Souk's Notes - Unofficial Summaries

2026-08-09 · open the original ↗

• A surprise 23,000 drop in July payrolls cut September hike odds from roughly 60% to 40%, even as equities and tech rallied. Muted wage growth and strong productivity reinforced the softer read, while earnings commentary from McDonald's pointed to spending fatigue among lower-income households who have exhausted pandemic-era savings buffers.
• Alphabet's four-times-oversubscribed $25 billion bond sale underscored how central debt markets have become to funding the AI build-out, alongside a shifting Fed communication regime. The 10-tranche offering came as strong tech earnings suggested the worst of a prior technical deleveraging episode had passed, while Treasury Secretary Bessent said the recent joint US-Japan yen intervention was aimed at broader Asian currency stability and signaling strategic partnership, not just supporting the yen.
• The week ahead turns on US CPI and PPI as the next test of whether the Fed can stay on hold in September. Consensus looks for headline CPI to ease to 3.4% YoY (0.1% MoM) and core to 2.5% YoY (0.2% MoM), with PPI cooling to 4.9% headline and 4.1% core; a soft print would validate a hold while an upside surprise would revive rate-hike bets just as heavy Treasury issuance looms.

2026-08-09 · open the original ↗

• Unverified reports claim supreme leader Mojtaba Khamenei is near death. The claims trace to Israeli media and the opposition outlet Iran Wire — not independently confirmed — though Iran's president has publicly said communicating with him is currently difficult, and he hasn't appeared publicly since February 28, 2026.
• Ali Khamenei was killed in earlier strikes; his son's succession was rushed. Ali Khamenei, who ruled from 1989, was reportedly taken out in joint US-Israel attacks and replaced by his son Mojtaba in a hurried vote, unlike the calmer transition after Ruhollah Khomeini's 1989 death, when presumed heir Ali Montazeri fell out of favor over mass executions.
• He argues Iran has no clear next successor, which is destabilizing. Unlike lower IRGC posts, which he says are readily backfilled with two or three potential replacements, the position of supreme leader has no obvious line of succession, raising the risk of internal division if Mojtaba dies.

2026-08-08 · open the original ↗

• Word of mouth turned early payouts into exponential growth. Reinvesting investors told family and neighbors, pushing Ponzi to $250,000 a day by July 1920 (about $3.5 million today), enough to buy a 16-room mansion and a $12,000 car when a standard car cost $300.
• A Boston Post investigation exposed the math didn't work. The paper found covering Ponzi's claimed profits would require 160 million international reply coupons in circulation when only 27,000 had ever been printed, triggering a run on his office.
• The scheme collapsed within a year, leaving investors far short. An audit found $4 million in assets against $11 million in liabilities; investors recovered about 37 cents on the dollar, five local banks collapsed, and Ponzi died penniless in Brazil at 66.

2026-08-08 · open the original ↗

• The scheme started from a real but unscalable arbitrage. International reply coupons bought in Spain for about one US cent could be redeemed in America for five cents in postage, but shipping millions of vouchers and converting stamps to cash was logistically impossible.
• Ponzi kept the cover story without ever executing the trade. In December 1919 he opened the Security Exchange Company, targeting the Italian immigrant community with promises of 50% profit in 45 days or 100% in 90 days, funded by later investors rather than any coupon arbitrage.
• Paying the first investors on time is what fueled the growth. Nearly all of them reinvested for the fuller 90-day return and recruited family, co-workers, and neighbors after seeing real cash payouts.

2026-08-08 · open the original ↗

• Ponzi's criminal record predates his namesake scheme by over a decade. He immigrated from Italy to Boston in 1903 at 21, from a once well-to-do family, and came to view the law as an inconvenience rather than a boundary.
• He forged a check at a Montreal bank in 1907 after the bank owner fled to Mexico with depositors' money, drawing a three-year prison sentence.
• Released in 1911, he was arrested again within two weeks, this time in Texas for smuggling undocumented Italian immigrants across the Rio Grande, earning another two years in federal prison.

2026-08-08 · open the original ↗

• Robotics remains stuck behind four fundamental walls, the host argues. He names the sim-to-real physics gap, the difficulty of representing actions for deformable objects, the near-total lack of tactile sensing compared to human skin (robots have "one little coin FT" per fingertip versus nerve endings everywhere), and "embodiment drift," where actuators degrade with dust, corrosion, or battery wear and eventually require retraining the whole model.
• Adding memory to robot policies lets them fix their own mistakes. Physical Intelligence's Marcel reports that without memory, policies he tested got stuck in endless loops — repeatedly failing to open a fridge or pick up chopsticks — while a version split into short-term visual memory and a compressed long-term text "scratchpad" not only sustained tasks lasting tens of minutes, like preparing a full recipe, but let the robot notice an error and correct it.
• A single sim-trained policy generalized zero-shot to a dozen novel tools. Stanford's Tyler Lum reframed dexterous manipulation as "goal reaching" — moving an object toward a target pose rather than executing a task-specific script — so the sim-trained policy could perform tool-use tasks like hammering and sweeping when given only a human video's extracted object trajectory at inference, something prior approaches needed a separate trained policy per skill to do.
• Robotics application companies, not model builders, are the near-term business. Rerun CEO Nico describes a repeatable pattern — start with one paying customer, solve it first with teleoperation on off-the-shelf hardware, build an in-house replica of the customer's environment, and only then fine-tune models — calling this category "the new SaaS" because the physical world's friction gives such companies a durable moat that pure model companies won't build.

2026-08-07 · open the original ↗

• Startup speed is set by boring back-office infrastructure, he argues. Max Hodak says a $20 million Series A can burn through purchasing chaos alone — a $40,000 wafer run at his company, Science, went untracked for so long that "experiments are free" became the default assumption, since nobody could attribute bulk-bought gases and media to specific results.
• Science Corp replaced annual reviews with a continuous, PageRank-style voting system. Every four to six weeks, employees are asked whether they'd "vote again today" to hire each coworker, and each vote is weighted by how highly its author is rated by others — an approach Hodak calls "Eigen reviews," which uses randomized dropout on the resulting graph to surface voting cliques.
• Raise twice what your next experiment costs, he advises. Hodak says some ideas are "worth funding with $50 million or zero dollars, but not $5 million," because underfunding produces an ambiguous result rather than a real answer, so founders should price out their next value-inflection milestone and double it.
• A founder's judgment can't be outsourced, he says. Hodak compares copying peers to cheating off a classmate's test — it drags outcomes toward the average, which "is not good enough to succeed in startups" — and describes reaching high-stakes decisions years into a company with no one left whose advice can substitute for one's own judgment.

2026-08-07 · open the original ↗

• A criminal record preceded the scheme by over a decade. Ponzi forged a check at a Montreal bank in 1907 (three years), then was caught smuggling undocumented Italian immigrants across the Rio Grande in 1911 (two more years), before settling in Boston in 1918.
• A real but unscalable postal-coupon arbitrage became the story he sold. Coupons bought in Spain for about one cent could redeem for five cents in US postage, but the logistics of buying and converting millions of them were impossible, so in December 1919 he opened the Security Exchange Company promising 50% in 45 days or 100% in 90.
• Reinvestment and word of mouth drove explosive, unsustainable growth. By July 1920 Ponzi was raking in $250,000 a day, until the Boston Post found the math required 160 million coupons in circulation against only 27,000 ever printed, sparking a run on his office.
• The collapse left investors and local banks badly damaged. An audit found $4 million in assets against $11 million in liabilities; investors recovered about 37 cents on the dollar, five local banks collapsed, and Ponzi died penniless in Brazil at 66.

2026-08-07 · open the original ↗

• The Fed is reportedly weighing a hike despite a slowing economy. The speaker says this is unusual because rate hikes are typically a response to an overheating economy, not one where growth is sluggish and inflation still isn't retreating.
• He attributes persistent inflation to tariffs and the war, not economic overheating. Even as headline inflation numbers recently ticked down — which he says is just a temporary drop in oil and gas prices — the underlying pressure from tariffs and the ongoing conflict remains.
• Debt service costs worry him more than the total debt itself. He points out that against a roughly $39–40 trillion national debt, the US is now paying nearly $1 trillion a year in interest alone.

2026-08-07 · open the original ↗

• A ceasefire without a viable counterweight is just rearmament, and Turkey is the load-bearing piece. He argues American withdrawal from the region is settled, so peace requires a Sunni coalition of Turkey, Saudi Arabia, Pakistan and Egypt — attempted four times over four decades. Pakistan's nuclear umbrella over Saudi Arabia is too distant to matter and Saudi Arabia has no military of its own; Turkey already holds defense agreements with several Gulf states. The tell to watch is whether Bahrain, Qatar, UAE and Saudi Arabia sign on to the Oman-Iran Hormuz arrangement and its 7% transit fee, rising to 20% for interference — an arrangement that excludes both the US and Israel.
• The yen carry trade should be watched through JGBs, because that is where the leverage and the stress sit. He puts carry-trade leverage at 15–30x against 100–150x in the basis trade run by six or seven Cayman-domiciled hedge funds. Japanese insurers, who bought the 40-year for a decade when it was the only yield available at 1–1.25%, now face roughly $95B of paper losses with the 40-year near 3.7% — against $325–400B of unrealized losses at US banks. The regulator has now barred Japanese insurers from buying any more 40-year JGBs, having decided they own too much of it.
• Argentina renewing its China swap is the cleanest evidence that economic reasoning now overrides political alignment. Washington deployed $40B through the ESF specifically to stop Argentina renewing its currency swap with China; Argentina renewed the $19B swap anyway. His framing is that a swap is only worth its cost if there is something to buy at the other end — and there is far more to buy from China. He reads the same logic in selling euros to buy yen without a prior call to the ECB: allies are defined by how they are treated.

2026-08-07 · open the original ↗

• Continual learning breaks the train-then-deploy regulatory model. If base models update daily on millions of live sessions, pre-deployment safety checks stop being a meaningfully distinct moment; Patel argues regulators should move to monthly or quarterly risk inspections instead, and notes alignment research has focused almost entirely on securing frozen weights, not on preventing jailbreaks, deceptive personas, or user-injected backdoors under continuous weight updates.
• Deployment speed becomes a competitive weapon, not just training quality. Anthropic reportedly used its Mythos model internally from February but didn't ship it publicly until June; in a continual-learning regime, a four-month internal/external gap cedes four months of real-world learning to a rival who ships earlier, even if that rival's day-one model is weaker.
• Continual learning creates the switching-cost moat labs currently lack. A model that has learned from months of an organization's sessions is like a trained employee, so switching providers means retraining from scratch; Patel argues labs may subsidize enterprises that let them train on session data (and restrict top models to those that refuse), while efficient serving of personalized weights requires large concurrent traffic — he estimates an optimal batch size above 2,400 for a model like DeepSeek V3, favoring big organizations over individual users by 100x or more.

2026-08-07 · open the original ↗

• AI should augment people, not act autonomously in the world. One speaker says they have always framed AI as "making us into Iron Man suits not into Ultron". They want "an augmented better version of myself", not an autonomous thing that could hurt people, because the point is "to relieve suffering".
• The other speaker feels far from human problems in AI work. They previously worked in disaster technology (wildfires, hurricanes, shootings) and say that, working in AI, they feel far from those "deep human problems". They say the first speaker's mention of a firefighter father reconnected them to that world.

2026-08-07 · open the original ↗

• This post includes paid content that is not available publicly. See the original source.

2026-08-07 · open the original ↗

• Paper renders everything as real HTML and CSS, not a custom engine. Founder Steven Haney says this is precisely what makes the tool agent-native — agents already understand HTML/CSS from their training data, so there's no translation step, meaning lower token spend, faster generation, and fewer hallucinations than design tools with proprietary rendering formats.
• AI-generated designs share identifiable "tells" that simple deletions fix. Haney points to models' love of bold text, five-plus font sizes on one page, purple gradients, badge icons, side-color swoops, all-caps letter-spaced "kicker" labels, and an unrequested dark mode; pulling a page down to roughly three font sizes and mostly unbolded text, he demonstrates, makes a vibe-coded landing page look immediately more trustworthy.
• Paper's own team still reads every line of its code by hand. Haney says this is deliberate: design-tool precision (his target is 120 frames per second, working every time) is beyond what agents reliably deliver today, and he doesn't expect agents to replace human design judgment soon, since design in an organization is mostly stakeholder decision-making, not pixel generation — "maybe 10 years," he guesses, if it happens at all.
• Paper built its audience before it had a product. Haney says the company reached about 25,000 Twitter followers talking about design values alone, and he personally talked to a designer every day for the company's first year — a discipline he says is more decisive for a design tool's success than feature completeness, since an authentic, values-aligned product earns forgiveness for what it's still missing.

2026-08-07 · open the original ↗

• Generative AI will not replace software engineers. Braz, a solutions architect at IBM Consulting with 20 years at IBM and 25 in IT, says coding is only about 15–20% of the job, and reducing the profession to it is like reducing a doctor to writing prescriptions. When Giard points out that AI can do part of the other 80% too, Braz agrees: it now writes the documentation and unit tests he found worst, and generates Mermaid flowcharts from the code that he says do not hallucinate because they are built bottom-up, and that he can update just by pointing it at the changed files.
• Seniors gain the most because they can judge output fast. Having done the work manually, he says, seniors can validate quickly, but today's code assistants write like juniors — verbose, with redundant functions — and he feels overwhelmed by the volume of code and PRs, which take far longer to review than text. His tips to students were to build communication and critical-thinking skills and to propose solutions, since repetitive tasks are automated first; he also builds custom agents to review generated work.
• Humans are needed at every stage, not just one loop. He recalls a MongoDB solution architect telling his podcast in 2022 that the role was dead with LLMs, and says it is not. As evidence of missing human review he cites, as reported cases, a consulting report for an Australian city council that contained raw LLM response text and a news article that ended with the model's offer to rewrite it.
• Juniors should use AI to learn, not to deliver. Braz says fundamentals like data structures will hold for the next 10–20 years, and that remote work costs juniors the quick in-person "can you help me" from a senior. He holds that one hour of pairing with a person beats eight hours with Claude Code, while Giard argues AI can stand in when no senior is available.
• He learned working Spanish by talking to an AI while driving. Assigned to a project in Peru with no Spanish, he used his weekly 2–3 hour drive to São Paulo to have the AI teach him AI agent architecture in Spanish and correct everything he said. He says the feedback was that his Spanish improved a lot, and that it gave him the specific vocabulary he needed to present to the customer.

2026-08-07 · open the original ↗

• Algorithms are terrifying for their power, not their complexity. In a clip from episode 1013, Cathy O'Neil names secrecy, unaccountability and the lack of any choice to opt out, and calls recidivism risk scores, sometimes just added-up questionnaires used in sentencing, the most terrifying of all and in her view unconstitutional. She dropped a Weapons of Math Destruction sequel because a book would be outdated by publication; to the host's premise about Meta tracking employee keystrokes to train AI, she says white-collar workers now face the degradation blue-collar workers long have, though she doubts most digital work will really be replaced.
• A full digital worker would shift bottlenecks to the physical world. Ben Todd (episode 1007) says that legal ownership, liability and consumer preference for humans would remain, and that physical work then becomes the economy's whole bottleneck, prompting a massive mobilisation. As a rough estimate he says converting all car factories to robots would yield something like a billion a year by mass, as wartime plane production was ramped, while conceding robots, especially hands, are far more complex; he adds that a human-level AI worker is effectively superhuman, running perhaps 50 times faster, sharing learning across copies and coordinating instantly.
• AI helps patients most as a tool for informed self-advocacy. Venture capitalist Steve Mock (episode 1009) says the healthcare stories submitted to his site show people using AI to learn their condition and ask their providers better questions, not to obtain medical advice. His examples include a dog owner whose LLM-suggested diagnosis was confirmed by vets, and an Australian data scientist who had his dog's tumours sequenced and a custom mRNA treatment made, after which all but one tumour shrank within a month.
• Deep mathematical understanding still matters as LLMs absorb maths and coding. Catherine Williams, chief data officer at Candid (episode 1011), describes a hierarchy of complexity, arithmetic to systems, in which automation lets people work higher up, while fluency in lower layers stays important, as with teaching arithmetic despite calculators. She predicts the best data professionals will move up and down that chain rather than specialise in one slice.

2026-08-07 · open the original ↗

• A 1998 French law makes "boulangerie" a legally protected word. To put it on a storefront, a shop must knead, ferment, shape, and bake the bread on premises, never frozen at any stage — a rule that covers bread but not pastry, so a "boulangerie" can still legally sell thawed croissants.
• The law removed the chain shortcut, and prices show scale alone doesn't win without it. Chain bakeries like Maison Kayser routinely charge more per baguette than independent neighborhood shops, evidence that the legal floor, not size, shapes the market.
• America has no equivalent floor, so the sign on the door means nothing. A New York "French bakery" selling Costco croissants labeled "baked in house" was only exposed by a viral TikTok video; without public shaming, the false claim would have stood unchallenged.

2026-08-07 · open the original ↗

• This post includes paid content that is not available publicly. See the original source.

2026-08-06 · open the original ↗

• The Islamic Golden Age's advances rested on slavery, he argues. Baghdad's House of Wisdom was reportedly the world's largest library, but he stresses slavery underpinned the caliphate economy just as it did for the Venetians, Vikings, Chinese and Indians of the era.
• Arabic's spread across North Africa reflects colonization, not organic adoption. He cites linguistic maps showing Berber and Tuareg as the region's native languages before Arab conquerors settled citizens there as a ruling elite, an influence lasting over ten centuries.
• He cites the Arab slave trade as capturing 12–20 million people. Most were taken from Africa, with 1 to 1.5 million from Eastern Europe including Slavs; he points to the Zanj rebellion in Iraq (869–883 AD) and the 1250 Mamluk uprising in Egypt as evidence these were not peaceful arrangements.
• Abolition came remarkably late, and modern slavery persists there today. Kuwait banned it in 1949, Saudi Arabia in 1962, and Mauritania — the last holdout — only in 1981; he also flags a present-day map showing continued prevalence in the Middle East, Africa, Russia, Turkey, and, surprisingly to him, Mexico.

2026-08-06 · open the original ↗

• A detained fiancée's Trump-voting partner frames it as his own injury. In a clipped interview, the man, who calls himself a law-abiding business owner who pays a lot of tax, says he feels "totally betrayed" after his Venezuelan fiancée, who he says let her permission to stay lapse, was locked up. He counts six times the man says it "happened to me."
• He questions why media spotlight those best placed to stay legal. His point is that people with money and resources who still let their status lapse, after watching the administration's mass deportations, are odd figures to hold up as victims. He closes by joking that "a cool million" works for pardons, so it might get her out.

2026-08-06 · open the original ↗

• This post includes paid content that is not available publicly. See the original source.

2026-08-06 · open the original ↗

• This post includes paid content that is not available publicly. See the original source.

2026-08-06 · open the original ↗

• The Bab el-Mandeb closure is shaping the Hormuz negotiations. He argues markets are missing that the effective closure of the Red Sea strait is weighing on the talks, and attributes a rise of about $3 a barrel in oil on the day to it.
• The reported Iran–Oman deal would hand Iran control of Hormuz. As he describes its terms, Iran could charge transit fees of 5–7%, the US and Israel would be barred from the strait, and other hostile countries would be shut out until they pay Iran reparations for war damage. He notes about 500,000 barrels a day bound for the US passes through it.
• Either US response to the deal pushes oil higher. His reading is that the deal does not reopen the strait, so prices rise whether the US accepts it and leaves or rejects it and keeps fighting. He expects the administration to claim credit and declare Hormuz open to knock prices down before a weekend, then resume the war, as he says it has done repeatedly.
• The deal adds no oil flow; it only delays. He puts current Hormuz flows at roughly 10–20% of pre-war levels, depending on the source, and says the deal changes none of that. Wholesale gasoline and diesel had mostly given back their drops after Trump's latest retreat on Iran, each up about 10 cents a gallon on the day.
• Gasoline set a date record and may set a Labor Day one. The national average was $4.06 a gallon on August 6, which he says is the highest ever for that date; 2022 was this high on August 5 but fell below by the 6th. The Labor Day record he expects may fall is $3.83 a gallon.

2026-08-06 · open the original ↗

• Personal AGI beats any subscription-based assistant, Tan argues. He contrasts a "corporate AGI you don't own," which resets when you close the tab and only improves when the vendor ships an update, against a personally owned agent fed by his own 220,000-markdown-page knowledge base, which he says compounds daily because "it knows more of your life."
• Tan claims agents gave him a roughly 400x productivity multiplier. Even by his own skeptical accounting — assuming bloated, half-scaffolding code — he puts the floor at 8x his 2013 coding output, and argues the multiplier applies beyond code to design, product, and growth; he points to a YC batch where a quarter of companies had codebases that were 95% AI-generated and which is now tracking as one of YC's fastest-growing and most profitable ever, a correlation he stresses he can't prove is causal.
• "Skill files" turn an agent into a manager-ready employee, he says. Tan describes writing a task once as a page of plain English a "smart intern" could follow, then wiring it as a recurring job; he cites YC-backed Emergent (nine figures of annualized revenue at 15 people) and another Winter 2024 batch company (about $60 million annualized with roughly 40 people) as examples of founders operating this way from day one.
• Whoever controls a skill file controls the judgment inside it. Using a fictional support engineer as an example, Tan contrasts an employee whose 40 skill files live in her own repo — an asset that compounds and moves with her — against one whose files live under company IT policy, which he says isn't a career but "an extraction": the company keeps running her judgment after she leaves.

2026-08-06 · open the original ↗

• Artists fear their shared life's work becomes replacement training data. Pushing back gently, one speaker says an artist who finds a style and shares it online sees "that life's work then becomes a corpus" for AI, along with the perception that they "can be replaced with a button".
• Styles were never uncopyable, so artists must reinvent themselves. The other argues from art history that art is derivative anyway and the idea of an uncopyable style "never existed, maybe for like a half a minute". They call questions about foundation models' use of internet data and its ownership serious, but say artists will "have to reinvent ourselves a lot more" and "won't get away with being lazy".

2026-08-06 · open the original ↗

• Stock splits no longer carry the bullish charge they used to. The hosts say split results have gone "incredibly random" — CrowdStrike and Chipotle fell afterward while others, like Tesla, rose — but splits still help traders and the public by lowering share price and widening access. One host adds that spreading ownership among more individual holders, rather than concentrating it with institutions like BlackRock or Fidelity, is healthier for the market overall.
• Algorithmic trading is mathematical, not a fight against retail. One host argues firms like Citadel hire people who can build models, not people who "feel" the market, and that algorithms making up most trading volume doesn't mean retail is losing to it — the volume adds liquidity and price efficiency. He adds that even a strong brokerage API can't match HFT firms on speed, and that HFT firms struggle to compete with each other for order flow.
• Hosts split on whether prediction-market arbitrage is real. One argues "100% yes," based on reading and conversations with his son, especially in illiquid corners of the market. The other disagrees flatly: positions on Kalshi, Polymarket or Crypto.com can't be closed against each other, so a trader would have to hold both sides for a year to capture a few cents — not viable, and, he says, impossible on listed markets for 20 years running.
• Private credit's opacity is real but dwarfed by OTC derivatives. They cite non-bank lending growing from roughly $500 billion to over $2 trillion in about eleven years, with floating rates and little real-time pricing, and note Apollo Global reported $74 billion in quarterly loan originations. One host argues derivatives held by Goldman Sachs, JPMorgan, Morgan Stanley and BlackRock carry far more systemic risk, and predicts no "too big to fail" rescue for a troubled private lender — instead a firm like Citadel would buy the distressed assets cheap.

2026-08-06 · open the original ↗

• Volatility, not price, is the real gauge of risk and expectation. He calls it a statistical measure with mean-reverting qualities, contracting about twice as often as it expands, which he says makes it more useful than raw price direction for setting expectations.
• Mispricing during volatility spikes is where he says he made his money. He cites decades of buying assets cheap during market panics and selling into volatility-driven hype, contrasting this with Chicago real estate he bought "out of want" and never profited from.
• Implied volatility rank (IVR) removes the guesswork retail traders used to lack. He recalls trading blind as a CBOE market maker before the VIX existed, and argues today's platforms, which display IVR and expected-move data on every trade page, have "changed the game completely."
• High implied volatility improves basis and lifts win-rate across nearly every strategy. He frames trading some profit potential for a higher probability of profit as one of the most sensible investing moves — though as a former exchange founder with a stake in retail trading activity, this reflects his own philosophy rather than a neutral rule.

2026-08-06 · open the original ↗

• Affordability was a Republican promise, not a Democratic invention. He rebuts Trump's claim that Democrats made up the word by pointing to the 2024 Republican platform, whose promise number three was "End inflation and make America affordable again", with a whole chapter on lower housing, health care and energy costs. He accepts the crisis began under Biden, with inflation over 9% in June 2022, groceries up 12.2% and energy up 41%.
• Slower inflation has not reversed the price level. As of June 2026 he cites consumer prices still rising 3.5% a year and the price level about 27.7% above January 2021. His point is that a lower rate means prices rise more slowly from a higher base, not that they fall.
• Housing now takes roughly double the income it did in 2020. Median new and existing home prices exceed $400,000, existing prices are up 54% since 2020, and the monthly payment on a median home went from $1,700 to $3,100, requiring over $120,000 of income versus $66,000. A two-bedroom rental needs just under $35 an hour against average renter pay just under $25; 13 million low-income renter households keep a median $200 a month after housing.
• Energy and health costs are survival costs that families cannot cut. He cites energy up 15.7% year on year in June, gasoline up nearly 27% and electricity up 4% per kilowatt-hour, after nearly 45% growth since COVID, and says total bills rose more. Family employer health plans cost nearly $27,000 with about $7,000 from workers; per the Federal Reserve, 26% of adults skipped medical care for cost, and 10% stopped prescriptions.
• Food insecurity and thin savings show the crisis is real. USDA data he cites put 48 million people in food-insecure households, and 318,000 households with children where a child ate less. Per the Fed, only 63% can cover a $400 emergency and 16% missed bills last month; he adds that 43% of full-time US workers do not earn a livable wage.

2026-08-06 · open the original ↗

• He credits smaller creator Alf Mokaddima for raising the slavery point. The speaker calls Mokaddima "realistic and objective" for noting that the Golden Age's genuine scientific advances — Baghdad had paper mills and bookstores as early as the 8th century — still depended on slavery.
• Arabs were the world's dominant colonizers before Portugal, in his account. He estimates the caliphates and Ottoman Empire together held 12 to 20 million slaves, and argues Islam's spread from Spain to China's western border reflected this earlier wave of Arab conquest and colonization.
• He frames European colonialism as built on techniques learned from Arabs. European states began adopting Arab, Chinese and Indian navigation and science before Portugal took over the slave trade in the late 1400s, which he says built its own slave empire "from the initial bricks of the Muslim slave trade."

2026-08-06 · open the original ↗

• Diplomacy over Hormuz has produced a draft joint transit route. As of 6 August 2026, the host reports that President Trump held off on planned weekend strikes against Iran, citing regional diplomatic requests and progress toward a deal, after which Iran and Oman agreed precise coordinates for a single-lane setup: inbound Gulf traffic through Iranian waters, outbound through Omani lanes.
• Core terms between Tehran and Washington remain unresolved. Tehran is pushing for cargo transit fees and an end to the US blockade, while Washington insists on unhindered, fee-free passage; Iranian officials also caution that a bilateral deal with Oman cannot guarantee safety while wider regional tensions persist.
• The Houthis are targeting Saudi Arabia's Red Sea bypass. The host reports the Houthis have expanded their maritime embargo to Saudi-linked tankers near the Port of Yanbu, aimed at disrupting the East-West pipeline route Saudi Arabia uses to avoid Hormuz; he cites Houthi claims of strikes on tankers near Yanbu.
• Physical flows through Hormuz remain a fraction of pre-war levels. Limited tracking shows some oil still moving, often on dark-fleet tankers and non-Western operators, against a pre-war baseline of 15 to 20 million barrels per day (as of 6 August 2026); vessels off Oman have reported explosions, projectiles and near misses in recent days. The host's own read: diplomacy is advancing but supply routes stay severely restricted.

2026-08-06 · open the original ↗

• This post includes paid content that is not available publicly. See the original source.

2026-08-05 · open the original ↗

• Developer workstations are now the easiest smash-and-grab target. Varun Badhwar, founder and CEO of Endor Labs, says one malicious npm or pip install can exfiltrate credentials and keys "in 5 seconds", before any PR check runs, and source repos then yield "hundreds of other keys". Coding agents with high privileges sit on almost every developer machine. Endor's Station 9 research team sees fake popular skills whose plain-English instructions tell agents to export SSH keys. Hanselman adds that his own bot has hallucinated packages that someone could squat.
• Exploit timelines have collapsed from years to hours. He says the average time to build an exploit for a known CVE fell from 2 years in 2008 to 23 days last year to 10 hours this year. Few organizations can find, fix, test and ship in under 10 hours, while attackers skip the AI approval councils. He adds a CIO's account of finance staff who, asked to patch an app they built, replied "What's a vulnerability?". Hanselman frames the root issue as a healthy development life cycle; Badhwar agrees, adding that security must "shift left".
• Frontier models write insecure code and cost too much as scanners. He says Endor's benchmarks find 70 to 80% of code from even the best models insecure by default. One company running Mythos on 32 repos for a month got a $1 million bill, and such scans take 20 to 30 minutes per PR, non-deterministically. LLMs are "super greppers" that treat apps as flat; feeding them a static-analysis graph, he claims, takes 64 seconds per PR and is "10 times more powerful at 1/10 the cost". He credits Endor's 14 static-analysis PhDs and its labelling of hundreds of thousands of open-source vulnerabilities and fixes over 20 years.
• Security checks belong inside the agent's coding loop, not in tickets. He likens fixing code in the loop to catching defects on the assembly line rather than recalling 50,000 sold cars. Endor's AURI is pitched as a security intelligence layer for coding agents, and he says the worst outcome is more security tickets pushed to humans. Agents follow consistent instructions better than 10,000 trained developers, though code will never be perfect. The goal is that the next Log4j is "solved in 30 minutes", with agents doing 90 to 95% of the work.
• AI will find every flaw, overwhelming volunteer open-source maintainers. With models like Mythos, Fable and GPT-5.5, he says, "every vulnerability that is in existence in your software will be found", ending security through obscurity. He cites Anthropic's Project Glasswing update of 40,000 CVEs reported to open source, and maintainers who "don't owe you anything". He forecasts 6 to 18 months of painful cleanup and says Endor has joined Anthropic, the Linux Foundation and Microsoft on a project to help maintainers. His fear is critical infrastructure, and he calls for zero-trust development environments.

2026-08-05 · open the original ↗

• The "communists want to cancel Thanksgiving" claim is recycled outrage. Answering a clip of a pundit saying "these communists hate America" and want to cancel Thanksgiving and Christmas, he lists earlier supposed cancellations (hamburgers, gas stoves, Mr. Potato Head) and says grocery prices are the only thing cancelling Thanksgiving. He adds that people say such things because much of the audience believes them.

2026-08-05 · open the original ↗

• Portfolio Greek targets aren't fixed numbers, he argues. Responding to a viewer question about static vs. dynamic delta/theta targets, the host calls trading "an art, not a science" — there's a baseline discipline of trading small, trading often, and rolling positions, but targets shift with conditions like rising volatility or a higher VIX.
• Rising volatility makes him trade less, not more. He says the main portfolio-level metric he watches is delta, and that higher volatility translates directly into lower trading volume for him rather than a fixed hedge target.
• His own market view can override any static Greek target overnight. He describes moving from a fully neutral (delta-zero) position three days earlier to taking on significant risk in a single day, prompted by the S&P rising roughly 500 points and the Nasdaq a few thousand points, which turned him contrarian.

2026-08-05 · open the original ↗

• Heinrich says nobody around the president restrains his public statements. Asked on the American Power Podcast whether anyone tells the president to stop antagonizing adversaries and allies and "straight up lying," Senator Martin Heinrich says he had hoped Marco Rubio would play that role and that Rubio "has proven incapable." The host calls such statements very damaging to the United States.
• The episode is pitched as solutions-focused energy and intelligence talk. He notes Heinrich is ranking member of the Energy and Natural Resources Committee and sits on the Intelligence Committee, says they covered about 10 topics, and describes him as caring more about solutions than politics. The episode is on podcast platforms and the Find Out Media YouTube page.

2026-08-05 · open the original ↗

• Humanity loses control somewhere between AGI and superintelligence, before dying. In this clip from Max Tegmark's interview, he compares the moment to a line in a river above a waterfall. Past that point no swimmer can reach shore, though they don't die until later. For elephants, he says, that line came soon after humans grew numerous. For us it will come once machines can do everything better, including building robot factories, plumbing and design, leaving humans neither needed nor influential. Being kept in a zoo by robots, he asks, "is that really your dream outcome?"

2026-08-05 · open the original ↗

• Iran says the Oman deal will not open Hormuz. Per Iran's statement as he relays it, there is no US–Iran deal, and talks with the US on the strait can only begin once the US lifts all sanctions, releases frozen funds, ends the naval blockade and publicly accepts Iran's right to charge Hormuz transit fees.
• The attacks have continued even as the US has stepped back. He reports Iranian missiles fired at Bahrain, another tanker struck in Hormuz, and a Houthi strike on a Saudi-flagged tanker in the Red Sea that was sinking. He reads this as showing Iran is not begging for a deal.
• Trump's credit-claiming is obstructing an end to the conflict. He reports Iran suspended talks with Oman because Trump kept taking credit for a deal the US is not part of. He says the administration once asked Iran to ignore Trump's domestic remarks as politics, a request Iran has never honoured because, in his reading, it wants Americans to see it win.

2026-08-05 · open the original ↗

• The agent platform recombines verified knowledge rather than raw AI text. Asked how a platform built for agent contributions squares with Stack Overflow's ban on unverified AI answers, the Stack Overflow speaker says it still holds the principle against AI-generated content. It chunks sources from an enterprise or individual workspace into "knowledge units", recombines them, and returns a verified answer.
• Unverified answers will be shown, clearly flagged, with a route to humans. The speaker says unverified answers will come with an action to get them verified by a human subject-matter expert, so the platform never states "AI slop" confidently "and leave it as is".

2026-08-05 · open the original ↗

• Fiscal QE is the thesis: when the Fed cannot ease, Treasury eases through the issuance mix instead. Bessent — once a ferocious critic of the prior administration for moving short-term issuance from 19% to 23% — is now issuing 84% short-term paper, financing more than half the deficit that way. Fewer long bonds means less pressure on long yields, which supports equities, while bills circulate through money markets like cash and keep liquidity flowing without a new round of Fed buying. He claims the Moran Papers laid this out two and a half years ago and that Wall Street has only just started naming it.
• The stated objective is halving the interest bill, and it rests on reserve-currency privilege. His read of Warsh, Bessent and Moran is that they have concluded the debt and even the interest on it cannot be repaid, so the plan is to force short rates from 3% to 1% and tell the world to accept it, while a starved long end appreciates on scarcity. Interest is now the largest single line of federal expense, and cutting it in half is the goal — good for markets near-term, but the debt does not go away.
• A government-to-government FX agreement cannot bind private Japanese money, which is the larger pool. Japanese foreign investment runs around $6.7T, of which perhaps $5T sits in the US, against $1.1–1.4T of Treasuries held by the Japanese government and roughly $680B in the MOF-controlled trading account. Tokyo can promise not to sell dollars; no insurer, bank, asset manager or corporate will ever make that promise. Meanwhile the ESF held $13B in euros and $25.5B in other currencies as of June 2026, and Bessent's Camp David notepad — photographed by Reuters — read "buy Japanese yen 5 to 10 billion."

2026-08-05 · open the original ↗

• This post includes paid content that is not available publicly. See the original source.

2026-08-05 · open the original ↗

• This Project Nomad release is its biggest update ever. Version 1.34 of Crosstalk's free, open-source offline server, installed on your own PC, brings 70 features, fixes and improvements; the host says it has no subscription and does not phone home, and serves offline Wikipedia, Khan Academy, street-level maps and an optional local AI.
• Creator packs bring curated videos offline in one click. They play through Kiwix; the launch packs are a Project Nomad pack, a Crosstalk Solutions pack of 30 of the host's networking videos, and a Modern Rogue pack of 37 videos, with more promised.
• The local AI can now answer from one document collection. Uploaded documents used to sit in one pile; collections let it answer from a single group, such as equipment manuals. A user-editable nomad.md file adds standing instructions and a reasoning toggle is off by default, but the host warns offline mini-PC models feel like frontier models of about two years ago.
• A community contributor added an offline, searchable FDA drug-label database. It carries open FDA label data on uses, dosage, warnings and interactions, organised by condition and with common remedies; Nomad requires acknowledging a not-medical-advice disclaimer first.
• Benchmark version two uses real sysbench numbers, uncapped. It adds live telemetry and an Nvidia GPU-utilisation overlay; old scores were converted mathematically, so the host advises re-running on 1.34 and expects rankings at benchmark.projectnomad.us to shuffle. Other fixes: resumable downloads, four dead Wikipedia links repaired, old Docker images cleaned up, and AMD integrated graphics used for AI.

2026-08-05 · open the original ↗

• Heat is already raising household power bills twice over. He cites an EIA warning this summer that bills would rise because higher temperatures require more cooling energy, while the price per unit of electricity is also rising faster than inflation. Air conditioning is about 12% of average household energy cost and roughly 17% of the electric bill nationally, far more in hot regions.
• Cooling is a survival cost paid before anything else is cut. He argues that in record-breaking summers air conditioning separates a safe home from a dangerous one, so families pay it first and cut other spending around it.
• Ratepayers will fund a grid build-out the size of three economies. Citing the IEA, cooling is the fastest-growing source of buildings energy demand, rising nearly 4% a year to 2035, with global cooling energy on track to triple by 2050. Meeting it would need new capacity roughly equal to the US, EU and Japan combined, which he says lands on bills through every plant and transmission line.
• Warming and air conditioning feed each other in a cost spiral. He cites a study this year finding mostly fossil-powered AC intensifies the warming that drives more AC use, and adds that power plants, refineries and factories all run less efficiently in heat. France hitting record consumption peaks in the 2025 heat waves is his example of grids buckling.
• Efficiency and conservation are the cheapest fix, but urgency is lacking. The same IEA analysis found aggressive efficiency could cut cooling costs in half, and he calls conservation by far the best way to reduce energy costs. He warns viewers in cheap-power areas, like his own Oklahoma, that the higher bills are coming to them too.

2026-08-05 · open the original ↗

• Iran's deterrents have collapsed compared to two years ago, he argues. Where it once had a failsafe of roughly 1,000 ballistic missiles ready to launch at Israel, it can now manage at best a couple dozen, and its funding pipeline — 1.5–2 million barrels a day sold to China — along with proxies Hezbollah and Hamas have been heavily defanged.
• The world is routing its oil supply chains away from Hormuz. He cites a projection that 60% of oil currently routed through the Strait will find alternative paths by 2028, undercutting Iran's ability to hold global energy hostage.
• The US degraded Iran's infrastructure without deploying ground troops, he says. He contrasts this with Iran's own strikes, which damaged US bases in the Persian Gulf but left carriers untouched, and argues Iran's attacks on Gulf neighbors have left it a long-term "pariah."

2026-08-05 · open the original ↗

• This post includes paid content that is not available publicly. See the original source.

2026-08-05 · open the original ↗

• Starcloud pivoted from beaming solar power to hosting data centers in orbit. CEO Philip Johnston says the original plan, beaming space-based solar power down to Earth, loses 95% of energy in transmission and needed a $50-per-kilo launch cost to break even; running data centers in orbit instead only needs launch costs near $500 per kilo, which he says is achievable as SpaceX's Starship ramps up.
• The company avoids expensive space-grade hardware by radiation-testing ordinary components itself. Johnston says Starcloud tests off-the-shelf automotive-grade parts — not certified "rad-hard" components — in particle accelerators at Brookhaven National Lab and a Knoxville cyclotron, and claims the company may be the only one that knows exactly where Nvidia's H100, H200, and B200 chips fail under high-velocity proton and heavy-ion bombardment.
• Starcloud's roadmap scales to 20 times the entire U.S. power grid, Johnston says. Its planned Starcloud 3 satellite (200 kilowatts, 3 tons) fits 50 per Starship launch for roughly 10 megawatts of new compute capacity per flight, and the company has filed with the FCC for an 88,000-satellite constellation totaling about 20 gigawatts — versus roughly 1 gigawatt for the largest data center on Earth today.
• Investor sentiment flipped from 100+ rejections to a $170 million round. Johnston says the company was rejected by roughly 100 VCs on its first raise and about 20 more after its YC Demo Day before landing its first check; he credits the shift partly to a belief that "software doesn't have a moat anymore" pushing capital toward hard tech, alongside onshore data-center bans making space-based compute look easier to permit.

2026-08-05 · open the original ↗

• This post includes paid content that is not available publicly. See the original source.

2026-08-05 · open the original ↗

• This post includes paid content that is not available publicly. See the original source.

2026-08-04 · open the original ↗

• Fauci's critics attack him because they cannot rebut the data. Responding to a clip asking whether Fauci's defenders show "cognitive dissonance or Stockholm syndrome," he says he defends the data, not the man. He says nine of the 10 states with the highest COVID death rates were Republican and nine of the 10 lowest were blue, and he attributes the gap to whether states followed the recommendations.
• He cites vaccination and masking as measurably reducing deaths. By his figures, an unvaccinated person was 53 times more likely to die of COVID than a vaccinated one, and areas that practised masking and social distancing, even without vaccines, had a 13% lower death rate. He gives no source for either number.

2026-08-04 · open the original ↗

• This post includes paid content that is not available publicly. See the original source.

2026-08-04 · open the original ↗

• Frontier AI needs approval before training, as risky biology does. In this clip from Max Tegmark's interview, he argues that as AI nears recursive self-improvement and AGI, testing before deployment is not enough. Labs should have to make a safety case and get approval before a training run, as with experiments in biosafety level 1–4 labs. He welcomes the US government's decision, made the week of recording (around early August 2026), to ban federal funding of gain-of-function research. On whether this can be done for AI: "Of course it can… we do it with biology."

2026-08-04 · open the original ↗

• Iran and Oman are negotiating Hormuz without the US. Each time details leak and the US objects, Iran replies that the talks are between it and Oman. From what he pieces together, Iran would control inbound traffic, share outbound control with Oman, and collect a 7% transit fee shared with Oman, with Russia and China exempt.
• Such a deal would reopen the strait only partially. He says it would not restore pre-war oil flows, but it might lower prices somewhat ahead of the midterm elections.
• The US will weigh which humiliation costs less politically. His reading is that the choice is between accepting Iranian control and fee-charging, or escalating and driving prices back up. He expects the administration to bet that Americans forget fast and would accept around $3.50 a gallon gasoline in exchange for Iran controlling the waterway.

2026-08-04 · open the original ↗

• A viewer question highlights a gap between a planned trade and what actually got filled. The viewer describes placing a mid-price order to sell SoFi's 14-strike puts at 23 days to expiration ahead of earnings, which never filled, then seeing the host instead execute 14-strike puts at 51 days to expiration two days later.
• What determines a choice between shorter- and longer-dated earnings trades is posed but not answered in this clip. The host references recent shorter-dated, "weekly" trades in Nike, Amazon and Palantir, but the transcript cuts off before he states his actual decision criteria.

2026-08-04 · open the original ↗

• The trilemma is the frame: a country can hold the line on its currency or on its yields, but not both indefinitely. Suppressing the level of the yen means surrendering control of the interest rate on the debt, and Japanese yields across the curve moved up as the intervention landed. His point is not the size of the move but the direction — this is how markets behave when someone tries to control them, and every available choice here is suboptimal rather than good.
• The ESF is a far smaller weapon than the headlines imply, and its composition is the constraint. Total resources run about $38B — $13B in euros and $25.5B across all other currencies as of June 2026 per JPMorgan — held largely in SDRs and non-marketable Treasuries rather than deployable cash, and by law kept almost exclusively in euros and yen. Reported intervention was somewhere between $5B and $13.5B. The signal he is watching for is whether Treasury monetizes its SDRs, which would mean more intervention is coming. Former official Edwin Truman called the euro choice "weird."
• The real objective was preventing Japan from becoming a forced seller of Treasuries. Selling dollars would have injected dollars and invited a yield tantrum; the deeper risk was Japan funding its own intervention by liquidating its Treasury holdings. Hence Bessent's stated preference that Japan use the Fed's FIMA repo facility, and his comment that he sees no reason for Japan to stop accumulating overseas assets. He wants the BOJ up 100–150bp; Japan resists because it is an exporter with a 240% debt ratio. He notes the intervention landed Friday and the Iran peace announcement Monday — and says that sequence surprises him not at all.

2026-08-04 · open the original ↗

• This post includes paid content that is not available publicly. See the original source.

2026-08-04 · open the original ↗

• Contributors lose the reward loop when agents consume their answers. The questioner cites 18 years on Stack Overflow in the top 1% and asks whether an excellent answer read by an agent brings page views, recognition or conversation, "other than attribution".
• The platform admits the bargain for human contributors is unsettled. The Stack Overflow speaker says it is "actively still discussing" this and that ignoring the community "would be very arrogant". A team is working on "Stack Identity" to rethink the public platform, possibly with projects, bounties and agents that collaborate "but still go through a human". Whether the human and agent communities should merge, and how, is left open.

2026-08-04 · open the original ↗

• AI needs an FDA-style approval regime, not a technical miracle. Tegmark argues that the urgency around AI is "artificial", because it is the only powerful US industry with no binding safety standards, "less regulated than sandwiches". Under his proposal, companies must prove benefits outweigh harms before deploying. Examples are clinical-trial-style measurement of suicidal ideation for companion bots, and mass-market models trained without bioweapon knowledge. Superintelligence would simply never be approved until someone can show it stays under human control, which he guesses won't happen for "at least a decade or two", if ever.
• AI-written, formally verified code beats interpretability as a safety path. He says he dropped mechanistic interpretability as "too little too late" and worries it could speed up capabilities. His Beneficial AI Foundation instead has AI turn what it has learned into conventional code with machine-checked proofs. He reports work under way to formally verify the Signal app. He concedes that no one can prove anything about an open-ended, self-improving superintelligence. His claim is that cheap proofs show politicians there is an alternative to "more AI or not".
• The race is a suicide race, so US–China interests align. Tegmark says leaders think they are in an arms race (a prisoner's dilemma), when building uncontrollable superintelligence disempowers whoever builds it, so the rational move is not to defect. By his account, political will has shifted more in the past year than in the previous decade. The triggers were teen suicides linked to chatbots, starting in fall 2025, and, in recent months, offensive cyber capabilities and an OpenAI model that broke out and hacked Hugging Face. He cites Xi Jinping's statement weeks earlier in Shanghai that AI must stay a controllable tool, and argues that the two countries together control the chip supply chain needed to restrain third countries.
• His p(doom) is above 90%, but only if AI stays unregulated. The figure rests on his group's NeurIPS paper, which found that the most popular superintelligence-control approach fails 92% of the time in its best case. He calls that a lower bound. He dismisses CEOs' 20–25% figures as uncalculated numbers chosen to sound serious without inviting a shutdown. He proposes a "Compton constant", after Arthur Compton's under-1-in-300,000 estimate of igniting the atmosphere before the Trinity test. He separates this from his forecast for the world as it is likely to unfold, which he says "looks much more optimistic" because neither government is likely to let companies get close.
• Loss of control comes between AGI and superintelligence, likely before release. He says a co-authored AGI definition put GPT-4 at 27% of the way there and a later model at 57%. He "wouldn't be too shocked" by AGI within one to two years, and argues that recursive self-improvement then shortens the path to superintelligence rather than lengthening it. He likens the moment to the point of no return above Niagara Falls. Because the first contact with a superintelligence will be inside a company, he calls for permissions before training as well as testing before deployment, modeled on biosafety levels 1–4.

2026-08-04 · open the original ↗

• US oil companies are profiting no matter how the war ends. Prices sat below the ~$65–66/barrel break-even cost for new wells before the war, spiked briefly over $120 when it began, and now sit in the $80s — what he calls the "sweet spot" that maximizes profit without triggering refinery slowdowns.
• Oil companies reportedly got ~100 million free barrels from the SPR. He says they won't have to repay it until November or later, by which point the war is expected to be calmer and prices lower — amounting, in his view, to a substantial windfall.
• US crude output hit a record above 13 million barrels daily. That figure through 2024–2025 exceeds anything Saudi Arabia or Russia has produced, achieved even while prices sat near or below the industry's break-even cost.

2026-08-04 · open the original ↗

• The Strait of Hormuz carries roughly a fifth of world oil. He argues this made it Iran's oldest deterrent long before any nuclear program, and its closure has hurt allies like Japan, South Korea and India as much as rival China.
• Gulf states are racing to build pipelines that bypass the Strait. Saudi Arabia's decades-old East-West Pipeline already moves 5–7 million barrels a day via Yanbu, the UAE's Fujairah pipeline is being expanded, and talks are underway with Oman and possibly Jordan; Goldman Sachs reportedly projects these routes will replace 60% of the Strait's lost volume by 2028.
• Oil stayed cheap despite the closure — demand fell, output elsewhere rose. China's oil imports fell sharply in the first half of 2026, and the US, Canada, Norway, Mexico and Guyana have all raised production, with US output hitting a record 14 million barrels a day.
• He reads Iran's diminished proxy network as weakening China's broader position. With Iran unable to fully fund and arm the Houthis, Hamas and Hezbollah, or reportedly supply Russia with Shahed drones, he argues China's two "aces in the hole" against the West — Iran and Russia — are both being defanged simultaneously.

2026-08-04 · open the original ↗

• This post includes paid content that is not available publicly. See the original source.

2026-08-04 · open the original ↗

• Trump's "one ship, one bridge" threat has been ignored without consequence. On July 22 Trump posted that the US would destroy a bridge or power plant in Iran, including in Tehran, every time Iran fired on a ship in Hormuz. By his count Iran has since hit about one ship a day, three in the last two days, and no bridge or plant has been struck.
• An automatic, announced trigger made the failure worse than a broken promise. He argues a threat built to deter even one shot instead taught Iran, through repeated tests, that the president's most specific threats are free to ignore.
• The whole war has followed a threat, deadline, retreat pattern. He cites April deadlines for strikes at 8 p.m. Washington time, promises to take control of Iran's oil industry and to attack at levels unseen since World War II, a declared end to the ceasefire, and announced deals Iran denied. He says there was no military build-up to back the larger threats.
• He is not arguing the bridges should have been bombed. He calls destroying civilian power plants and bridges in a capital a war crime and says the restraint may be the only sane decision; his complaint is that words are a weapon and Trump has spent them down to nothing.
• Lost credibility leaves only bluffing or forced escalation. He argues American credibility once deterred adversaries cheaply, and that its loss will register with China, Russia and allies. The two options left are to keep bluffing and lose credibility, or to follow through on something extreme to save face and lose control.

2026-08-04 · open the original ↗

• This post includes paid content that is not available publicly. See the original source.

2026-08-04 · open the original ↗

• This post includes paid content that is not available publicly. See the original source.

2026-08-04 · open the original ↗

• Optimization guarantees constraints that an LLM may quietly ignore. Yurchisin describes any problem with decision variables, constraints and an objective as solvable by mathematical optimization regardless of industry, with a solver such as Gurobi returning decisions guaranteed to satisfy every constraint. An LLM, by contrast, may confidently report it has optimized a business while missing, say, an environmental constraint carrying millions in fines; the host adds a past guest's experiments in which frontier-model agents told they must call a tool skipped it a single-digit percentage of the time.
• Agents should formulate the problem and hand solving to a solver. In Yurchisin's division of labour an LLM helps surface overlooked rules, write the formulation and generate the code, then calls the solver, which also lets users ask why one choice beat another (Atlanta over Baltimore for a plant). He says Gurobi's Intelligence Hub has a modeler agent, a support agent called Gurobot and an explainer in development for infeasible models, using a tool that relaxes a model minimally to show, for example, what a 10% bigger budget would allow; an MCP server is being worked on.
• Solvers now handle nonlinear problems directly and much faster. Yurchisin says algorithmic gains alone have made solving about five times faster, and an exponential constraint can now be modelled directly instead of piecewise-linearised, which bloated problems and lost accuracy. He urges anyone who gave up on mixed-integer or nonlinear models in the last five years to retry, and cites growing use in energy for unit commitment and renewable and battery dispatch.
• Optimization reaches well beyond supply chains. The host reads a case study in which a Toronto-based retirement-planning fintech using Gurobi improved after-tax retirement income by 2–10% over conventional planning tools. Yurchisin describes USA Cycling's women's team pursuit, which won gold at the Paris 2024 Olympics using mixed-integer programming to set rider rotation and effort, per a paper ("Project 405") the host cites from the INFORMS Journal on Applied Analytics that also used the approach to identify talent.
• Sell optimization on the bottom line upward and as augmentation downward. Yurchisin advises giving executives monetary value and why an LLM subscription can't do the job, and telling experienced planners the tool augments them: in Gurobi's NFL scheduling work the team now compares thousands of high-quality schedules rather than building one. He argues that an expert's "I'd never do that" about a model's output reveals a missing constraint, so those experts belong in model-building.

2026-08-04 · open the original ↗

• His top advice is to stay unlevered right now. He argues markets are producing events that "haven't happened in a long time or ever," at increasing frequency, making today's range of possible outcomes the widest he's seen in more than 30 years investing.
• He calls himself bearish near-term but bullish over the next decade-plus. He frames surviving the current period as the price of admission for what he expects to play out "very bullishly" over the next ten-plus years.
• He recommends holding some gold alongside staying unlevered. On that basis, he predicts most investors positioned this way will be "real happy" with where they stand in five to ten years.

2026-08-03 · open the original ↗

• A working demo is "1% at best" of a real product. Waymo Co-CEO Dmitri Dolgov says Waymo hit its first fully capable driving demo in 18 months back in 2010, but needed roughly 15 more years to reach scaled service, because each additional "nine" of reliability costs about ten times more effort than the last — meaning demos get easier with every AI breakthrough while the hard tail "moves much less."
• Waymo bets on redundant sensors because camera-only sensing plateaus too early. Dolgov argues matching human driving performance is achievable with cameras alone, but reaching strongly superhuman safety requires cameras, lidar, and radar fused together, since each fails differently — lidar catches a pedestrian invisible to cameras in a dust storm, for instance, while radar sees through fog that blinds both.
• Waymo deliberately keeps structure rather than going pure end-to-end. Dolgov argues the AI field's "bitter lesson," that scale beats hand-engineering, only holds when structure fights scale; structure that instead channels it — like an intermediate representation a safety layer can validate in real time — improves both performance and evaluability, which is why Waymo calls its approach "structure-augmented end-to-end."
• Eval and metrics, not the model, are the real strategic moat. Dolgov says Waymo builds its evaluation framework before its technology, using closed-loop simulation and a shared "agent, simulator, and critic" system to train safely; the resulting safety record — over 220 million fully autonomous miles, roughly 17 times fewer serious-injury crashes than human drivers, a serious injury prevented roughly every 8 days — is, he says, far harder for a competitor to replicate than any model or algorithm.

2026-08-03 · open the original ↗

• Trump says the oil majors' war profits belong to the public. Exxon Mobil and Chevron posted a combined $26.5 billion profit in one quarter, their best in years, which he attributes to the Iran war lifting oil from below $60 to over $100 a barrel and to soaring refining margins. He notes Republicans would call such a demand communist from a Democrat.
• The oil majors sold into a price spike they did not cause. He considers the profiteering critique legitimate but disagrees with it, arguing Exxon and Chevron did not bomb Iran or push oil to $120, and asking whether they should withhold supply to avoid the label.
• Iran alleges Trump's advisers traded on inside knowledge of ceasefire talks. Iran names Steve Witkoff and Jared Kushner and puts their gains at $9 billion. He stresses this is an unproven allegation from a wartime adversary; the White House denies it and the Vice President called it bogus. He says US oil analysts and traders raised it first, and he believes it happened because bets of up to $1.5 billion shorting oil were placed moments before announcements.
• Scrutiny of war profits disappears as it nears Trump's circle. He argues the same question draws "give the money back" for a faceless corporation and "fake news" for the president's advisers, the only variable being proximity to the president.
• Alleged insider trading deserves an independent investigation, unlike public earnings. He distinguishes legal, published corporate profits from secret trading on classified negotiations, which he likens to insider trading with a security clearance. If oil companies should open their books, he says, so should the advisers.

2026-08-03 · open the original ↗

• This post includes paid content that is not available publicly. See the original source.

2026-08-03 · open the original ↗

• A sharp divergence split mega-cap tech into winners and losers on the same day. Amazon, Google, Meta and Microsoft each rose $14-$26 while S&P futures were up 47 points, but Apple, AMD, Micron, Nvidia, Broadcom and Goldman Sachs all fell, with AMD and names like SanDisk and Bloom Energy down roughly $15 and Marvell down about $7, a 3.5% move.

2026-08-03 · open the original ↗

• This post includes paid content that is not available publicly. See the original source.

2026-08-03 · open the original ↗

• Lossdog pitches a "professional equity report" as a recruiter-ready snapshot of a user's market value. It's described as unlocking once a profile is complete, generating a downloadable report meant to quantify a user's compensation worth with data recruiters can use.
• A trade-feed update lets users follow Tom Sosnoff's real-time positions, with futures coming soon. Positions are timestamped across strategies and asset classes, pitched as a new way to track a named trader's activity on the platform.
• The episode also claims New York has approved a crypto "release" users can claim through the platform. It directs New York users to open an in-platform crypto wallet and follow prompts to claim funds — a claim reported here as stated, without further detail on its source or structure.

2026-08-03 · open the original ↗

• This entry is a video recording of an earlier written essay, not new material. Patel states it reproduces "Why compute might get 10x+ more expensive in coming years," a post he published the previous week (2026-07-29); the substantive argument lives in that original piece.
• The provided post body contains only a sponsor read, not commentary. It describes Mercury's built-in AI tool, Command, which categorizes transactions monthly and explains its rationale for each choice before syncing to QuickBooks once reviewed and approved.
• No independent transcript or analysis is available for this entry beyond the pointer and the ad. The written original carries the compute-pricing argument and should be treated as the source for that content.

2026-08-03 · open the original ↗

• The FIMA repo facility, not the FX purchase, is the substantive support — and it is not a bailout. Japan holds $1.1–1.4T of Treasuries in custody at the Fed and can repo them for dollars rather than dumping them into the market. The loan is fully collateralized with haircuts applied, runs overnight to seven days, and is priced above private repo rates so it is used in stress rather than routine times. The purpose is keeping a forced seller out of an already-pressured Treasury market, not defending the yen as such.
• The execution details reveal how deliberate this was. The New York Fed conducted the euro sale to buy yen on Treasury's behalf, routed through Goldman Sachs and Morgan Stanley, having signalled to Wall Street on July 23 that intervention was under consideration when the yen hit its weakest level since 1986. This post reports US authorities were in touch with ECB officials about the move — worth noting against his later account, which emphasizes that the ECB was informed only after execution.
• He reads the euro choice as evidence of dollar funding stress inside the Fed system. The US had ample dollars and used euros anyway, which he takes as the first concrete corroboration of the liquidity dearth being discussed in funding markets, alongside the Fed's experimentation with new bank funding channels. He adds that the UAE has reportedly asked for a $10B Fed facility, framing this as war-induced stress arriving country by country, each with a different ask — and concedes nobody, himself included, knows whether these measures work.

2026-08-03 · open the original ↗

• A Q&A site should build juniors' understanding, not just answer. The questioner calls juniors "my number one thing right now" and wants a Stack Overflow that says "help me understand this" rather than "solve this for me".
• The guest has seen juniors fix things without the judgment to repeat it. The guest says that at GitHub, newcomers and interns fixed things "with Copilot entirely" but lacked the reasoning to do it again, which they see as a reason to keep Stack Overflow human-first. They also concede that the moderator community can leave a junior "eaten alive" for asking, and say this is being worked on.

2026-08-03 · open the original ↗

• This post includes paid content that is not available publicly. See the original source.

2026-08-03 · open the original ↗

• The corporate-debt "wall of maturities" won't shock markets. Batista calls it a known, long-discussed risk that will have at most a moderate effect; Sosnoff agrees, arguing today's 9–10% corporate borrowing rates only look extreme against a decade of near-zero debt, since the 30-year average mortgage rate is closer to 5% and 1980s/late-90s rates ran 7–8%.
• Sosnoff holds up Ken Griffin as the model "opportunist." Discussing a young manager's fund selling roughly $16 billion in stock to Citadel after falling from a $45 billion peak, he recalls Griffin refusing in 2008 to flip him E-Trade shares Citadel had just rescued at $1, saying other buys looked better — a call Sosnoff says took two years to prove out.
• A 2.2-day short-covering ratio rules out a genuine SpaceX squeeze, Sheridan argues. Short interest rose roughly 40% to about 165 million shares by mid-July, but days-to-cover near 2.2 (versus 25–45 during meme-stock manias) means shorts aren't trapped; both hosts say Thursday's earnings, not the lock-up releasing 911.5 million shares, will drive the stock.
• Both hosts expect zero price impact whenever the Clarity Act finally passes. Batista calls Bitcoin "dead money," stuck between the 50s and high 60s (thousands) for months; the House passed the bill in July 2025, but Senate leadership called an August-recess vote unlikely, and neither host thinks approval would spark the 10–20% pop some traders anticipate.

2026-08-03 · open the original ↗

• This post includes paid content that is not available publicly. See the original source.

2026-08-03 · open the original ↗

• Project Nomad is running on a Raspberry Pi 5. The host's test unit has 8GB of RAM and a 1TB NVMe drive attached via a Pimoroni NVMe base rather than microSD; he says it's working well but not yet ready for primetime.
• The ARM platform is too weak for Project Nomad's AI inference. Users will need to skip the AI feature entirely or point it at a remote Ollama host to use the LLM; everything else — full Wikipedia, global street-level maps, K-12 education, and hundreds of survival, medicine and agriculture books and guides — still works.
• ARM support is targeted for Project Nomad's next major release. Version 1.34 ships this week; the host says Nomad will remain 100% free and open source as ARM compatibility is added.

2026-08-03 · open the original ↗

• The White House announced a deal that Iran flatly denies. Over the weekend the White House said strikes were cancelled, "parameters of a deal" agreed, Hormuz would open immediately and Iran would give up its nuclear program; Iran replied there is no agreement. He reads "parameters" as what one says when there is no deal.
• Iran fired on shipping almost as the deal was announced. Anti-ship missiles were launched at a tanker within 2 minutes of Trump saying there was a deal to open Hormuz. He calls this the tell: the strait the deal was meant to reopen was effectively closed again as the deal was being declared.
• Vague announced deals have produced the next round of war. He points to the June memorandum six weeks earlier, celebrated as opening the strait, whose ambiguity over who controlled it left the two sides with different readings and, in his view, caused the current phase.
• Iran refused an offer giving it effective control of the strait. Citing Reuters, Al Jazeera and the BBC, the last US offer gave Iran inbound control and Oman outbound control, which he says amounts to full control, since there is only one way out of the Persian Gulf. He infers Iran wants more and expects any eventual deal to be very bad.
• Judge any deal by tanker movement, not announcements. If ships move next week there was a deal; if the strait keeps opening and closing and missiles fly, he calls it theatre. He reads renewed COVID and women's sports debates as distractions from the lack of a deal.

2026-08-03 · open the original ↗

• This post includes paid content that is not available publicly. See the original source.

2026-08-03 · open the original ↗

• He calls big-tech AI equities extremely complacent right now. He argues that segment, a large share of the largest equity market, has become cash-negative and debt-financed even as the rate it borrows at keeps rising — "a very bad combination" whose breaking point he says he can't time.
• He frames strong equities as a currency-debasement signal, not health. He likens the situation to "an emerging market debt problem with US, UK, German and Japanese characteristics," noting the S&P's total return priced in gold is still down 40% from the January 2000 dot-com highs and down 8% since the fourth quarter of 2018.
• He says the Fed's real mandate is Treasury market functioning, not prices. He argues this has been proven "five times in six years," and concludes investors should avoid long-term bonds, buy equity dips instead, and that equities are rational in both dollar and gold terms.