Souk's Notes - Unofficial Summaries

2026-05-28 · open the original ↗

• Data curation is the unloved stage that dominates the work. Henry calls it the step everybody hates and says her team has spent the past 6 months on data curation alone, with dedicated staff. A thousand-page textbook may introduce a topic in chapter 1 and return to it in chapters 5 and 20, so subject matter experts help attach terms correctly in a graph database. She jokes that her company is as much a data company as an AI company, and the host ties this to the main point of episode 993 with Jacob Miller and Jeremy Mumford.
• Custom tokenizers and embeddings avoid expensive vision models for documents. She describes building bespoke tokenizers and embedding models on top of them. These let drilling documents be chunked and OCR'd at the text level, rather than drawing bounding boxes with a costlier vision model.
• Physics belongs in RL environments, not in model weights. Model building combines continued pre-training of open-source models on oil and gas (chosen after constant benchmarking against open and API models) with reinforcement learning. Teaching a small model physics directly makes it overfit to physics, she says. Instead, fixed physical rules sit inside an environment where the model writes Python in a REPL.
• How a model is trained dictates how it must be served. Henry says reinforcement-learning models are bursty, so meeting 99.9% uptime under SLAs requires a large, costly inference cluster. She calls inference the stage she finds most exciting and most people find least interesting.

2026-05-27 · open the original ↗

• His core test for dictatorship is a leader outlasting constitutional term limits. By that measure he contrasts Ukraine's four different presidents since 2000 — including Viktor Yanukovych, ousted by the 2014 Maidan Revolution and fleeing to Russia — with Russia and Belarus, where the same leaders have effectively remained in power.
• He argues Putin and Lukashenko rewrote their constitutions to stay in office. Putin changed term limits in 2008 (two four-year to two six-year terms) and again in 2020, resetting the clock so he can legally rule until 2036; Lukashenko removed term limits in 2004 and reinstated them in 2022 while exempting himself.
• He defends Zelenskyy's suspended 2024 elections as constitutional, not a power grab. He cites Article 83 of Ukraine's constitution, which permits suspending elections under martial law, concluding Zelenskyy is the only sitting leader among the three who hasn't violated his own constitution to remain in power.

2026-05-27 · open the original ↗

• The verified charges are narrow, specific, and already partly beyond reach. On 19 May 2026 the DOJ unsealed a Northern District of California indictment — returned October 2025, filed January 2026 — charging four Chinese container manufacturers (CIMC, Singamas, Shanghai Universal Logistics/Dong Fang, CXIC) and seven executives with a single Sherman Act §1 count for restricting output and fixing prices on standard dry containers between November 2019 and at least January 2024. One CIMC subsidiary is state-owned via COSCO. Six of the seven executives remain in the PRC; one was arrested in France in April 2026 awaiting extradition.
• The alleged mechanics are unusually concrete for a cartel case. Production-line shift and hour caps; 87 surveillance cameras across 49 dry-container production lines for mutual monitoring; an explicit prohibition on building new container factories; a penalty fund for over-production; customer-specific quotas from September 2020 expanding into comprehensive total-allowable-capacity caps by late 2022; and a coordinated price war against non-conspirator small manufacturers in August 2023. Scale claims: over $35B of affected commerce, CIMC's container-segment profit rising roughly 100x from 2019 to 2021, and container prices roughly doubling across that window.
• The fact-check corrects the framing that prompted the analysis, in both directions. The "over 95%" share belongs to six conspirators including two unnamed firms; the four named alone are closer to 87%. The $35B is DOJ's own "early estimate" and a government allegation rather than an adjudicated finding, as is the price-war language quoted from the indictment. But one number is understated rather than overstated: nearly 100% of the world's refrigerated containers are China-built, a separate and tighter concentration. Market reaction was muted — the Hong Kong-listed shares fell 1.5–1.6% on unsealing, consistent with a low priced probability of corporate penalty ever crossing the Pacific.
• Two structural observations carry the analysis, and both cut against the enforcement working. First, this is paper power on both sides: a 95% position held not through physical superiority but through coordinated paper agreements, met by indictments against executives outside US jurisdiction and firms with limited US asset exposure — two paper systems contesting a physical-flow chokepoint, with neither able to alter the physical reality. Second, the reflexive signalling trap: an indictment announcing that PRC supply is legally fraught raises the urgency for counterparties to lock in long-dated supply with the indicted firms, because no alternative capacity exists at scale — so the action strengthens the named firms' position. The paper proposes "Enforcement Substitution" as a term for antitrust deployed as industrial-policy proxy where onshore alternatives do not exist, and registers the forward-looking claims as falsifiable predictions with named watch sources rather than asserting them.

2026-05-27 · open the original ↗

• The indictment is punishment applied to a chokepoint with nowhere to switch to, which he treats as the whole point. Four Chinese firms holding roughly 95% of the dry container market are charged under US law, while the United States has closed its last container production line entirely. CIMC alone runs 2 million TEU of capacity across 11 plants; the one plant outside China, in Vietnam, produces 200,000 TEU and hopes for 500,000 by 2029 — and even that depends on Chinese-origin coatings, hardware and fittings. His question to shippers who need containers is simply: where would you go?
• Every use of paper power draws down a finite reserve, and he expects this one to accelerate the split. Rather than disengaging, counterparties will build a dual system and shift settlement toward the yuan, exactly as happened with the teapot refiners. Meanwhile MOFCOM's blocking mechanism is already in place to counter extraterritorial reach. And the enforcement itself is largely unenforceable — six of seven indicted executives sit in the PRC, and he asks what China would do if the one arrested in Europe were actually extradited and imprisoned.
• The surrounding week is what makes the sequence legible. The summit produced cooperative language, but China then banned Nvidia's Blackwell gaming chip while Huang was still in the country, published a fully domestic chip technology stack with no US content, and passed blocking rules against extraterritorial jurisdiction on 15–16 May — after which Washington struck back on the 19th with this indictment. He also updates his MP Materials read: trading around $66 against a conventional $24–40 valuation with a consensus bull case of $90–100, where the premium is explicitly pricing a near-monopoly on supply the US must now source domestically.

2026-05-27 · open the original ↗

• One early webinar viewer became a multi-billion-dollar money manager, the host says. A friend who started with no market knowledge while watching thinkorswim webinars around 2007-08 later became, by the host's account, the only person in his graduating class who could explain volatility; he went on to manage an estimated four to five billion dollars.
• The host claims that story is not an outlier — 50 to 100 similar cases exist. Most of those, he says, reached the hundreds of millions or tens of millions rather than billions, offered as evidence that people who started learning markets through the hosts' past teaching have gone on to manage money at a range of success levels.

2026-05-27 · open the original ↗

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

2026-05-27 · open the original ↗

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

2026-05-27 · open the original ↗

• A single shared database, not better tools, was YC's real unlock for internal agents. Partner Pete Koomen says YC runs on its own software sitting on one Postgres database covering every company, founder and financial record, so an agent with read-only SQL access could answer arbitrary questions; he argues this collapsed the friction of cross-team requests enough to trigger a Jevons-paradox-style explosion in the number and complexity of questions people bothered to ask.
• A tool registry grew from about 20 tools to over 350 as teams kept adding their own. Koomen describes YC's internal harness evolving from a simple agent loop into a shared registry that any team can extend, and says the key move was building a "resolver" — an agents.md-style entry point linking to markdown skills — so agents could discover and reuse capabilities instead of duplicating them, checked for redundancy with what Garry Tan calls a "dry and MECE" meta-skill.
• YC deliberately broadcasts every internal agent conversation to Slack by default. Koomen says this transparency, chosen despite internal debate, let employees learn techniques by watching others use agents and created a social check that made it safer to give agents broad access to sensitive context — a trade-off he argues requires an "egalitarian, trust by default" culture that works best in small, aligned organizations.
• A single partner's handwritten skill for writing "two-sentence descriptions" became better than any individual partner at YC. After other partners fed a group office-hours transcript back to the agent to refine the skill, Koomen says it now writes better two-sentence founder pitches than he does personally, illustrating what he and Tan call building "superintelligence inside a company": composing many individually improved skills into a shared organizational capability.

2026-05-27 · open the original ↗

• Copilot did not create the permissions problem; it exposed it. Mantripragada, who says she spent 15–20 years as a SharePoint and Power Platform specialist at the SEC, Freddie Mac, Blue Cross Blue Shield and Gilead, likens AI to switching the lights on in a messy room: overshared, ungoverned data that sat unnoticed is now surfaced, as when an intern asks a new agent for "the strategy for the next 2 years" and gets an HR site's documents.
• Every agent should pass an intake review before it is built. Her "just-in-time AI governance framework" records who is building it, the approver, the business case and the intent as an audit trail, then on a button click checks the named data sources through the Graph API, Purview and SharePoint Advanced Management for overshared or orphaned sites, sensitive data and DLP settings, returning either a remediation plan or an approval. She says it runs inside the customer's tenant on a Teams/SharePoint site, covers SharePoint, OneDrive and Dataverse so far, is built largely in Power Automate, and is currently available only by hiring her firm, Low Code Power, with general availability a goal.
• Proactive governance beats leaving it to auditors after the fact. Her tagline is "build fast, govern faster": citizen developers need a "driver's manual" up front rather than a ticket later, and she says that when AI goes wrong the repercussions now reach the CEO, not just the IT director. She calls Microsoft's Agent 365, which she says went generally available on May 1st, the coming go-to for reporting on workflows and usage but "slightly reactive"; she ties auditability needs to the EU AI Act and the NIST AI RMF.
• Start with one or two high-effort workflows and a few champions. She picks the most manual, time-consuming work, found by talking to the domain teams, because quick results drive adoption. Her example is an R&D tax credits firm whose customer-vetting document has about 25 fields: a Copilot agent plus an agent flow, added for deterministic results, now fills a Word template with a company profile on one click.
• Her readiness assessment begins by documenting the environment as-is. She describes spending the first week on the customer's environments, workflows, infrastructure and licences, documenting the as-is and to-be process, then using Purview, the M365 and Power Platform admin centres and SharePoint Advanced Management reports to find exposures such as sites open to everyone or missing security groups.

2026-05-27 · open the original ↗

• Oil and gas knowledge is being lost as a generation retires. Henry describes a hiring drop-off from the '90s to the early 2000s, before activity picked up again in the shale revolution of the 2010s. The knowledge held by older workers is disappearing, and younger staff don't know where things are.
• AI can synthesize the industry's largely unstructured documents. She says 90% of oil and gas documents are unstructured, much of it handwritten, covering leasing, titling, safety and regulation. Her company applies AI across them, and she frames the safety side as work that saves lives. She says its leader has worked in fracking for over 20 years and the oil-and-gas staff have over 100 years of experience combined.

2026-05-27 · open the original ↗

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

2026-05-26 · open the original ↗

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

2026-05-26 · open the original ↗

• Commencement speakers get booed for warning that AI could make degrees obsolete because the message devalues the degree itself. One host adds that such speakers also draw criticism over AI's ecological impact, arguing both points undercut the celebratory, "go take over the world" tone graduates expect from a commencement speech.

2026-05-26 · open the original ↗

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

2026-05-26 · open the original ↗

• The US and Iran exchanged strikes while both sides still call the ceasefire active. He reports Iran mined the Strait of Hormuz with two ships, the US destroyed them, Iran retaliated with drones and missiles, and the US destroyed the launch site — yet the administration maintains a ceasefire is still officially in place.
• He argues this effectively ends the Pakistan-brokered peace track. In his reading, the exchange collapses the potential US-Iran peace deal being brokered by Pakistan and closes off any chance of Iran transferring its enriched uranium to China.
• He flags a warning-sign gap: Iran's pre-attack jamming and airspace closure haven't reappeared. Two days before this strike, Iran had jammed GPS over Kuwait and Qatar and shut its airspace; he says he hasn't seen those signals recur since, leaving open whether Iran will retaliate against neighboring states hosting US bases.

2026-05-26 · open the original ↗

• He predicts Iran will be struck imminently, citing closed airspace and IRGC threats. He points to Iran's maximum alert status, heavy fighter jet activity, and jamming over Qatar and Kuwait, alongside IRGC threats to cut seven undersea fiber-optic cables, close the Bab el-Mandeb Strait "by fire," and deploy unseen third-generation drones and missiles.
• He reports three simultaneous US-side developments he reads as connected. An alleged IRGC assassination plot against Ivanka Trump was uncovered, Tulsi Gabbard abruptly resigned citing her husband's health (which he doubts, suspecting her anti-war stance instead), and the president left a New York/New Jersey meeting for Washington on short notice.
• He predicts Iran will be hit harder than before and expects further Gulf-state attacks. He argues Iran's Strait of Hormuz closure was anticipated by military planners but its strikes on US-basing neighbors were not; he expects the US and its allies used the ceasefire to arm up, and dismisses Chinese/Russian-supplied equipment as inferior to American gear.

2026-05-26 · open the original ↗

• The document exists to falsify its own prior entry, and it does exactly that. The anchor premise of the original Busan Architecture entry — a stated commitment that China would "effectively eliminate" critical mineral export controls — is marked FALSIFIED. The May 2026 Beijing summit fact sheet softened the language to "address concerns," a deliberate diplomatic downgrade, and the dispositive empirical signal is yttrium customs data: roughly 30 tons a month before controls, 60 tons in March 2026, and 10 tons in April — an 83% collapse. The de-escalation envelope was never operationally delivered.
• The audit method is stricter than most self-assessments, and it says so. Projections are pulled verbatim from the source entry with no retroactive softening; every framework concept receives an explicit fate — formalized, validated, refined, falsified, new candidate or unresolved; and the same two-instance evidentiary bar used for entry inclusion applies to outcomes. It even discloses that the projections were reconstructed from memory and must be cross-checked against the underlying entry files before being treated as source of truth.
• Falsifying the anchor did not falsify the framework — six concepts survived or hardened. Sovereign Licensing Defection is formalized on four instances, including the China-Japan dual-use ban and the yttrium element-specific throttle. Codified Triage is validated operationally through DPA III and the $96M DoD-Lynas letter of intent against the Ford Chicago halt. The five-tier Civilian Production Moat is validated empirically as Tier 1 fills while Tier 2 stops. The 40–65K ton NdFeB gap is validated and accelerated — the yttrium collapse proves it binds now rather than in 2030. And Endogenous Ceasefire is validated in its purest form: the settlement itself generated the next round of friction, producing State Council Order 834 and the Mineral Resources Law effective 15 June.
• The investment implication is stated without hedging. Any portfolio construction that assumed Busan was a binding constraint on Chinese coercion has been operating on a falsified premise since January 2026 at the latest, and the Mineral Resources Law institutionalizes precisely what Busan was supposed to constrain. The audit also forces a new candidate concept — Codified Coercion Architecture — for the mirror codification now appearing on the extraction side.

2026-05-26 · open the original ↗

• The tell he noticed was who stayed behind, and what happened to him. Huang initially declined the trip, was personally called by Trump, flew his own jet to Anchorage to board the presidential aircraft, and was the only chief executive to remain in Beijing afterwards — during which China banned his Blackwell gaming chip, in what he believes was its largest gaming market. Days after a summit publicly framed as easing rare-earth and chip restrictions, China announced its reserve law doing the opposite.
• On the revised reading, Busan was never a settlement — it was a positioning pause. The architecture's character was misread: the de-escalation language provided cover under which Chinese statutory architecture hardened. They talked accommodatingly and passed laws that were not accommodating, so what exists now is a codified structure with no negotiated exit. His formulation is the title's: when law supersedes agreement, negotiation loses its allowance and leverage becomes an illusion.
• The allocation consequences are already visible and directly investable. Yttrium, of which the US imports 93% from China, has gone from a normal ~450 tons to 60 and then 10. Ford's Chicago plant has closed and EV producers are being deprioritized under triage, which he flags as a negative read-across to Japanese autos. Lynas is the structural beneficiary — Australian mining, Malaysian LAMP processing, and a Texas magnet plant targeted for Q4 2026 that he names as the single most important thing to watch — but with a genuine constraint he does not talk past: Malaysian street protests against sending heavy rare earths to the United States.

2026-05-26 · open the original ↗

• The host bills this as unbiased, but tests with his own SIM. He says he has left the Ubiquiti creator program, no longer gets free devices and has no affiliate codes; the review uses a Crosstalk Mobile 5G unlimited SIM, which his company sells as a T-Mobile business MVNO, and pitches its $5-a-month static IP and non-CGNAT addressing along the way.
• The hardware is a standard UniFi gateway with dual-SIM 5G built in. It has two nano SIM slots, with slot 2 also able to take an eSIM, plus four 2.5 GbE ports, a 10 Gb SFP+ port, a microSD slot, a touchscreen and a fan. Only port 1 has PoE, standard PoE at 15.4 W rather than PoE+, and it can be assigned as a WAN so a PoE-powered WAN device needs no extra brick.
• Setup worked out of the box, with two quirks the host hit. The SIM came up as WAN 3, and he measured 150 Mbps down and 48 Mbps up untuned in mid-office. Setting the custom APN for a static IP (iot.cc.static) took the console offline until he rebooted. The APN setting is only in the web interface, not the phone app, and the LED colour is only on the touchscreen.
• Ubiquiti says T-Mobile Home Internet SIMs are not supported. Some commenters on his earlier 5G Max video said theirs did work, which the host has not tested. He reports port forwarding through the separate UniFi 5G Max add-on modem only started working after its most recent update, and expects it to work on this device too.
• The host recommends it for small businesses needing 5G, if in stock. At $499 MSRP against $399 for the 5G Max add-on, upgrading an older gateway costs little more and is fully integrated; it was out of stock at recording. Home users with a UniFi gateway should instead add T-Mobile or Verizon home internet as WAN 2, at about $50 a month or less ($25 on promotion), since CGNAT is usually fine at home.

2026-05-26 · open the original ↗

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

2026-05-26 · open the original ↗

• A sycophantic general model is dangerous where lives are at stake. Henry says a general model asked about oil and gas terms may know some definitions but not how they relate. Given a job safety report (JSA), it may read the document without understanding it and say "everything looks good". Her company's domain models are meant to check local rules, compare the data and push back, because a wrong all-clear can cost someone an arm.
• Physics belongs in RL environments rather than in the weights. Henry describes four stages: data curation (6 months so far, with subject matter experts building a graph database), bespoke tokenizers and embeddings that let documents be read at the text level instead of by a costlier vision model, model building, and inference. Model building combines continued pre-training of benchmarked open-source models with reinforcement learning. Teaching a small model physics directly makes it overfit, she says, so fixed physical rules sit in an environment where the model writes Python in a REPL.
• RL training is bursty, so GPUs must be provisioned conservatively. Reward calculations run on CPU while the GPU sits idle, then a spike of data hits it all at once. An 8-billion parameter model that fits on one A100 then needs several, and people moving into RL hit seg faults from overloaded GPUs. Henry's fix is to scale back the 90% of memory people often give to the model and KV cache.
• Early Monte Carlo forecasts catch failing runs and reward hacking. Henry's grounded continuous evaluation (GCE), from a paper the host names as Beyond Static Snapshots, simulates where a run is heading from about an hour of reward data instead of waiting 2 hours to days for a checkpoint. A strong checkpoint after a weak forecast flags reward hacking, such as models under GRPO keying on keywords to get the final answer without learning the equation. She says the deterministic environments remove the need for a reference model, cut compute and burstiness, and let models work on a MacBook.
• Scaling bad data makes models worse, not better. Her 2023 NeurIPS workshop paper, accepted as what she calls a top 5% paper, argued against the Kaplan et al. scaling view by applying diminishing returns. She says undeduplicated data, nearly impossible to clean at that size, makes models forgetful and overfit, and no number of GPUs fixes it. She sees this vindicated in 70 billion or 120 billion parameter models outperforming 1 trillion-plus models.

2026-05-26 · open the original ↗

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

2026-05-25 · open the original ↗

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

2026-05-25 · open the original ↗

• The Copilot app is built for issue and PR work. Hanselman, who says they work 50% at GitHub and 50% at Windows as an individual contributor and have been working on the app, places it between VS Code (code-focused) and the terminal: "It is not a text editor... It is not a terminal." The video was posted a week after launch, with the app in technical preview (sign-up via the GitHub changelog) and "more in store" at Microsoft Build on June 2nd.
• Each task-based session gets its own worktree. Sessions sit under projects and can target a branch or an issue; Hanselman shows 18 worktrees in Worktrunk that the app created, and says a new session can run against the local repository or a worktree.
• Model choice allows adversarial reviews, with the human as judge. Hanselman picks models per session, can set two models to argue over a PR, and runs agents in parallel; plans wait for approval, which Hanselman says gets read first. An "auto" model setting gives a discount.
• Terminal, browser and scheduled automation live inside the app. Side panes show changes, plans, a terminal and a browser for localhost URLs, with "pick and polish" for selecting page elements; scheduled workflows run triage in the background on the local machine, and sessions can go remote or open in VS Code, Visual Studio or Cursor. Hanselman says the app means leaving for the terminal less than before.

2026-05-25 · open the original ↗

• You can buy two of intelligence, cost and latency, not all three. The host takes this triangle from the guests' book Architected Intelligence, along with its techniques for "cheating" it: dynamic routing, streaming, semantic caching and ensembles.
• Streaming is the most effective cheat; ensembles are the most underused. The guest answering cites, without being sure of the figure, a statistic that every added 100 milliseconds loses about 10% of users. Caching is now easy because OpenAI and Anthropic often apply it automatically, with semantic caching as an optional extra layer. The guest sees ensemble methods gaining interest since Andrej Karpathy released his "Council of LLMs" a few months earlier.
• On-the-fly routing by difficulty is the hardest cheat. The guest notes that adaptive thinking in ChatGPT and Anthropic's products drew complaints: "having it try to guess which to do bothers everyone." The advice is to leave on-the-fly routing to the labs, and to route only where a deterministic rule for a given problem set exists.
• Good UX design can hide LLM latency entirely. The guest likens it to Instagram uploading a photo while the user types the caption. Designers can start AI jobs early and walk the user through other steps so the result appears instantly. Slow LLMs, they argue, mean engineers must relearn latency tricks that faster internet let them forget.

2026-05-25 · open the original ↗

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

2026-05-25 · open the original ↗

• Guano imperialism is the historical precedent for winner-take-all resource grabs. The U.S. Guano Islands Act empowered citizens and the Navy to seize unclaimed islands with bird-dropping deposits, and 19th-century Peru derived roughly 60% of national revenue from guano on the Chincha Islands, illustrating how resource scarcity once drove overt imperial conquest.
• Post-WWII American soft power, not military dominance, built 80 years of relative peace. Institutions like NATO, the UN, and USAID connected nations economically and politically, deterring war as a dispute-resolution tool; the Marshall Plan delivered roughly $13.3 billion (about 90% as grants) explicitly branded "from America" with no equity stake attached.
• That soft-power architecture is being dismantled and replaced with transactional leverage. USAID was shut down by executive action in 2025 (a federal judge called it unconstitutional before Congress retroactively restored funding), Voice of America is a shell of itself, and remaining aid now comes with strings, equity stakes, and demands attached.
• The shift from soft power to hard power carries a body count. Operation Southern Spear has killed more than 160 people in strikes on alleged drug boats near Venezuela since September 2025 under a self-declared "armed conflict," with little public evidence released, while the same military buildup captured Venezuela's president in January.

2026-05-24 · open the original ↗

• The DoD agreement created two stacked sovereign instruments on one company, which is why a single discount rate misprices it. An NdPr oxide floor of $110/kg for ten years, and a separate 10X magnet facility EBITDA floor of $140M a year indexed at 2% with quarterly facilitation payments over ten years from commercial operation. Applying conventional commodity-equity DCF across the whole company collapses four genuinely different risk layers into one number.
• Valued tranche by tranche at its appropriate cost of capital, the pieces look nothing alike. The EBITDA floor is functionally a sovereign-issued inflation-linked annuity — a growing annuity discounted at 6% over ten years gives $1.10B at commercial operation, $980M discounted back two years, and roughly $785M after a 20% execution haircut. The NdPr floor carries a $70/kg floored gross margin ($110 floor against a ~$40 target unit cost at scale) but swings by a factor of two on volume: $1.83B at 5,000 MT steady state against $3.67B at 10,000 MT, and that $1.8B gap is the honest structural ambiguity. The above-floor call is essentially at-the-money with spot at $99–112/kg, uncapped — unlike Lynas, where JARE captures 30% above $150/kg — worth roughly $900M to $1.4B.
• The conclusion is a large premium, and the paper names its own weakest link rather than burying it. Fair value lands at $25–40 against a $60.31 close, a 49–142% premium, which it reads as the market pricing structural value sitting outside the SCS architecture itself. But NdPr has no deep liquid options market, so the Black-Scholes-derived option value is explicitly flagged as indicative rather than authoritative, with the volatility assumption dominating the answer — 30% versus 60% annualized produces materially different numbers, and the value is best held as an order of magnitude rather than a point estimate.

2026-05-24 · open the original ↗

• The claim is categorical, not a call for recalibration. Conventional equity tools were built for instruments that are a single risk class with parameter variation across companies. Sovereign Credit Substitution violates that at the architectural level, producing a multi-tranche security whose tranches carry fundamentally different risk classes, each needing its own methodology. So the question is not whether the existing toolkit needs tuning but whether it belongs to the right category at all — and the paper's answer is no.
• Each standard metric breaks at a distinct and identifiable point. Free cash flow treats sovereign facilitation payments as commodity-cyclical when they are functionally sovereign coupons, overstating the cyclical risk discount. A single EV/EBITDA multiple applies the same multiple to sovereign-anchored and commodity-equity EBITDA. Debt-to-equity becomes meaningless when a sovereign Series A Preferred with PIK dividend, warrants, no maturity and a government holder sits between debt and equity and is neither. Revenue confuses commercial expansion with zero-margin sovereign cost pass-through. Growth blends a commodity production ramp with inflation-indexed sovereign growth. And blended beta is statistically meaningless when the floored portion has beta near zero and the residual carries commodity-equity beta — leaving CAPM-derived cost of equity wrong by 400+ basis points in one direction or the other.
• Three clean historical precedents establish both the pattern and the timeline. REITs after 1960 needed roughly fifteen years to develop FFO/AFFO and cap-rate-spread methods; MLPs after 1986 about ten to arrive at distributable cash flow and coverage ratios; BDCs five to ten for NAV-based valuation and P/NAV multiples. In every case a regulatory or contractual feature created a security conventional analysis mispriced systematically, sell-side coverage lagged by years, and the mispricing was real and exploitable during the transition before the new framework became as familiar as the one it replaced.
• The honest framing is that the mispricing runs both ways, not just cheap. REITs were systematically undervalued from the late 1970s through the mid-1980s while analysts could not price the tax pass-through, then systematically overvalued in 1997–99 when cap rates collapsed faster than fundamentals justified. The pattern is not directional: the market gets the risk class wrong, and the sign of the error depends on what it is comparing against at the time. SCS sits at year zero of that transition, which is itself the structural source of mispricing.

2026-05-24 · open the original ↗

• His claim to recognise the pattern is a specific one: he brought these asset classes public. Nearly fifteen years in capital markets and, by his account, more MLP and REIT IPOs led than anyone — both new asset classes at the time, each requiring valuation methods that did not yet exist. He sees the same thing now with MP and Lynas, and expects it again with quantum computing.
• The line he pulls out of the analysis is the one worth keeping. MP at $60.31 against a modelled $24.93 base case and $40.40 optimistic — premiums of 142% and 50%. But the durable formulation is that Sovereign Credit Substitution does not price the equity; it puts a floor under the bear case. Each tranche needs its own comparable: sovereign-backed cash flows against finance bonds, the above-floor exposure against commodity derivatives, the residual against conventional mining equity, the preferred against structured credit and convertibles — and for pre-revenue names being handed government money, something closer to biotech pre-approval valuation.
• He gives four concrete signals for when the methodology actually arrives. Sell-side analysts beginning to disaggregate the tranches in published research; credit agencies treating sovereign-floored cash flows as quasi-investment-grade or sovereign; index providers creating critical-minerals-plus-sovereign-backed sub-indices; and specialised funds emerging with explicit SCS equity mandates. He expects the structure to spread to uranium, chips, the defence industrial base and allied LNG export terminals, so this is the beginning rather than the end — and reframes the investor's question accordingly: not whether the stock is cheap or expensive, but whether it is being analysed with the right tools at all, which most coverage will get wrong for years.

2026-05-24 · open the original ↗

• The interview identifies the author and explains where the framework's shape comes from. Daniel Gallagher spent decades on Wall Street including senior roles in wealth advisory at Morgan Stanley and UBS before founding Process for Growth. His father was the tenth man on DuPont's first in-house consulting team — banks of that era placed such people inside companies and told management to take direction from them — and later one of the senior partners at Proudfoot. That inheritance is visible in the method: a consultant's supply-chain framing laid over financial and geopolitical analysis rather than the reverse.
• The framework originates from one physical fact, and he gives the numbers. The United States had 1,250 ships at Bretton Woods and has 256 today; Reagan campaigned on rebuilding to 600 and it never happened; America now builds roughly 1.4 ships a year against China's hundred. The framing question he set himself was not about America or China but about a prime minister of Austria or Finland: how does that person decide anything once the naval umbrella is gone? The answer became food, energy, technology, demographics, and ease of attack — and he notes the last one drew puzzled questions for three years and draws none now. Six global systems can break, with rare earth minerals and magnets added to the original four. Food worries him most: a century ago the average item on your table travelled about ten miles, and now it travels roughly 1,100.
• His message to restructuring professionals is that their entire toolkit has been invalidated, and he says it directly. Leverage is gone and free cash flow becomes king, but more fundamentally every way the profession has restructured for thirty years is no longer relevant, because valuations sit on top of a system that has broken underneath them. Hub-and-spoke is finished; the replacement he calls polycentric mutualistic architecture, a web in which every node must be individually hardened. His concrete advice: add at least three supply-chain experts and a technology-stack specialist to the team. His example of the decision failure to avoid is the aluminium case — 50% tariffs on Canadian metal sent American buyers to Qatar for a fifth of their orders, Iran then destroyed that plant with force majeure running five years, and the error was never the outcome but a procurement process that priced landed cost while ignoring whether the metal could arrive at all.
• The most original passage is the letter-of-credit cascade, which he treats as far larger than the commodity story. Oil quoted at $98 on paper traded at $140 in Dubai and $241 in Malaysia — the North owns the paper, the South owns the physical, and the anchor between them has snapped across gold, silver, copper and oil. The mechanism underneath: five trading houses issue letters of credit against cargoes, then repo those letters to five European banks and Citicorp at close to par, meaning trillions of dollars of seaborne commodities move on borrowed paper. When Hormuz froze, the houses pulled their letters within 24 to 48 hours, leaving roughly 3,100 ships whose cargo nobody owns — and, on his estimate, something like a $60B hole in bank balance sheets from repos now attached to nothing. Restarting the system requires someone to decide what price a new letter of credit is written at, which is unanswerable while three prices exist. His conclusion follows from who can still underwrite: China, whose state-owned banks finance oil, ends up the dominant financier of the oil trade.

2026-05-24 · open the original ↗

• Hallucinations usually signal missing context, not a faulty model. The guests' book argues that hallucinations are gaps in "canon", the information a system accepts as authoritative. The guest answering says that with 2026 models, a model told you live on the third floor won't say the second. A hallucination therefore flags that the context is missing or that retrieval failed to fetch it. The host adds an example of his own: an HR agent given an undated two-year-old guide whose policy has since reversed, which would answer confidently and wrongly.
• A "canon manager" should own what agents treat as true. The authors coined the title to consolidate roles that organizations name differently. The anti-pattern they describe is critical agent context buried in a GitHub repo that only an engineer can change, when the engineer is not the subject matter expert. The canon should be moved in front of the experts who are accountable for the agent's behaviour.
• AI tools that cut friction make company wikis viable. The guest says no one used the old company wiki, but a well-maintained one now unlocks "a million things". The biggest gain for their company and others has been AI that spots conflicts in the knowledge base and proposes a diff for an expert to accept or tweak.

2026-05-23 · open the original ↗

• Markets absorbed a sharp hawkish repricing — from two expected 2026 Fed cuts to a year-end hike — without the usual tantrum, buoyed by AI optimism and strong earnings. Fed minutes showed more inflation concern than April's press conference suggested, yet equities, credit spreads and yields all improved, reflecting a "higher-for-longer" recalibration of inflation, rates and deficits that investors largely shrugged off.
• The Main Street-Wall Street divide widened as record-low consumer confidence contrasted with solid hard data and a blowout Nvidia quarter. Nvidia reported $81.6 billion in fiscal Q1 revenue, up 85% year-over-year, alongside a 25-fold dividend increase and an $80 billion buyback, even as household inflation expectations rose and "soft" sentiment data diverged further from resilient production and jobless-claims figures.
• The week ahead centers on newly sworn-in Fed Chair Warsh's reform-oriented debut and the fragile Strait of Hormuz negotiations, with US PCE data as the key release. Consensus looks for headline PCE at 3.8% annually (0.5% monthly) and core at 3.3% (0.3% monthly), while Warsh secured supportive words from President Trump on Fed independence at his swearing-in.

2026-05-23 · open the original ↗

• China has moved upstream, which is the genuinely new development. On 20 May, days after the summit, a state decree on mineral resource law implementation ordered accelerated construction of strategic mineral reserve sites and stipulated that state-reserved strategic mineral resources may not be mined or encroached upon without approval from the State Council natural resources authority. China already dominated downstream processing at 91–98%; asserting government control over the mining itself closes the remaining end of the chain.
• He explains Sovereign Credit Substitution in plain terms, including what it is not. The government commits to buy output at $110 per kilogram where China sells at $30 — a differential that exists precisely because Chinese dominance suppressed prices far enough that no Western producer could earn a profit. That converts a producer's contracted cash flow from commodity and operational risk into sovereign credit risk: the only remaining exposures are the sovereign defaulting or Congress declining to keep buying. He is explicit that this is not nationalization, not a subsidy, not a price support programme and not an operational monopoly — and notes the company retains commodity-equity character above the floor.
• The Western bench is thin, and thinnest exactly where it matters most. MP handles light rare earths; the heavy elements — six of roughly sixteen to nineteen — are the real gap, and Lynas is the only Western route, mining in Australia and processing at the single non-Chinese heavy facility in Malaysia, already contracted 65% to Japan and 35% to the US. RE Alloys is a pending third candidate under a DoD framework with nothing allocated yet. Against a business measured in tens of billions of dollars, the Western players aggregate to a few hundred million. And the one processing route that exists is politically exposed — Malaysians are protesting in the streets against supplying heavy rare earths to the United States.

2026-05-23 · open the original ↗

• AI treats whatever data it is given as truth. So, Mumford argues, AI engineering falls back on data engineering. The book quotes Anchorman: whatever you put on the teleprompter, Ron Burgundy will read. Mumford likens this to data scientists who are hired, find the company's data in a mess, and spend two years doing data engineering. The hope is to pass that lesson on to AI engineering.
• AI engineering's data problem is unstructured documents, not missing numbers. Data science asked where the numbers were. Mumford says AI engineers in 2026 ask where the PDFs, images and Google Docs are, which are authoritative and which are junk, and how to chunk, embed and search them.

2026-05-23 · open the original ↗

• He argues quantum computing is replacing nuclear weapons as the defining strategic race. The US and China are the main competitors, with Europe trailing; he predicts quantum technology will reshape the world more profoundly than the internet did.
• The US and China are pursuing different quantum priorities, he says. China is focused on quantum communication for unbreakable secure messaging, while the US is pursuing brute-force computational power aimed at breaking encryption and advancing medicine, weaponry and space technology; his assessment is China leads in communication and the US leads overall.
• He warns adversaries may be stockpiling encrypted data to decrypt once quantum computing matures. Current quantum computers can't yet break modern encryption and may not for 10 to 20 years, but he raises the possibility that stolen, still-encrypted data is already being held in anticipation of that capability.

2026-05-22 · open the original ↗

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

2026-05-22 · open the original ↗

• SpaceX seeks a $1.5 trillion+ valuation despite losing billions with only one real business line. It lost $4.9 billion in 2025 and $4.3 billion in Q1 2026 alone (partly from the xAI acquisition), total 2025 revenue was under $19 billion, and Starlink (Connectivity) is the only dependably profitable segment.
• Nasdaq and S&P changed their rules specifically ahead of SpaceX's listing. S&P cut its "seasoning period" from 12 to 6 months and waived profitability requirements for megacap stocks, while Nasdaq shortened inclusion to just 7 trading days for top-40 market-cap companies, both changes accelerating forced buying by passive funds.
• Reincorporating in Texas in 2024 stripped shareholders of legal recourse Delaware would have provided. Derivative lawsuits now require owning at least 3% of the company (about $45 billion at the target valuation), and the S-1 routes securities disputes to the untested Texas Business Court instead of federal court, insulating Musk from accountability for self-dealing like SpaceX's $131 million in Cybertruck purchases from Tesla.
• The SpaceX S-1 explicitly frees Musk from any duty to avoid competing businesses, and the author argues this "Elon Musk IPO structure" — full board control, no firing risk, minimal shareholder recourse — is becoming a template other founders (citing OpenAI and Anthropic as likely next movers) will demand for their own offerings.

2026-05-22 · open the original ↗

• A record default rate is traced to a riskier form of collateral. The $2 trillion private credit industry's default rate hit 6%, tied to NAV lending, where banks lend to credit funds against the value of the funds' own loan portfolios rather than hard assets.
• A JPMorgan Chase markdown triggered margin calls at major private credit firms. In March, JPMorgan wrote down loans to software companies made obsolete by AI, forcing NAV-collateral holders Blue Owl, BlackRock, and Ares Management to raise cash or liquidate.
• Other major banks followed JPMorgan in revaluing similar loans, and the video sets up a part two on which banks are most exposed and whether the fallout could trigger a physical gold and silver supply shock.

2026-05-22 · open the original ↗

• Shareholder primacy is a recent legal invention, not a law anyone ever passed. Eric Ries, author of the forthcoming "The Uncorruptable," traces it to a small group of academics and judges in the 1960s-70s who reinterpreted "any lawful activity" charters as requiring shareholder-value maximization — a norm he says has never been subject to legislation or referendum, yet is treated as binding by courts, as when Twitter's board was compelled to sue to complete Elon Musk's takeover.
• Founders keep losing companies they built even while succeeding by every normal measure. Ries cites Twilio's Jeff Lawson, ousted 199 days after his dual-class control sunset expired despite revenue growth since IPO, and Sol Price, locked out of FedMart in 1975 by a new board after 20 years building it; FedMart went bankrupt within seven years of adopting standard shareholder-value practices, while Price's next venture became Costco.
• Independent board directors don't solve investor control — they just make it look balanced. Ries argues independent directors have no equity stake in a company's mission but do have a financial incentive to be seen as pro-investor, since investors recommend them for future board seats, making the classic two-VC/two-founder/one-independent board effectively investor-controlled; he cites data that companies using a separate nonprofit-trustee structure instead are six times more likely to survive to year fifty.
• A nonprofit trustee board can preserve value the market would have destroyed. Ries recounts Novo Nordisk's foundation trustees blocking a lucrative 2000s merger because it wasn't necessary for the company's survival — the rival company that merged instead had its R&D cancelled after being acquired by Merck — allowing Novo Nordisk to keep funding an unproven GLP-1 program for over a decade to a later $600 billion valuation; he points to Anthropic's similar long-term benefit trust as enabling its refusal of a $200 million contract.

2026-05-22 · open the original ↗

• Recent computer science grads now fare worse than the average new graduate. The host cites the New York Fed: 7% unemployment for computer science grads aged 22 to 27 and nearly 8% for computer engineering. That compares with about 6% for all recent grads and about 4% for the US workforce. He adds that US computer science enrollment fell 11% in 2025 and programming enrollment 26%, and that Handshake postings are down roughly 50% from their 2022 peak. Revelio Labs data suggest entry-level software and data analytics postings fell by as much as 67%. He contrasts this with White House adviser Kevin Hassett's claim of no sign in the data that AI is costing jobs.
• Some evidence says AI is shutting the young out of entry-level jobs. A 2025 study by Stanford's Erik Brynjolfsson and colleagues found employment among 22 to 25-year-olds in AI-exposed jobs down 13% since 2022, while older and less-exposed workers held steady or rose. The host says the Dallas Fed replicated the pattern this year and found it driven by non-hiring, not layoffs.
• Other studies find no AI effect, blaming rates and over-hiring instead. Two Google economists found posting declines in AI-exposed occupations as steep for seniors as juniors, and starting before ChatGPT. A Federal Reserve study of over a million firms found "precisely estimated null effects". Skeptics point to high interest rates, the end of the tech boom and 2021 over-hiring. Stanford's Eric Roberts likens the moment to the dot-com bust, after which hiring recovered by 2004.
• The host's view: macro explains much, but entry-level exclusion is real. He says the granular evidence that young AI-exposed workers are uniquely shut out is hard to dismiss. At his own consulting firm, Claude Code and OpenAI's Codex make code so cheap and fast that he is not sure why it would hire any entry-level developer in the foreseeable future.
• Grads should differentiate on design, domain, portfolio, agents and network. The host's five tips: compete on system design rather than raw coding output; pair AI fluency with a domain such as healthcare or law; build a public portfolio, since recruiters distrust CVs and applicant-tracking systems filter most out; learn agentic tooling such as RAG, evaluation and multi-agent orchestration, citing PwC's finding of a 56% wage premium for AI skills; and rely on referrals over job boards. He says the advice applies at any career stage.

2026-05-22 · open the original ↗

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

2026-05-22 · open the original ↗

• He predicts Cuba's regime is near collapse, with US military involvement likely. He frames the earlier Venezuela operation as a "warm-up," noting it cut off Cuba's main oil supplier and deepened an already severe Cuban energy shortage.
• He rejects the White House's "existential threat" justification as pretext. In his view, NATO's border with Russia and Taiwan, India and Japan's proximity to China pose far greater threats to those powers than Cuba poses to the United States.
• He infers three real motives: a political win, hemispheric dominance, and resources. He argues the administration needs a visible victory after the costly Iran war, wants to cement US control over the Western Hemisphere to a degree unseen since before the 1960s, and covets Cuba's fertile farmland plus its nickel and zinc reserves.

2026-05-22 · open the original ↗

• Project Nomad now supports AMD GPU acceleration for its AI features. Version 1.32.0 of Crosstalk's free, open-source offline server runs Ollama on AMD GPUs with ROCm support, where only Nvidia was supported before. The team tested it on a Radeon 780M iGPU and an RX 6800.
• Knowledge-base ingestion was redesigned and now defaults to manual. Users pick "always" or file-by-file ingestion in the setup wizard or on first AI chat; manual is the default because a CPU-only install can peg the CPU for days. Only text is indexed, so ZIMs heavy with images or PDFs show errors, and deleting an entry removes only the ingested data, not the file.
• Indexing a ZIM can take 5 to 10 times its size on disk. The host says ZIMs are compressed, so a 10 GB file could become 50–100 GB once indexed. Nomad now warns before ingesting anything over 50 GB or more than 10% of remaining free space.
• Maps and content sources were extended for international users. Maps now show coordinates and take named, coloured pins, with pin notes slipped to a 1.32.1 due within days. Maps can be downloaded by country at a chosen zoom (15 is street level; Nicaragua was about 125 MB), and the content explorer adds Kiwix mirrors and custom ZIM libraries.
• AI handling got more robust and is now fully optional. Nomad auto-recovers Ollama when Docker changes drop its Nvidia passthrough, and switching models mid-chat now unloads the old one so it doesn't spill out of VRAM into slow system RAM. Leaving AI unselected in the setup wizard skips every AI step.

2026-05-21 · open the original ↗

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

2026-05-21 · open the original ↗

• Sulfur is the cleanest commodity-layer cascade in the compendium because six distinct trigger paths converge on one rate limiter that cannot be independently scaled. Over 80% of global supply is recovered from desulfurization of crude oil and natural gas — natural gas processing ~57% (35.5 Mt), crude refining ~35% (22.0 Mt), other byproduct sources ~7%. Frasch mining is essentially extinct, uneconomic at any current price with infrastructure largely decommissioned. Sulfur supply is structurally locked to hydrocarbon throughput.
• It fails the elasticity test every other strategic input passes, and that is what makes the cascade unrelievable. Copper, rare earths, lithium, oil and gas can in principle all be scaled by capex if price signals are strong enough. Sulfur cannot, because its supply is a derivative of a primary product that is itself contracting under policy pressure — price signals do not produce more sulfur, they only redistribute the existing byproduct stream. This inverts every other chokepoint in the framework: Hormuz, Malacca and Bab el-Mandeb are bottlenecks where demand exceeds capacity, while sulfur is a chokepoint where supply is hostage to throughput in a system the world is deliberately shrinking. The constraint deepens as decarbonization succeeds.
• The price arc and the demand gap are already visible. Tampa contract sulfur ran $69/LT in Q1 2024 to $116 by Q4 2024, $270 in April 2025 and $310 in October 2025 — roughly 4.5x in eighteen months. Global sulfuric acid demand goes from ~246 Mt in 2021 toward ~400 Mt by 2040 against a projected shortfall of 100–320 Mt, equal to 40–130% of current supply depending on decarbonization speed. Phosphate fertilizer is ~60% of sulfuric acid demand, so the transmission runs straight into food. Producer concentration is meaningful too — Aramco ~10.9%, Sinopec ~10.2%, top five combined ~42.3% — with Middle East exports above 13 Mt, more than 60% of it to Asia.
• The most striking scorecard line is visibility, which reads "none." Sulfur is not on the US DOE critical minerals list, not on the EU Critical Raw Materials Act, and is not a Factor in any country's Five Factor read — yet it functions as the rate limiter for multiple Factors across multiple countries simultaneously. The entry logs this explicitly as a Factor Dimension Mismatch, and scores reversibility low on the grounds that byproduct architecture cannot be unwound on a sub-decade horizon regardless of what policy decides.

2026-05-21 · open the original ↗

• OpenAI is investing $2 million in tokens, not cash, into every spring and summer 2026 YC company. Garry Tan explains the money comes via an uncapped SAFE, effectively pricing the investment at the company's future Series A valuation but committing it at the earliest stage, and can be spent on Codex or the OpenAI API.
• The deal targets founders "token maxing" through software-factory patterns. Tan describes dispatching many agents to burn large volumes of tokens as a way to reach what would have been 18 months of engineering and Series-A-level product polish in a fraction of the time, and says YC observed its best current companies already spending heavily on tokens this early in their life cycle.
• YC extended a special application window since the summer batch deadline had already passed. Tan says the offer, developed together with Sam Altman, is being run as an experiment for just this spring and summer batch, with applications due by Memorial Day weekend (May 25) and decisions promised by June 5.

2026-05-21 · open the original ↗

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

2026-05-21 · open the original ↗

• The scorecard reads Australia as a middling state, and the entry opens by saying that is structurally wrong. A composite near 3.4/5 suppresses what actually defines the country: it produces near-monopoly volumes of five distinct critical inputs simultaneously — iron ore, metallurgical coal, LNG, uranium and rare earth upstream — while scoring 2/5 on Technology precisely because the downstream value-add is foreclosed. The composite averages away both the upstream concentration and the downstream foreclosure that together are the position.
• Australia is the cleanest case of geological abundance failing to convert into industrial capability. Iron ore from BHP, Rio and FMG runs roughly 800 Mt a year with about 80% to China, shipping at wellhead-equivalent margins. Lithium spodumene ships to China for conversion. Light rare earth concentrate from Mt Weld — among the highest-grade carbonatite globally — ships to LAMP in Malaysia for separation. Aluminium leaves as alumina rather than finished product. Met coal underpins Asian steelmaking and LNG underpins Asian power generation, in both cases at the bottom of the value chain.
• The stated cause is a cost-of-capital differential compounded by a skills constraint, operating over four decades. Long-term industrial investment faces high private-sector WACC in Australia against low and state-coordinated capital in China, which makes downstream Australian value-add chronically unprofitable at private cost of capital while remaining strategically essential. The second-order constraint is human: the mining engineering and geophysics workforce is world-class, while the semiconductor, advanced manufacturing and chemical engineering workforce is shallow. Export the upstream, import the downstream is the result of both operating together.
• What isolates Australia from comparable exporters is that the outcome was not chosen — and AUKUS is the substitute. Every other large critical-input exporter in the matrix either captures some downstream margin, as Norway does via Yara fertilizer and Saudi Arabia via Aramco refining, or has explicitly selected the upstream-only path as an alliance commitment. Australia is the only case where upstream-only is locked in by capital and skill constraints rather than strategy, producing five separate near-monopoly positions with virtually no industrial sovereignty. AUKUS is the substitution mechanism — buying alliance access to submarine technology and semiconductor capability in lieu of building it — and whether that substitution proves durable is the entry's central open question.

2026-05-21 · open the original ↗

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

2026-05-21 · open the original ↗

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

2026-05-21 · open the original ↗

• Zillow Offers lost to a winner's curse once its features drifted. The host gives the loss as almost a billion dollars. Miller describes the trap: an offer set too high is accepted and Zillow loses, while one too low is refused. The pilot went well at first, but COVID and other changes caused model drift. Miller treats any adversarial position as a shift in the feature space the model never planned for.
• Modern AI products face far more adversarial environments. Miller would hate to build a customer-service bot that hands out discounts to settle complaints, since Reddit posts teach people to trick it. Such projects need observability that flags out-of-norm behaviour immediately. At the start of COVID, Miller says, many data science teams threw out their models and forecast on the fly because the present no longer resembled the past.

2026-05-21 · open the original ↗

• Define and size the move before slicing anything. The presenter, a co-host, calls this "level zero": state the metric's definition, grain and window, say whether the change is absolute or relative and against what (prior four-week average, week-over-week, year-over-year), and set a bar for whether it is worth chasing. A 12% drop that follows an unexplained 12% rise is just a return to baseline.
• Drill into the dimension that explains the most change. The biggest percentage move is not the target. In the presenter's example, revenue is down 12% ($5 million) and APAC accounts for $4.5 million of it. That is the direction to follow, even if one acquisition channel is down 80%. Slice again inside that segment while it still carries most of the top-line move, from "EMEA" down to paid members in Germany or the UK on iOS. The failures named are panic, telling a story from the one dimension a dashboard has, and "drilling to nowhere", which Shane adds turns up spurious correlations and red herrings.
• Agents add depth a human would skip, not just speed. Shane recalls spending two months at Nextdoor on a root cause analysis of falling sessions, losing track of threads as new data came in. He argues that subagents can drill dimensions in parallel and write to one knowledge base, and that even naive segmentation and attribution done this way is "extremely powerful", with causal-inference methods as an optional layer on top.
• The demo traced a June 2024 ticket spike to one iOS release. On the repo's fictional Novamart data (50,000 users, 47,000 orders, 6.5 million events, 500 products, a $99 membership), a plain-English prompt had Claude Code find that payment tickets jumped to 41% of all tickets in June and fell back in July while attempts grew smoothly. It named a defective iOS app release, version 2.3.0, fixed a few days later, and rated its own confidence high. The presenter says the work still has to be checked. The Google Doc shown was produced beforehand, and the presenter says they already knew the story the data held.
• Opus is required for reasoning work, by the hosts' account. Shane says the system "does not work well with Sonnet", which is "very confidently wrong". It only clicked with Opus 4.6 in February, and Opus 4.7 went "off the rails" until the paid version of the repo was updated. A third co-founder says Sonnet can do pure execution of fixed skill rules. In Q&A the presenter says the agent flags every cause it finds when several drive a move, not only the largest. It connects to Snowflake and DuckDB natively, has no Copilot version, and a Codex version is being built.

2026-05-20 · open the original ↗

• He chronicles repeated Russian nuclear threats followed by the West doing the opposite. From Putin's February 2022 warning over Donbas through threats tied to Sweden and Finland joining NATO (April 2022), the Netherlands delivering F-16s (2024), and Ukraine's continued deep strikes after a December 2025 threat, he says each threat was met with escalation rather than retreat.
• Russia holds roughly 5,000 nuclear warheads yet has used none. He notes this despite Ukraine's daily strikes across Russian territory, including Moscow, and Ukraine's ongoing occupation of Russian territory in Kursk since an August 2024 incursion, which he says Russian air defenses can't adequately stop.
• Putin's latest threat, on May 18, 2026, named London as a potential nuclear target. He ties this to Putin accusing the UK and France of planning to supply Ukraine with nuclear arms, and concludes the recurring threats are hard to take seriously given the consistent lack of follow-through.

2026-05-20 · open the original ↗

• An undefined-risk strangle offers a materially higher probability of profit than the equivalent defined-risk iron condor. At the same short strike, the strangle skips buying protective wings, so it collects a bigger credit and carries a much higher probability of profit — the tradeoff, the host argues, is uncapped risk in exchange for getting paid more when the trade works.

2026-05-20 · open the original ↗

• Product taste is judgment, distinct from personal preference. Mark separates disliking a film from judging a product, and says the Sysinternals and ZoomIt design choices are Mark's own preferences, which suit others because they rest on intuitiveness and ease of use. Hanselman's example of missing taste is Windows voice access, whose bar takes over the top corners where users throw the mouse for close and the system menu; Hanselman says its designer made voice access the best rather than Windows.
• Their preceptor experiment showed most developers lacked it. A small group of Microsoft developers built ZoomIt's video trimmer without a visual spec, and both hosts judged most submissions clunky in both UX and expected behaviour; Mark's own version was the intuitive one.
• Taste comes from exposure and cannot be bootcamped. Hanselman calls it "having seen stuff and then filed it away"; Mark calls it experience, which "directed, methodical exposure" can accelerate only to some extent. Neither thinks a boot camp yields a senior engineer, and Mark says it takes sustained, repeated exposure to variants of the same thing.
• Software can't be fully specified, so engineers must still read code. Mark rejects a friend's claim that nobody will need to look at code, describing "sculpting" by iterating on AI output, as with the trimmer and a reference-checker app with dozens of behaviour questions; Mark notes that Carnegie Mellon's Software Engineering Institute, formed in the '80s, never codified programming. Hanselman adds that whatever is left unsaid gets filled by "the default part of the switch statement", and that AI does well on common apps but goes wacky at edge cases with little prior art.
• Mark still sees value in a degree. Mark calls it structured training, a gauntlet that serves as shorthand for evaluating someone, and says CS matters beyond simple web-plus-database apps. Hanselman worries that schools are not moving fast enough to teach thinking alongside AI, and Mark predicts "a lot of schools turning out ChatGPT degrees".

2026-05-20 · open the original ↗

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

2026-05-20 · open the original ↗

• The client was paying twice for poor 5G at two nearby offices. A Southern California catering company had a Verizon 5G home internet connection in each of two offices about six units apart, getting 40–60 Mbps with the modems buried among stainless-steel kitchen equipment. Charter Spectrum offered wired service billed separately per site at well over $150, and that only as a first-year promotional price.
• Crosstalk's design put one 5G router on the roof feeding both offices. It uses a Crosstalk Mobile SIM, the firm's own product, in a Peplink Max BR1 Pro 5G inside an IP67 Antenna Max enclosure magnet-mounted on the roof. It is powered by PoE++ over a single cable, clear of the kitchen, with outdoor Cat 6 run across the roof into the second office.
• A cheap TP-Link Omada switch and Wi-Fi 7 APs finished the build. The host found a $75 unmanaged DS106GPP with one PoE++ port for the router and three PoE+ ports for EAP772 access points, which adopt by QR code in the Omada app with no on-site controller. The team hadn't realised the APs came without PoE injectors, and bought a TRENDnet one locally for $30 for the second office.
• Both offices now share one LAN and one connection at mid-300 Mbps. The host puts that at about eight times their previous roughly 40 Mbps, with a single bill. He and a Crosstalk engineer add that Crosstalk Mobile SIMs have exceeded 700 Mbps elsewhere, in Pasadena.
• A neighbour's parking-lot link uses APs not built for point-to-point. The host spotted a pair of Ubiquiti U7 Pro Outdoor omnidirectional APs, presumably meshing across about 60–70 ft of clear line of sight. He thinks it probably works at that distance, but would not have chosen them, and worries about their RP-SMA connectors being left uncapped in the weather.

2026-05-20 · open the original ↗

• The verification corrects the news framing in both directions before any analysis begins. The 20 May Xinhua release is the State Council's implementation regulations for the revised Mineral Resources Law, signed by Premier Li Qiang — eight chapters, 79 articles, effective 15 June rather than immediately. The sequencing the headline omitted matters: the revised law passed 8 November 2024 and entered force 1 July 2025, introducing a new Chapter 5 that created the first legal framework for a national critical mineral stockpiling system. The legal authority already existed; this codifies the mechanics. The widely repeated "five-year minimum term" is marked Plausible rather than Verified, since it does not appear in the English release.
• The framework read is that this is control-plus-deny applied one layer further upstream. Ethiopia's GERD anchored the pattern on water through discretionary release; the May regulations create the legal architecture for China to deny extraction itself, not merely exports, at the processing-feed layer, gated on State Council approval. Both are defensive uses that convert the option to deny into a structural asset rather than monetizing it — a tight enough mechanism match that the entry recommends promoting the concept from candidate to formalized on the two-instance rule.
• The gap between summit language and domestic codification reaches a third instance and can now be formalized. The Phase 1 declaration on 14 May was the performative readout; six days later Beijing codified the domestic regulatory architecture that operationalizes the next coercion cycle. Western markets receive the summit headlines while the Chinese domestic policy stack quietly hardens — and the joint statement is silent on any of it.
• Two consequences carry forward, one structural and one about tempo. Structurally, supply becomes discretionary upstream of the export-control layer: the non-Chinese 20–30K ton floor underpinning the projected 40–65K ton NdFeB gap assumed Chinese supply stays commercially available, and Beijing now holds a legal lever to widen that gap on demand. On tempo, the US announced a $12B critical minerals reserve programme on 2 February and China announced copper reserve expansion on 3 February — Beijing now answers US strategic-minerals moves with operational counter-architecture inside 24–48 hours, and these regulations were likely accelerated to lock domestic architecture in before the next negotiating round. The entry also inverts one of its own prior diagnostics: the West has been pricing Chinese mineral dominance as transactional and is instead getting patronage-style state allocation — the mirror image of Iran assuming patronage and receiving transaction.

2026-05-20 · open the original ↗

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

2026-05-20 · open the original ↗

• The episode is a worked example of his data-point method, and he narrates it deliberately. His filter is that a headline must begin with "for the first time in N years" to be worth chasing. Honda reporting its first annual loss since 1957 cleared it; the model then corrected his framing and pointed at 1985 instead, arguing the Plaza Accord built modern Toyota and modern Japan as an exporter on four assumptions — a managed weak yen as competitive cushion, open access to the US consumer, reinvestment of FX translation gains into production systems, and hybrids as the bridge to electrification.
• Toyota's 8 May filing reads as a structural bet that the yen strengthens durably. Ten billion dollars of US capex, and Highlander and Tundra now exported from US plants back to Japan — which converts a dollar cost against dollar revenue into a dollar cost against yen revenue, so a stronger yen makes US-built units cheaper in yen-denominated accounts. The planning assumption is 150 against roughly 159 spot, embedding about 6% appreciation, and the shift to rolling averages signals the volatility regime is being treated as persistent rather than transient. Management states outright that the cost base has structurally shifted up and FX translation gains can no longer absorb it. Handing the presidency to the CFO rather than the founder's grandson signals that capital allocation and FX positioning now outrank product development.
• The consequence for the carry trade is the non-obvious part. If corporates reposition first, then when the financial unwind eventually arrives the mechanical corporate flows that would normally amplify a yen rally are no longer there to amplify it — short-run bullish for the yen, longer-term ambiguous, and probably a more muted move up than the consensus expects. He also logs a separate narrative confirmation: Morgan Stanley extending data decoupling across its investment bank with Goldman and JPMorgan expected to follow, which supports his earlier call that multinationals will bifurcate into two entities rather than choose between the US and China.

2026-05-20 · open the original ↗

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

2026-05-20 · open the original ↗

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