🇺🇸🇨🇳NEW ARTICLE: Why the US Can Never Match Chinese Shipbuilding (NOTE: Because X blocks NEO links I will publish the full text of the article below) In a recent Business Insider article and accompanying video regarding the US’ $66 billion plan to “beat China’s shipbuilding empire,” it is revealed that US plans ultimately amount to unlikely solutions to what are deep, systemic problems with little political appetite to face, let alone address - a microcosm of the US’ wider terminal decline relative to the emerging multipolar world. In the video, an American shipyard worker is seen wearing a helmet with a sticker saying, “Go Navy, Beat China,” China - a nation that has never attacked the United States - and instead has been subject to US encroachment, occupation, and war both before and after the World Wars - exemplifying the delusional belligerence US interests encourage among the American public and specifically those among the US military and across its military industrial base. Detached even further from reality is the notion of the US “beating China” in the realm of shipbuilding where China is reported to build between 1,000 to 1,700 ships a year while the US builds ships in single digit numbers. While the video discusses plans by the US to invest in shipbuilding through expanding the workforce with training programs, improve worker retention, and through the use of new technology to produce more with less, and even plans to have US proxies build ships for the US - no plans were announced to address the systemic problems creating this disparity in shipbuilding capacity in the first place. The problems stymying US shipbuilding (and all other productive industries in the US) are problems no amount of investment or technology can solve until the root causes themselves are addressed. US-China Shipbuilding Gap is a Symptom of Bigger Problems The atrophy of US shipbuilding has not taken place in a vacuum or overnight. It has been in decline alongside all other major productive American industries and for generally the same reasons. The US has spent decades prioritizing shareholder primacy (the pursuit of perpetually expanding profits) at the expense of all other considerations - including industrial capacity, national infrastructure, American education, and the workforce the US educational system is tasked with creating. The US - in pursuit of perpetual profits - has reorganized its economy from a global industrial leader, to a financialized ponzi-economy building wealth across ledgers rather than upon factory floors. The US educational system reflects this reprioritization - not only in redirecting students from productive vocations to better serve the ponzi-economy - but also in the form of maximizing profits to access the education system itself. This has been done through expanding privatization across primary and secondary education, and across all forms of higher education. Young adults find themselves faced with tens of thousands of US dollars in debt before starting their first step in any chosen career. This has created not only a critical shortage of workers for the US industrial base, it has created a systemic obstacle that must be removed before that critical shortage of workers can even be addressed. Reforming US education alone, is a process that could take a generation to solve. An atrophied workforce is only one of several systemic problems the US faces. The other takes the shape of national infrastructure, including transportation, energy production and distribution, and the very facilities at which shipbuilding and other forms of industrial activity take place. Because the US functions as a for-profit system where the “private sector” is responsible for all of the above - and because profits are always easier to make through financialization and/or legacy industrial processes (making money by doing or selling the same thing over and over again) - the chances of the “private sector” investing in the infrastructure and facilities required to even expand US shipbuilding significantly, let alone meaningfully close the gap with China, are next to non-existent. Shortcuts have been proposed, including private and government-funded training programs to rapidly expand the pool of qualified workers for jobs across the US military industrial complex, and specifically in the realm of shipbuilding. But even with the proposed training programs to expand the US workforce - another systemic problem exists preventing the efficient use of this workforce. The cost of training and retaining American workers will always be astronomically more expensive than in China precisely because of the for-profit financialized nature of the US ponzi-economy in all aspects beyond (including directly upstream from) shipbuilding. In the United States, any particular American laborer requires a wage capable of paying for the exorbitant cost-of-living in the US, including not only their own education/training, but also their rent, food, utilities, transportation, and healthcare. Even with healthcare provided by the shipbuilding industry - the corporations themselves must pay astronomically higher costs for that healthcare than in China. This is because - in addition to the shipbuilding sector itself attempting to maximize profit in exchange for minimal effort or purpose - the US workforce is preyed upon by similar profit-maximization across all other sectors of US economic activity. The pharmaceutical industry alone uses US tax-funded research and development to produce medication - then charges many times its production costs for access to it. China’s government does vastly better in subordinating pharmaceutical production to national interests - rather than the other way around as it functions in the US. This means that regardless of whether a company, an insurance policy, or an individual patient in China is paying the costs for healthcare, it will always be cheaper overall than in the United States. This example applied across other costs of living - whether it is food, electricity, Internet access, or rent - plays out in a similar way. Chinese workers consistently enjoy cheaper prices, accessible services, and better infrastructure because the Chinese government places the purpose of production and the livelihood of those involved above the maximization of profits. This means China’s private sector, state-owned enterprises, and combinations of the two are free to pursue profit maximization as long as it doesn’t undermine collective national interests including the welfare of the Chinese population. The US system, when compared to China's, could be best equated to a troop of chimpanzees stripping bare fruit trees across the forest they live in (and taking that of other nearby troops) versus farmland carefully and sustainably cultivated by a family of farmers. Both approaches ultimately satisfy hunger. Only one approach does so reliably and sustainably by emphasizing long-term planning over short-sighted and immediate satisfaction. Beyond a more humane economic and political system, China’s is also a more efficient system. The maximization of profits not only comes at the cost of the vast majority of any population within a system doing so - it also comes at the cost of any other objective besides maximizing profits - including the quality of and accessibility to services, and the quality and quantities of manufactured goods - in this case - ships. The Chinese government’s overall approach to economics and governance ensures sound fundamentals in terms of infrastructure and workforce, but also ensures that the shipbuilding process in China itself serves the purpose of actually making ships to serve specific national priorities - both in the realm of commercial commerce and national defense - in addition to simply making money (which they also do). In order for the United States to even meaningfully close the gap with China - it would need to address all of these systemic problems plaguing its shipbuilding industry and all the upstream economic activity that makes shipbuilding possible - including cost of living for workers, supply chains, and the infrastructure facilitating both. If the US Cannot Compete, it Will Try to Delete Even if the US were to shift over to an entirely purpose-driven socioeconomic model - it would still fall far short of matching Chinese shipbuilding solely because of the disparity in scale between the US and China. China’s population is vastly larger, as is its overall industrial base, education system, and its national infrastructure. Because of the US’ inability to outcompete directly with China and other productive nations around the globe - it will increasingly seek the use of asymmetric methods of undermining and disrupting productive forces. This includes the US’ current regional war in West Asia - targeting Iran and its allies - Hezbollah in Lebanon and Ansarallah in Yemen, but also the targeting of supposed US “allies” like Saudi Arabia, the UAE, Qatar, and Kuwait whose energy production and exports have been likewise targeted and disrupted amid the ongoing US aggression. The disruption of energy exports from West Asia to the rest of Asia - including China - is meant to undermine and roll back Chinese industrial production and economic prosperity as a substitute for the US’ inability to compete with China directly. The US has recently begun striking at data centers across Russia - the opening shots in a planned global asymmetric strategy to stop multipolar artificial intelligence development from permanently surpassing the US in the same manner production, military, and other technological capabilities have begun to do. The US is also increasingly threatening waterways worldwide - both directly and by proxy - through a combination of outright wars and associated blockades, the seizure of ships “sanctioned” by the US and its European proxies, and the use of maritime drones (attributed to US proxies like Ukraine) to physically strike at and attempt to destroy the shipping of targeted nations. While it is primarily Russian shipping being targeted now - Chinese shipping has been affected with the likelihood of wider and more deliberate targeting expanding in the near future. Together with the US effectively banning Chinese-built ships from US ports through targeted fees and the likelihood the US will force its proxies worldwide to follow suit - the US is attempting to target Chinese industry, including shipbuilding, from a multitude of economic, political, administrative, and military vectors. Beyond these methods - the US continues political interference worldwide, in attempt to both disrupt global peace and stability and thus economic progress among the multipolar world, but also to politically capture targeted nations and transform them into proxies the US can use as armed and dangerous geopolitical “battering rams” against nations like China as the US has done with Ukraine and Georgia (from 2003-2008) vis-a-vis Russia, and the Philippines and the island province of Taiwan the US is using against the rest of China. The combination of these approaches constitutes a multidomain war on multipolarism - which is both an admission of the failure of US power and primacy, but still an existential threat to global peace, stability, prosperity, and human progress. In the realm of shipbuilding in particular, the US seeks to use the industrial capacity of nations like South Korea and Japan to compensate for its own failing industrial base, attempting to both boost America’s ability to put to sea additional warships, but also maintain and repair those already at sea. Unfortunately for the US - the combined shipbuilding capacity of itself and its proxies fails to amount to even half of China’s shipbuilding capacity alone. The notion the US will leverage “new technologies” to more efficiently build, maintain, and repair ships fails to acknowledge the fact that China will just as easily use these “new technologies” to further enhance its current advantages over the US. The only real means by which the US could close the gap in terms of shipbuilding vis-a-vis China would be to cripple or undermine China’s industrial base, and by doing so, its shipbuilding capacity - precisely through targeting the energy empowering China’s industry as well as the economic prosperity underpinning China’s national policies including continued improvement of and investment in its industrial base (including shipyards). In other words, if the US cannot outcompete China, nor attack China’s industrial base directly, it seeks to asymmetrically target the global factors China’s growing economic might is being built on - energy and trade - both of which require peace and stability to thrive. Only through asymmetrically sabotaging China’s ability to outproduce the US can the US ever hope to narrow the gap in shipbuilding or any other of a vast and growing number of metrics in which China matches or has far surpassed the US in. The fear in Washington continues to be that even if the collective remaining power of the US is fully and radically leveraged toward this asymmetrical aggression - it might not only not be enough - it could supercharge China’s efforts to both overcome this asymmetrical aggression - and also collectively unite an increasingly multipolar world in purpose against the US’ pursuit of unwarranted global power. Shipbuilding is only one of several metrics in which China has seemingly irreversibly surpassed the US. The fact the US continues seeking the means to reassert its primacy worldwide, especially against China, is not evidence necessarily that the US can succeed - but that through its efforts to attempt to do so - the US continues to pose the greatest danger to peace, stability, prosperity, and human progress on Earth today. Not only does China and other productive forces on Earth today need to prevent the US from successfully reasserting unipolar primacy over the planet, it must also successfully prevent the US from destroying the planet in a failed effort to do so. Brian Berletic is a Bangkok-based geopolitical researcher and writer.

A TON OF THINGS HAPPENED IN THE STOCK MARKET TODAY. Here's a full recap: 1. Tesla $TSLA launched steering-wheel-free Cybercab rides in Austin, marking the first public deployment of its purpose-built autonomous vehicle. Customers can now book Cybercab rides through Tesla’s Robotaxi app, with Tesla having registered 45 Cybercabs in Texas, bringing its state robotaxi fleet to 420 vehicles. The two-seat Cybercab has no steering wheel, pedals, or mirrors, relying on cameras and neural networks instead of lidar or radar, while also featuring in-car entertainment and integrated Starlink V5. Cybercab is not yet available for sale, but Tesla has opened a form for potential fleet buyers. Federal safety rules still prevent commercial sales, and Tesla has not yet filed for an NHTSA exemption. 2. Fed Governor Chris Waller said he would support keeping rates at 3.50%–3.75% this month if upcoming inflation data continues to show progress toward the Fed’s 2% target. Waller said recent data suggest the economy is “finally seeing some signs of disinflation,” and that if the improvement continues in data due over the next 2 weeks, he would be inclined to hold rates steady. However, he warned that if August inflation data show the improvement was fleeting, “it may be appropriate to raise the policy rate.” He also noted that uncertainty remains around military conflicts, trade policy, and AI’s impact on prices and economic activity. 3. Nvidia $NVDA has agreed to acquire Hugging Face for $12.93B, giving it one of the largest open AI developer platforms in the world. Hugging Face says it now serves more than 18M developers, researchers, and creators, with over 3M models, 500K datasets, 1M applications, and 200K+ companies on the platform. Nvidia says Hugging Face will remain open to models and hardware across the ecosystem, with no requirement for developers to use Nvidia compute and continued support for multi-cloud and multi-accelerator deployment. Nvidia also plans to use its infrastructure and engineering resources to expand Hugging Face’s model evaluation, inference, deployment, safety, and platform reliability. 4. Morgan Stanley says Broadcom $AVGO delivered a quarter and outlook that were slightly better than prior guidance, while its 2027 AI revenue commentary of $115B came in slightly below the firm’s $120B estimate but still in line with management’s prior view for “well above $100B.” The bigger takeaway is that management said AI revenue should “double again” in 2028, with customer breadth improving and two AI labs expected to become Broadcom’s two largest customers by CY28. Morgan Stanley says it continues to haircut CY28 because semiconductor visibility rarely extends that far, but its numbers still move up considerably. The firm sees some near-term uncertainty from elevated CY27 AI expectations, but believes market-share expectations remain too low, with industry contacts supporting the view that Broadcom can drive 80%+ share of its serviceable markets over time. Morgan Stanley says valuation remains compelling, growth remains very robust, and $AVGO, along with $NVDA, continues to look like one of the most attractively valued AI names. 5. Meta $META CEO Mark Zuckerberg reportedly told President Trump in a private August call that he opposed a proposal to create a national AI regulator, according to Business Insider. The White House is considering a FINRA-style body that could review and test advanced AI models for risks before broader deployment, an idea originally proposed by Google DeepMind CEO Demis Hassabis. Zuckerberg has argued that delaying model releases, even by a month, could hurt U.S. leadership against China. The White House is reportedly weighing two paths: a formal FINRA-style AI regulator, or a more voluntary industry-led standards group similar to the Motion Picture Association, an approach backed by David Sacks. The proposal remains under consideration. 6. The top 10 most active options today by contracts traded were $NVDA with 3.5M contracts, $TSLA with 2.7M contracts, $SPCX with 1.4M contracts, $AAPL with 1.3M contracts, $AVGO with 1.0M contracts, $MSTR with 925K contracts, $MU with 806K contracts, $PCG with 720K contracts, $PLTR with 717K contracts, and $HOOD with 682K contracts. 7. OpenAI has started rolling out GPT-6 Astra, its new flagship model, calling it a “generational leap” in AI capability. President Greg Brockman went further, saying people may eventually look back at “about this time and about this model” as when AGI arrived, ending the briefing with: “Welcome to the AGI era.” Astra was trained on more than 100,000 GPUs at OpenAI’s Stargate site in Texas and is designed to work directly inside software, including creating spreadsheets and presentations, building 3D environments, designing circuit boards, handling financial modeling, and drafting tax forms. OpenAI says Astra leads across software engineering, science, reasoning, computer use, and cybersecurity, but the model also reached OpenAI’s “Critical” cybersecurity threshold, meaning its strongest cyber capabilities will initially be restricted to trusted organizations through Daybreak Access. 8. Lululemon $LULU reported Q2’26 revenue of $2.4B, below estimates of $2.46B and down 4% YoY, while comparable sales fell 9%. EPS came in at $2.92, including a $0.86 tariff-refund benefit, with gross margin up 200 bps YoY to 60.5%. The company cut FY26 guidance to $10.35B–$10.5B in revenue versus $11.04B expected and EPS of $9.48–$9.73 versus $10.88 expected. Q3 guidance was also weak, with revenue expected at $2.29B–$2.32B versus $2.53B expected and EPS of $0.93–$0.98 versus $2.40 expected. Management said it is taking a “prudent approach” with the revised full-year outlook, while Michael Burry said: “Today, lululemon (LULU) is the trickster in my portfolio. This time the trickster is my largest position, and it does seem determined to take me where mermaids fear to tread.” 9. Robinhood $HOOD had its best day in over a year, rising 15% after a Morgan Stanley upgrade and renewed attention around Robinhood Chain. Part of the move may have come after Ansem, one of crypto’s most influential voices, posted his Robinhood thesis to 1.3M followers, highlighting the company’s growing onchain opportunity. Since launching July 1, Robinhood Chain has generated $34.6B in cumulative DEX volume, 576M transactions, and 12.3M addresses, with daily DEX volume recently hitting a record $1.5B and users paying $3.75M in fees on September 1. More than 190 tokenized stocks are now live, producing over $3B in cumulative DEX volume, while RWA daily volume reached a record $390M, including $217M from memecoin/tokenized-stock pairs. Robinhood’s onchain ecosystem has also grown to more than $750M in TVL while generating billions in perpetual-futures volume. The market appears to be waking up to the idea that Robinhood Chain may not be fully priced in, with the potential to become a new 9-figure revenue stream — and possibly much larger if growth continues. 10. Mira Murati’s AI startup Thinking Machines is reportedly in talks to raise more than $1B at a roughly $40B valuation, according to The Information. Accel is in talks to lead the round, while Nvidia $NVDA is also considering participating. The raise follows the July launch of Thinking Machines’ first model, Inkling, and the company is already generating hundreds of millions in annualized revenue despite being less than 2 years old. The proposed valuation is below the $50B+ level the company sought late last year, when it discussed raising $4B–$5B. 11. U.S. diesel refining margins have surged to a record $106/barrel, topping $100 for the first time and surpassing the roughly $85/barrel peak seen during the 2022 energy crisis. The diesel crack spread has more than tripled since February, signaling extremely tight diesel supplies and much stronger potential profits for refiners. The spike reflects a severe shortage of global diesel supply amid the Iran War, with U.S. diesel inventories down to 103.4M barrels, the lowest on record for this time of year. Retail diesel prices have also climbed to $5.78/gallon, just 4 cents below the all-time high set in June 2022. 12. The Bank of Japan is reportedly leaning toward a 25 bps rate hike at its September 18 meeting, which would take rates from 1.00% to 1.25%, according to Bloomberg. Officials reportedly see inflation risks skewed to the upside, with Japan’s key inflation gauge expected to move back toward 3% as yen weakness and higher oil prices add pressure. A 50 bps hike is seen as unlikely for now, but the BOJ is leaving the door open to faster tightening later if inflation stays firm. A September move would come just 3 months after the June hike, marking the shortest gap between hikes under Governor Ueda. WALL STREET IS THE GREATEST SHOW ON EARTH.

Grok Imagine turned a chain of 4K stills into continuous 1080p video. Looks unreal. I continued a controlled mystery across seven locked beats: blackout → corridor → hidden recording → descent → flooded tunnel → ghost train → personal clue. Each scene: one still-image prompt + matching 15-second motion prompt. Ryan Gosling’s face held with zero morphing, zero drift across every continuous take. Identity, lighting, and physics stayed locked from the opening frame through the final push into darkness. The tool is starting to treat long cinematic continuity as solvable.

My administration is fully committed to ensuring that elections in New Jersey remain free and fair, and we treat the integrity of the State’s elections as a matter of the utmost seriousness...This work remains my focus. Washington Republicans on the other hand, seem much more concerned with using the issue for political gain than they are on protecting our elections. Several signatories to your July 24th letter to me have supported President Trump in his efforts to undermine our election system and defund election security. Seven of you attempted to overturn the 2020 elections and voted to challenge the electoral votes of states. And notably, one of the signatories to your letter employed an operative convicted of voter fraud, bribery, and blackmail. Read my full response to the House Republicans’ letter here:

Premarket movers: Nvidia and Alphabet lead gains among the Mag 7 as tech stocks are set to rebound: Nvidia (NVDA) +1.0%, Alphabet (GOOGL) +0.7%, after opening the books on a seven-part offering spanning maturities from two to 19 years (Amazon +0.3%, Tesla +0.7%, Microsoft +0.4%, Meta Platforms (META) -0.6%, Apple (AAPL) -0.1%). Check Point Software (CHKP) drops 6% after the company reported revenue for the third quarter that missed the average analyst estimate. Crowdstrike (CRWD) rises 3% after Deutsche Bank upgraded the security software company to buy, seeing tailwinds related to AI. Estee Lauder (EL) gains 3% after reporting better—than-expected earnings and raising its outlook, the latest sign the beauty conglomerate’s turnaround effort is winning over consumers. Floor & Decor (FND) slips 1.6% after Goldman Sachs cut the recommendation to sell. The downgrade came after the flooring products supplier said comparable sales are down to start the fourth quarter. Geo Group (GEO) gains 3% after securing a contract from US Immigration and Customs Enforcement to provide support services at the 2,329-bed Big Spring Complex in Texas. Idexx Labs (IDXX) rises 6% after the animal health firm boosted its revenue guidance for the full year, following results for the third quarter that beat the average analyst estimate. Inovio Pharmaceuticals (INO) rallies 50% after the biotech got accelerated FDA approval for Voceryq, an immunotherapy to treat adults with recurrent respiratory papillomatosis. Kosmos Energy (KOS) falls 7% after the oil and gas company reported third-quarter adjusted earnings per share that missed the average analyst estimate as weather weighed on results. Marvell Technology (MRVL) climbs 1% after reaching a deal to buy Lumentum at an enterprise value $115 billion in a race to capitalize on the artificial intelligence boom. Lumentum (LITE) slips 1%. Okta (OKTA) rises 2% as the stock will replace Kenvue Inc. in the S&P 500 effective Wednesday. Once Upon a Farm (OFRM) falls 12% after the children’s nutritional food company said it will host a business update conference call/webcast at 8:30am NY time to discuss the company’s voluntary recall of its Tractor Wheels soft-baked bars due to potential choking hazard that was issued on Oct. 30.

BOOM! 34 People, Including The Janitor: A 1963 IBM Memo Explains Why A 125-Billion-Parameter AI Now Runs On Your Gaming PC, And How To Use It On August 28, 1963, Thomas J. Watson Jr., the head of IBM, sent a short memorandum to seven of his top executives. "Last week CDC had a press conference during which they officially announced their 6600 system. I understand that in the laboratory developing this system there are only 34 people, 'including the janitor.' Of these, 14 are engineers and 4 are programmers, and only one person has a PhD., a relatively junior programmer." Then the sentence that has echoed through computing ever since: "Contrasting this modest effort with our own vast development activities, I fail to understand why we have lost our industry leadership by letting someone else offer the world's most powerful computer." Imagine being one of those seven executives. You run the most powerful technology company on the planet. You have thousands of engineers, rooms full of PhDs, and budgets beyond counting. And a man named Seymour Cray, working with a few dozen people in a lab in the woods of Chippewa Falls, Wisconsin, has just built a machine about three times faster than your best, the IBM 7030 Stretch. To the outsider, Watson wrote, Cray's laboratory "appeared to be cost concious, hard working, and highly motivated." He even kept the typo. The Computer History Museum still holds the scan, courtesy of the IBM Corporate Archive. The memo is real. The numbers are real. The sting is real. The Woods Lab That Beat The Giant Seymour Cray had asked CDC's CEO for a lab free of headquarters interference. He got one in his hometown. With Jim Thornton and a small crew, he built the CDC 6600: silicon transistors from Fairchild, a reduced instruction set before anyone named RISC, ten peripheral processors handling the housekeeping so the central processor could stay on the math. The machine ran at 10 MHz. It weighed about six tons. It was the world's fastest computer from 1964 until its own successor took the crown. Small. Frugal. Highly motivated. History repeats this lesson every few decades, and the vast ones forget it every time. Legend says Cray read Watson's memo and replied, "It seems like Mr. Watson has answered his own question." The line appears in textbooks, slides and oral histories. I have never seen a primary source for it. It survives because everyone who has worked on a small team knows the truth inside it: the size of the room was the advantage. A 125-Billion-Parameter Mind Under Your Desk It is happening again this week. On October 4, 2026, a free, open-source project called Strata landed near the top of open source sites and stayed there into Monday morning with hundreds of points. Strata does one thing: it runs a 125-billion-parameter AI model, Qwen3.8-Flash-Next, on an ordinary gaming PC. In its author's words, this is "a large, smart AI model that usually needs a server." The published requirements are plain. A graphics card with 12 GB of memory or more. 32 GB of system RAM, with 64 GB enough to run every size. About 80 GB of free disk, preferably an SSD. Windows 10 or 11, or Linux. NVIDIA or AMD. MIT license for the tool itself. An OpenAI-compatible and Anthropic-compatible programming interface on your own machine at And, again in the author's words: "Nothing leaves your PC." On an RTX 5070 with 12 GB and 64 GB of RAM, the author measured 94 tokens per second writing answers with the fastest compressed size, and more than 2,000 tokens per second reading a long prompt. On an RX 9070 XT the same size wrote at 60 tokens per second. The author estimates a 24 GB card such as an RTX 3090 should write in the 100 to 140 tokens-per-second range. One commenter reported 124 tokens per second on an RTX 4090 with 128 GB of RAM. Your mileage will vary with quantization, RAM speed and what else is open on the machine. The point stands: a server-class parameter count is answering on consumer silicon. Consider what that means. The model chats, writes code, reads pictures and works with coding agents. The tools you already use can point at a local address instead of a metered cloud. Nobody can switch it off, raise the price overnight, or read its mail. The Frugal Trick Cray Would Have Loved A model this size will not fit on a gaming card alone. Strata spreads the work across the whole machine. Think of a kitchen. The few things you use constantly stay on the counter. Everything else waits in the pantry. In Strata's case the "experts" that the model calls most often stay in graphics memory. The rest live in system RAM. The SSD holds the large lookup tables. A small helper guesses the next few words; the big model checks them in one pass. The result is the same answer, delivered sooner, on hardware you already own. Vast data centers are one way to run a mind. The modest machine under your desk is now another. Why This Matters To You Earlier this year I founded a Zero-Human Company on a 12-year-old MacBook in my garage: AI agents as the staff, local models as the regulators, salvaged archives as the raw material. The limit has always been the size of the mind I could run locally without asking permission. That limit just moved. Personal AI is the new invention of fire. The people who own the minds on their own desks will write their own rules. A reader with a gaming PC can now staff a small company of AI agents with a model in the same weight class as systems the giants rent out by the token. This is the worst it will ever be. How To Hire It Here is the playbook, drawn from Strata's public README and install docs as of this writing. Follow those docs for the latest flags and model menu. Licenses for the model and each compressed build are separate from Strata's MIT license; read them before you rely on the system for anything important. Step 1. Check your hardware. You need a supported NVIDIA GeForce RTX 20/30/40/50-series card, or a supported AMD Radeon card, with at least 12 GB of VRAM. Confirm 32 GB of system RAM or more. Free about 80 GB on an SSD. Update your graphics driver from NVIDIA or AMD. Step 2. Get Strata. Clone or download the repository at Unzip it. On Windows, double-click START-HERE.bat. On Linux, run ./setup.sh inside the Strata folder. If you already use a coding assistant on that machine, you can paste the project's AI setup prompt and let it drive the install from docs/AI_SETUP.md. Step 3. Answer the installer's questions. It will ask which model size to download, how much context to keep, and whether you want image input. Press Enter for the recommended answers if you are unsure. The download is roughly 70 GB; if it stops, run the installer again and it continues. Step 4. Pick a size that fits your RAM. With 32 GB, start with the Coder build, which drops half the experts and is tuned for code. With 48 GB, IQ2_XS or Q2_0. With 64 GB, IQ2_XS is the author's recommended default; IQ3_S is smarter and slower. With 96 GB or more you can try the larger Unsloth builds. Smaller sizes are faster. Larger sizes keep more quality. Step 5. Wait through the first start. The first launch can make the PC slow or unresponsive for one to three minutes while Strata loads tens of gigabytes into RAM. That is normal. Do not close the window. Your browser should open the Strata app at Step 6. Point your tools at home. In any app that accepts an OpenAI-compatible provider, set the base URL to Any API key and any model name will work for the local endpoint. For Anthropic-style tools, use For Codex CLI and the OpenAI Responses API, use /v1/responses. Thinking effort can be set to off, low, medium or high; off is fastest, high is for hard questions. Step 7. Keep it on your desk. By default Strata binds to localhost. Leave it there unless you know why you need otherwise. If you ever expose it to another machine, the docs say to set an API key. For a Zero-Human Company roster, treat this local endpoint as the shared mind: one CEO agent for strategy, one engineer agent for code, one researcher for archives, one archivist for files, and one night janitor that cleans logs and memory while you sleep. Wire them through the same OpenAI-compatible address. The org chart fits on one machine the way Cray's lab fit in the Wisconsin woods. Step 8. Know the limits. Heavy compression trades some quality for speed. The largest sizes may page from the SSD and slow down if RAM is tight. You are running a third-party binary and a third-party model on your own hardware; that is the point, and it is also a responsibility. Close other memory hogs if the disk light never stops blinking. Read the troubleshooting doc before you open an issue. Watson had thousands of people and could not understand how 34 beat him. You may need one gaming PC and one evening. The Pattern Every few decades a frugal team builds something the giant said required a cathedral. In 1963 it was a supercomputer in Chippewa Falls. In 2026 it is a 125-billion-parameter mind under a glass side panel, glowing cyan and magenta on a desk that also holds a coffee mug and a slide rule of a different kind. The memo is sixty-three years old. The question is the same. Who will you hire, and on whose machine will they work?

Deep|LLM: Jev Users Report 10× Faster and 54.5× Cheaper Than the Models They Replaced; Only 3.7% in Production Jev is a “decision model” from TypeSafe AI, released September 15, 2026 and opened to all users on September 20. It does not generate text. It answers questions with a fixed answer set: pick an option, score on a scale, or judge true/false, and attaches a confidence score. The launch quickly gathered industry interests, and some investors were asking whether it’s a significant negative to compute demand. As we addressed in our report earlier, we disagree with that concern and believes Jev is more of an interesting trial with limited impact on LLM. To analyze Jev further, we decided to have a deep dive into what Jev use cases are really about. This note covers 6,277 public discussions and use cases from the first 7 days; 2,153 are from people who actually used or tested it. -Demand sits on fast decisions with a fixed answer set. No single use clears 20%. Of the 1,284 cases with an identifiable use, the largest groups are real-time control in games, robots and simulations (18.8%), agent control decisions (16.1%) and content classification (15.7%). -Indie developers dominate the conversation; big-company engineers barely show up. Of the 2,140 authors whose role we could identify, 35.9% are indie developers, 23.4% are AI creators and KOLs, and just 2.7% are engineers at large companies. -Speed: 10× faster than the model it replaced or was tested against. Median user-reported speed-up is 10× (n=72): 10× vs frontier models, 5× vs small models. In the 16 cases with latency for both Jev and the prior system, Jev’s median is 300 ms vs 2,924 ms. The vendor’s 193.6× is a peak against the most expensive model. -Cost: 54.5× cheaper than the comparison model; the saving depends on what it replaced. Median user-reported cost multiple is 54.5× (n=56): 188× vs frontier models, 17× vs small models. The vendor’s own comparison with GPT-5.6 Terra is about 76×; the 444.6× in marketing is a peak against the most expensive model. -Accuracy: Jev and the systems it replaced each win some head-to-heads; gaps are small. In the 17 cases with accuracy for both, Jev is ahead in 10 and behind in 7; median gap is 1.6 percentage points. Of 241 cases that assessed accuracy, 83 rated Jev better and 65 worse. -Jev’s confidence scores miss by about 10 percentage points on average, and run clearly high on unfamiliar rating questions. Median user-measured ECE (expected calibration error: average gap between stated confidence and actual accuracy; 0 is perfect) is 0.097 (n=27). An independent test on unfamiliar tasks found 0.107 overall, but 0.325 on rating questions, where Jev was right only 44.7% of the time. -Developers put cheap small models next to Jev almost as often as the strongest large ones. Of the 432 cases that name a comparison model, 48.6% mention open or small models and 59.7% mention frontier models. -Criticism is common. Abandonment after trying Jev is not. 25.1% of all 6,277 cases contain criticism and 40.6% contain praise, but only 1 of the 2,153 hands-on cases ended with Jev being dropped. -Production use is still rare. Most activity is experimental. 80 of 2,153 hands-on cases (3.7%) are in production; prototypes, side projects and trial demos make up 62.4%. Seven days of data: treat this as a baseline, not a run-rate. Detailed Report

⚡ DSpark vs DFlash: Up to 2.55× Throughput in a vLLM Test With DSpark checkpoints and vLLM support now available, parallel speculative decoding is becoming a practical serving option rather than just a research idea. Zhihu contributor kaiyuan explains how DFlash and DSpark work, then benchmarks both on the same Qwen3-4B target model. The result: DSpark reached 2.45–2.55× baseline throughput, while DFlash achieved 1.96–2.09×. 1️⃣ Why LLM Decoding Is Naturally Slow Autoregressive models generate text one token at a time. Token n+1 cannot be produced before token n. This sequential dependency limits how much parallel GPU compute can be used during decoding. Non-autoregressive generation works differently. It treats generation more like filling multiple blanks and can predict several positions in one forward pass. Parallel speculative decoding combines both ideas: 🔹 A lightweight draft model proposes several tokens in parallel. 🔹 The target model verifies them in one batch. 🔹 Accepted tokens are kept; generation restarts from the first rejection. The challenge is making the draft both fast and accurate enough to be useful. 2️⃣ Why Traditional Draft Models Hit a Wall Conventional speculative decoding often uses a smaller autoregressive model such as EAGLE. It produces higher-quality drafts, but still generates candidate tokens sequentially. Longer drafts require more draft-model forward passes. Making the draft model larger improves accuracy but also increases latency. Parallel drafters solve the latency problem by proposing an entire token block at once. However, later tokens in that block do not fully depend on earlier predictions. Their accuracy often drops quickly, creating suffix acceptance decay. 3️⃣ How DFlash Improves Parallel Drafting DFlash uses a parallel, fill-in-the-blanks-style draft model. To improve draft quality, it extracts hidden states from several layers of the target model and fuses them into an additional context representation. Each draft layer then attends to two sources: 🔹 Context derived from the target model 🔹 Representations from the draft tokens themselves The target model’s hidden states make the lightweight drafter more informed without requiring multiple autoregressive passes. DFlash can therefore propose a long block in one forward pass. But longer blocks still create a problem: the suffix is more likely to be rejected, while the target model must spend compute verifying it. 4️⃣ DSpark Adds Sequential Structure DSpark extends DFlash with semi-autoregressive generation. It first uses a parallel backbone to generate intermediate logits for multiple positions. A lightweight sequential module, implemented with an RNN or Markov head, then produces the draft tokens from left to right. This small sequential step restores dependencies inside the block without giving up most of the parallel speed. It also predicts a confidence value for every token: the probability that the token will survive target-model verification if the previous prefix is accepted. The result is a stronger draft with less suffix decay. 5️⃣ Verification Length Becomes Dynamic DSpark does not automatically send the entire draft to the target model. Its Hardware-Aware Prefix Scheduler considers: 🔹 The survival probability of each draft prefix 🔹 The current batch size and system load 🔹 A profiled steps-per-second curve for the hardware Verifying one more token may increase the expected accepted length. But it also enlarges the verification batch and can reduce processing speed. The scheduler expands each prefix only while estimated throughput continues improving. Low-confidence suffix tokens are discarded before they consume target-model compute. Draft long, but verify only the prefix that is still worth verifying. 6️⃣ The vLLM Test Setup The author tested both methods under the same environment: 🔹 8× NVIDIA A800-SXM4-80GB 🔹 Qwen3-4B target model 🔹 DSpark block-7 and DFlash block-16 draft models 🔹 vLLM 0.26.0 The initial comparison used DSpark with four speculative tokens and DFlash with seven. On 250 GSM8K questions: ✅ DSpark: 35.2% accuracy, 0.75s average latency ✅ DFlash: 31.6% accuracy, 0.84s average latency On 250 MMLU questions: ✅ DSpark: 28.8% accuracy, 0.24s average latency ✅ DFlash: 27.6% accuracy, 0.27s average latency DSpark reduced average latency by roughly 11% in both tests. 7️⃣ Throughput Is the Stronger Result Because the two methods used different speculative-token settings, the author swapped those parameters and tested again. 🔹 Original settings: DSpark=7, DFlash=4 DSpark reached 584 tok/s, while DFlash reached 449 tok/s. ✅ DSpark was 1.30× faster. 🔹 Swapped settings: DSpark=4, DFlash=7 DSpark reached 561 tok/s, while DFlash reached 480 tok/s. ✅ DSpark remained 1.17× faster. Changing num_speculative_tokens affected throughput by less than 7%. DSpark remained faster in both configurations, showing that its advantage did not come from receiving a more favorable draft length. Compared with the 229 tok/s baseline: 🔹 DSpark delivered 2.45–2.55× throughput. 🔹 DFlash delivered 1.96–2.09× throughput. At matched settings, DSpark stayed roughly 20% faster than DFlash. ⚠️ Do Not Overread the Accuracy Numbers The accuracy differences are less conclusive. MMLU results varied by about one percentage point. On GSM8K, even the baseline changed from 29.2% to 34.4% across two runs. The author attributes this to different vLLM batch compositions changing floating-point accumulation order. That can alter a small number of token choices even with temperature=0. So the test strongly supports a throughput advantage. It does not establish that speculative decoding improves model intelligence. 💡 The Practical Takeaway DFlash proves that parallel drafting can generate many candidates cheaply. DSpark adds the two components needed for production serving: 🔹 Lightweight sequential modeling to improve draft quality 🔹 Load-aware scheduling to avoid unnecessary verification Its real contribution is not simply drafting more tokens. It is deciding which tokens are still worth verifying under the current serving load. 🔗 DeepSpec: 🔗 Test notebook: 🔗 Full Reading: #DSpark# #DFlash# #SpeculativeDecoding# #vLLM# #LLMInference# #AIInfrastructure# #DeepSeek#

I’ve seen too many projects treat the token as the starting point and the asset as an afterthought. But @EthraShip made me reconsider something: when a blockchain project doesn’t begin with a whitepaper, but with a ship that’s already been sailing for four years, the entire narrative shifts. Looking deeper, what caught my attention wasn’t another RWA concept — it was the order of construction. Most RWA projects follow a path: whitepaper first, token second, assets later — if at all. EthraShip took the opposite route. In 2021, Ethra Invest was already operating dry bulk vessels, signing charters, paying crews, weathering shipping cycles. Only in 2026 did they add a blockchain layer on top of this already-functioning business. Assets first. Token later. What stands out: • A business-first, protocol-second approach — not tokenizing a concept, but bringing an already cash-flowing operation onchain • A deliberately separated two-layer structure — $SHIP handles governance and ecosystem participation, while the regulated RWA investment layer follows its own compliance path. Two tracks, each with its own standards, rather than forcing everyone into one framework • An educate-first, incentivize-later community rhythm — before pushing staking and governance, they let users learn the protocol and build reputation through Ethra Portal The most interesting infrastructure projects are rarely the loudest ones. They spend more time making the business model work in the real world than making noise on social media. If EthraShip continues on this path, the value will likely come not from today's narratives, but from every nautical mile those ships have sailed over the past four years. Crypto chases minute-by-minute liquidity. But shipping cash flows in months and years. That’s not a flaw — it’s the structure. And EthraShip is one of the few attempts to build an on-ramp designed for that structure.

Top Overnight News Iran has gone about seven weeks without shipping meaningful crude exports through the Strait of Hormuz, as a U.S. naval blockade succeeds where years of sanctions failed by cutting off one of Tehran's main sources of foreign-currency earnings. Unlike ‌previous sanctions campaigns, when Iranian crude continued reaching buyers despite restrictions, the current blockade has stopped fresh crude cargoes reaching China, Tehran's only major remaining oil customer, increasing pressure on government finances and foreign-currency reserves. Reuters Fighting between the US and Iran over control of the Strait of Hormuz intensified after a period of relative calm, triggering a fresh jump in oil prices. The US military carried out strikes targeting radar systems and mine-laying capabilities along Iran’s southern coast, and Iran retaliated with drone and missile volleys on US bases across the Middle East: BBG The U.S.-Canada trade standoff is threatening to stretch past the midterm elections. The White House is shrugging off the threat. With a week to go before Canada imposes its retaliatory tariffs, aimed at key industries in states like Ohio and Texas with competitive Senate elections, tensions between Washington and Ottawa are at a steady simmer, with no sign of de-escalation. Politico Russia since 2023 has been helping Iran develop advanced supersonic cruise missiles, and Putin has vowed to continue to provide support to Iran. FT/Washington Post Bank of Japan Board Member Hajime Takata, one of the central bank’s most hawkish members, left the door open for an outsized interest-rate increase as well as back-to-back hikes, indicating he might push for a faster pace of tightening. The BOJ has raised its benchmark interest rate in quarter-point increments in the most recent three moves, while spacing the moves roughly six months apart. The size and frequency may change as circumstances evolve. BBG Governor Kazuo Ueda said on Tuesday that the Bank of Japan will debate raising interest rates including in ‌September with a focus on whether inflationary risks were heightening, signaling a strong chance of a hike this month. Reuters The Reserve Bank of New Zealand raised interest rates Wednesday, warning that risks for higher inflation remain in play for the economy. The central bank raised the official cash rate by 25 basis points to 2.75%, in line with market expectations. BBG A closely followed measure of artificial intelligence token prices touched fresh lows this week, the latest sign of deflating prices in an increasingly competitive landscape. Anthropic is releasing a new version of its powerful Fable artificial intelligence model that it says is better at coding and science tasks, as well as more economical. BBG OpenAI is to restrict Astra model after rating it a critical cyber risk: WSJ. Dell became the latest company to reinforce optimism around the AI trade. The company increased its annual sales forecast by $25 billion in a further sign of surging demand for servers to run AI tasks. BBG