, no holiday. Stock exchanges have hours. Futures have limited ability to discover price outside the main window. The result: when relevant information explodes overnight or on a Sunday, global capital needs a liquid place to express its read. Bitcoin is that place. The sequence is well known: new information → Bitcoin reacts → futures confirm → exchanges open adjusting prices. Those who only look at the daily chart conclude that Bitcoin "predicted" the market. It didn't predict. It priced first. This is the concept of price discovery. Markets exist to transform information into price. When they're closed, they can't. Bitcoin doesn't stop. It never stops listening to the world. But be careful: this doesn't mean every Bitcoin drop anticipates an S&P 500 drop. Quite the opposite. Leveraged liquidations, bankruptcies, regulatory problems, whale movements, stablecoin collapses — internal events drag Bitcoin down while stocks remain indifferent. The relevant signal isn't "Bitcoin fell." The right question is: why did it fall — and are other markets confirming? When Bitcoin, commodities, currencies, interest rates, and futures move coherently, the message is bigger: the global price of risk has changed. That's why the correlation between Bitcoin and the S&P 500 varies so much. In regimes of abundant liquidity and low interest rates, Bitcoin behaves like an ultra-high-beta asset, glued to speculative tech stocks. In generalized panic, it aligns with stocks because everyone cuts risk. At other times, internal factors dominate and the correlation simply disappears. The relationship isn't fixed. It's conditional on the regime. So stop repeating that "Bitcoin follows the S&P." The correct formulation is different: Bitcoin and stocks respond to the same forces — but crypto's uninterrupted structure allows Bitcoin to do price discovery first. Think of Bitcoin not just as an asset, but as a sensor. Global, liquid, continuous, hypersensitive to risk perception. A noisy sensor that produces false signals, suffers from leverage and speculation. But one that, precisely because it never closes, has something no other market has: it never completely stops listening to the world. Sometimes, stocks give the signal and Bitcoin follows. At other times, Bitcoin lights the alert during the night — and stocks merely confirm the message when they open a few hours later. Those who understand this asymmetry don't chase Bitcoin. They use Bitcoin to know what comes first.

Intelligence is the ability to anticipate what happens next in a way that reality confirms — not the skill of articulating ideas with sophistication. The world is full of verbally brilliant people who are consistently wrong. They craft elegant narratives, cite complex concepts, and sound deep in talks and posts. But when the real test comes — what happens next quarter, in the market, in politics, in the company, in life — reality proves them wrong. Verbal sophistication is cheap. It impresses rooms and feeds. Accurate prediction, on the other hand, is expensive: it requires direct contact with reality, the willingness to be proven wrong, and the humility to update your model when the facts don’t cooperate. The real test of intelligence isn’t “How well do you argue?” It’s “What did you bet would happen — and did it?” Those who only know how to articulate well live off reputation. Those who anticipate well live off results. The difference, over the long run, is brutal.

Trump says: “America First.” Putin shouts: “RUSSIA FIRST.” Trump believes he can end the war by buying diesel (and energy) from Russia. The reality is darker: the Russian economy has been reconfigured to depend on the war. Stopping (or drastically reducing) the conflict would trigger, all at once, three shocks the Kremlin considers more dangerous in the short term than the cost of continuing. 1. Collapse of demand in the military-industrial complex Companies in the sector live off long-term state orders (Putin himself has spoken of order backlogs of 5 to 10 years). There are already signs of contraction even with the war ongoing. Without orders come defaults, layoffs, and cascading bankruptcies. The “military Keynesianism” that sustained growth in 2023-24 has already run out (GDP of ~4.1% in 2024 down to ~1% in 2025; 2026 projections at 0-1%). 2. Social return of veterans and the income gap There are still hundreds of thousands of combatants at the front. Frontline pay is 2-3x civilian wages; death or disability compensation equals the price of an apartment. Mass demobilization of an army sustained by economic incentives (not classic conscription) creates an income shock, PTSD, and risk of instability. In 2025, with most still mobilized, there were already more than 1,000 homicide cases involving veterans. Extending contracts and keeping troops in military settlements is seen as the “more manageable” option. 3. Regions and elites dependent on war income Poor areas and military-industrial hubs (Tula, Samara, Buryatia, etc.) experienced a boom in income, consumption, and local finances thanks to bonuses and contracts. They form a political support base. Cutting spending means severe contraction exactly where the regime needs stability. Among the elites, the “commoditarians” prefer sanctions relief, but the statists (Belousov line) gain institutional power with continued militarization. The Defense Ministry itself argues that cutting defense would hurt growth because the economy already “depends on the war as its engine.” Classic path dependence. The buffers are running out (liquid assets of the Sovereign Wealth Fund fell from 6.5% to 1.8% of GDP; record bankruptcies), but the short-term political calculation favors continuing: the costs of exit still look higher than the marginal cost of prolonging. This is not about Putin’s “personal will.” It is about an economic machine and a set of interests redesigned to use the war as fuel. “Russia First,” in practice, means keeping the war running. Buying diesel does not dismantle that structure. Main sources: IFRI (Inozemtsev/Laruelle), Kiel Institute, KSE Institute, IISS, Allianz Trade, ERR, El Economista, Statecraft & Strategy, Friedrich-Ebert-Stiftung (2025-2026). @realDonaldTrump @KSE_Institute @kielinstitute @MarioNawfal @IISS_org @IFRI_ @Vladi_Moscou @elinaribakova @thebell_io @sanctionsgroup