Souk's Notes - Unofficial Summaries

2026-10-23 · open the original ↗

• The merger negotiation is theater with a known ending. Elon Musk himself called his recusal from the newly formed Tesla and SpaceX special committees an "odd kabuki dance"; the author takes that literally, staging an imagined committee-to-committee dialogue in which both sides already answer to Musk and the only real question is how much more control he gets.
• SpaceX's size claim rests on shaky foundations. In the satirical exchange SpaceX cites a $2.6 trillion peak market cap against Tesla's $1.6 trillion, and Tesla counters that SpaceX, public for under six months, also fell to $1.4 trillion over that year and depends on government contracts and bond offerings, and may have no free cash flow. A Democratic House, the author suggests, would bring investigations, subpoenas and regulatory inquiries.
• Governance points to SpaceX acquiring Tesla. Musk holds 84% of SpaceX's voting power through dual-class stock plus extra board-appointment rights, so SpaceX as acquirer extends that control over the combined entity. The author imagines Tesla's committee demanding a 20% premium to compensate shareholders for Musk's greater control, then caving to the "why not give him everything" logic.
• A premium paid in low-vote stock costs Musk little. The author's closing note argues that even a won 20% premium would be paid to Tesla holders in SpaceX's low-vote shares while Musk keeps his super-voting shares, so holders get economics without power. The committees will bargain hard, but the absent Musk is the biggest beneficiary.
• Combining the pay packages compounds unfulfilled promises. As satire, the committees simply add both companies' contingent packages into $2 trillion, justified by a SpaceX milestone of colonizing Mars with 1 million people. The joke's point is that the promise itself, not delivery, is what gets valued and paid.

2026-10-23 · open the original ↗

• Analysts carry the risk when executives demand chat-with-data. The speaker, credited in the title as Tableau's Nate Nichols, describes executives who have a striking conversation with ChatGPT and tell the analyst to put it on all their data. Analysts feel caught in the middle, because when it returns a wrong number it is their job on the line, not ChatGPT's, and that brings pressure and anxiety.

2026-10-23 · open the original ↗

• LLMs silently resolve ambiguity, so answers drift between asks. The speaker, credited in the title as Nate Nichols, says LLMs are so eager to be useful that, asked "who's my best salesperson?", they pick some meaning of "best" and answer. Tomorrow's answer may differ because it used number of deals instead of total deal value, so the speaker's team works to shake out the ambiguity before answering.

2026-10-23 · open the original ↗

• Only the decision at the end of the data funnel makes money. The speaker, credited in the title as Tableau's Nate Nichols, says companies pay to store, process and visualise data, but only a better or faster decision produces money. The funnel is lossy, and the step from a dashboard to a better decision has historically had a low conversion rate.

2026-10-23 · open the original ↗

• Users wanted their key number fixed, not a dynamic feed. The speaker, credited in the title as Tableau's Nate Nichols, says Narrative Science's Lexio, a natural-language insights product, was first designed like a social news feed that put the most interesting changes at the top. Users instead wanted the number they need always in the same place. The speaker calls this partly psychological comfort and partly real for people making decisions, and the balance between dynamism and stability an ongoing trade-off.

2026-10-23 · open the original ↗

• Generated stories pay off for tiny, invested audiences. The speaker, credited in the title as Tableau's Nate Nichols, recalls a big client that turned little league stats entered by parents into game recaps ("came from behind off a three-run homer"). Each story had an audience of about a dozen people, but they were reading about their own kids and grandkids, and the speaker says that model resonated.

2026-10-22 · open the original ↗

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

2026-10-22 · open the original ↗

• The reserves behind the Venezuela deal are overstated by about 90%. The White House claimed over 65 billion barrels of proven reserves in the area the US will control (300 billion for the whole country); the AP reports researcher Luisa Palacios puts proven reserves at about 10% of that, roughly 6.5–7 billion, and Francisco Monaldi at 10, possibly 15 billion.
• The 300-billion-barrel figure was always a Chávez fiction, he argues. He says no US oil company ever backed it and the US Geological Survey only repeats what countries report; drawing on three decades in the industry, he stresses that only oil recoverable under existing economic conditions counts, and political control changes neither the geology nor the cost.
• A 90% smaller prize puts the promised investment case in doubt. The White House announced $100+ billion of private investment and future dividends, all resting on profitable barrels; he asks whether investors will pull commitments, and whether Washington will then be pressured into guarantees and subsidies to keep the deal alive.
• Venezuelan oil will not deliver the promised cheaper fuel. It is mostly extra-heavy sour crude needing blending and heavy infrastructure, and consultancy Welligence expects output to rise only a couple hundred thousand barrels a day through 2027; he adds that this high-cost oil needs the very high prices the deal promises to lower.
• The intervention's full cost demands an independent reserve audit. Brown University's Costs of War project estimates nearly $5 billion for US operations from August 2025 to March 2026, including the Maduro capture; he asks what the 35% stake and at-cost purchase rights are now worth, and predicts an independent assessment would come in below 60 billion barrels.

2026-10-22 · open the original ↗

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

2026-10-22 · open the original ↗

• Venezuela's claimed world-leading reserves were never backed by data. He has argued for years that the 300 billion barrel figure was a dictator's unverified claim, putting the true number near 60 billion barrels. An AP story published today, citing six Venezuelan oil experts now in the US, put the whole country at 60–100 billion barrels.
• The acreage the US seeks holds a fraction of Trump's figure. Trump said that acreage alone held 65 billion barrels; per the AP, Luisa Palacios put it at about 6.5 billion, and Francisco Monaldi at perhaps 10–15 billion depending on the recovery factor.
• Inflated reserves mean inflated valuations of Venezuelan assets. He argues reserves drive how a country's assets are valued, so US companies and private equity firms invested there hold overstated positions; if reserves are about 20% of the claim, he reads asset values as about 20% too. He attributes experts' past silence to their financial exposure to firms operating in Venezuela.

2026-10-22 · open the original ↗

• Engineering leverage no longer comes from how much code people write. Verma, CTO of Ironclad, says that because code writing is no longer the bottleneck, smaller teams can now match larger ones, with fewer specialist teams and less coordination overhead, though specialisms such as reliability and scalability remain; what has not changed, in Verma's view, is that the job was always mostly framing problems, weighing alternatives and judging solutions.
• Juniors are still hired, paired with seniors for judgment. Ironclad deliberately kept entry-level hiring, with the CEO's backing, on the reasoning that today's graduates are tomorrow's principals; seniors can struggle to unlearn but have the taste to judge AI output, as when a reviewer caught an agent-written SQL query that Verma says would probably have brought production down.
• Code review is shifting to risk-based tiers with a different reviewing model. Ironclad introduced AI across the whole development life cycle to avoid moving the bottleneck, but manual review is still overwhelming the team; the approach being explored puts human eyes on risky code and lets AI review standard code, always with a different model from the one that wrote it to avoid shared biases.
• Measure customer outcomes over time, not efficiency or lines of code. Verma rejects the efficiency narrative because its end state is zero, preferring growth framed as outcomes delivered per team per timeframe at acceptable quality; Verma also argues adoption into real workflows diffuses more slowly than the technology is produced, and that outcomes are not keeping pace.
• Shift mental models first, through voluntary learning. Asked what a new CTO should change first, Verma recommends education over a mandate even though returns come later; Ironclad ran a 20-day program last year, with Verma teaching classes, ahead of a hackathon, and about 10 months on reports good, largely voluntary adoption.

2026-10-22 · open the original ↗

• Exiting when IV rank drops below a threshold saved money. The host says they closed a gold strangle once gold's IV rank fell to 16, the rule being to take positions off under 16. Gold then dropped about $80 (roughly $8 in GLD), a move the host says is bad for strangle sellers.
• IV rank predicted which earnings strangles would make the most. The host sold strangles into IBM, Meta and Tesla earnings, sized by IV rank: most in Meta (about 65), fewer in IBM (49), fewest in Tesla (28). All three won, and the profits came in the same order. A December Tesla trade forecast at 50 cents made 46. The host's argument is that there is no edge in trading, only mechanics to optimize, and that the mechanics say trade bigger when IV rank is higher.

2026-10-22 · open the original ↗

• Moving to gradient boosting loses the scorecard's shared explanatory language. The author argues SHAP is a valid accounting of model output, but it rests on choices the result doesn't show (background data, how missing characteristics are filled in, log odds versus probability). Feature-level SHAP also splits linked fields such as balance, limit and utilization. So people who can challenge a WoE trend in five minutes struggle to challenge it, and the model looks more like a black box than it is. Their answer, in a new paper with Denis Burakov ("Conditional Weight of Evidence for Gradient-Boosted Credit Models", arXiv:2610.24586), is to extend WoE and IV to GBDTs rather than replace them.
• Conditional WoE measures a characteristic's effect given the rest of the file. It keeps the scorecard units and sign convention (bads on top, positive is adverse) but starts from the average for applicants who match on everything else. Worked example: a 10% to 20% default estimate is an odds factor of 2.25, so WoE = log(2.25) = 0.81. Ordinary WoE is the special case where nothing else is known. Omitted characteristics take the mix of values that similar applicants actually have, not zero or missing, and identity-linked fields (balance, limit, utilization) are treated as one group. In a simulation with a deliberately planted interaction, file age above 50% utilization had a true conditional WoE of −0.45 (files under four years) and +0.25 (established files). The method estimated −0.41 and +0.23, while ordinary WoE gave +0.08 and 0.00 and had the right sign only 28% of the time. The author stresses this shows the method can detect the interaction, not that young files are good credit.
• Conditional IV gives incremental information directly; coalition IV adds up. Conditional IV can't be negative and is zero only when a characteristic adds nothing, which does in one step what clustering, VIF and stepwise selection do after marginal IV. In the simulation, installment debt fell from a marginal IV of 0.68 to a conditional 0.01, while file age rose from 0.07 to 0.16 because its effect reverses with utilization. In the FICO HELOC data, trade volume fell from 0.24 to 0.04. Coalition IV divides the whole model's IV among characteristic groups: the full file's 2.51 was matched exactly by the coalition IVs, against a 4.56 sum of marginal IVs. That gives a model document a characteristic-level IV table whose rows add to the total.
• Two estimators are practical, each with stated weaknesses. The more accurate one fits an extra GBDT per characteristic group to predict the production model's probabilities without that group, and needs no new outcome data. In the simulation it had errors of 0.09–0.18 logit units and the right sign for 99–100% of material cases, but it is only as good as the production model and is untested on a badly misspecified one. Feeding the production model a missing value is not a valid shortcut. The second, closer to scorecard practice, computes bad/good shares within a neighborhood of applicants who often share leaves. It beat ordinary WoE on every characteristic but needs a lot of data, and was sometimes worse than assigning zero for weak characteristics, so the author says to check it against that baseline.
• Reason codes depend more on the reference group than on model error. The authors adapt the Regulation B comparison against applicants at or just above the passing score: they swap each characteristic group into reference files and average over swap order and references. The contributions sum exactly to the score gap and reproduce points-below-average on an additive scorecard. A simulated applicant 90 days past due with 53% utilization gets delinquency as the whole reason against broad passers, but revolving credit as the main reason against tree-similar passers, who all have a delinquency and use 27% on average. Switching reference groups changed the leading reason in about a third of simulated declines, while fitted and true models agreed 93% of the time, so the author argues reference choice should be documented as policy. Conditional WoE is reported separately from the reason code. Conceded limits: the accuracy results are simulated, the HELOC cutoff was imposed, booked-only samples bias characteristics the earlier approval policy screened on, and the paper doesn't settle which reference group to use.

2026-10-22 · open the original ↗

• Code-writing is no longer where engineering gets its leverage. Verma, CTO at Ironclad, says that shift makes smaller teams as capable as larger ones, while the bigger part of the job (framing problems, weighing alternatives, judging solutions) matters more than before. She disputed the "typing part of coding is over" line Hanselman relayed from a Microsoft engineer: voice dictation could always have done the typing, and what is new is that the tools produce the code. She sees engineers splitting into those who loved coding and miss it and those who preferred the problem-solving.
• Ironclad kept hiring juniors and pairs them with seniors. Verma's reasoning is that seniors have the judgment to evaluate AI output but can struggle to unlearn, while juniors are all-in on the tools but lack the pattern sense. Her example: an agentic coding tool wrote an unoptimized SQL query that she says would probably have brought production down, and an experienced reviewer caught it and wrote a detailed review for the engineer who generated it. She says the CEO backed the strategy and has posted about it on LinkedIn, and she tells staff to always challenge the AI rather than treat it as ground truth.
• Teams are shrinking as generalists replace some specialist teams. Verma expects specialisms such as fault tolerance, reliability and scalability to remain for some time, but says smaller teams cut coordination overhead and "done well will have an advantage". She distanced herself from Sam Altman's one-person companies ("I don't completely subscribe"), and says Ironclad's reorganisation took time because it was mainly a shift of mental models.
• Review is the new bottleneck, and Ironclad is still working on it. Ironclad introduced AI across every stage of the development cycle so that speeding up generation would not just pile work downstream, yet Verma concedes manual review still overwhelms the team at times. The approach under consideration is risk-based review: riskier code gets more human eyes, standard code is reviewed by another AI, and the reviewing agent must be a different model from the one that wrote the code so the two do not share biases. Hanselman says he runs multi-model adversarial reviews across Claude, GPT and Gemini; Verma says Ironclad does the same on the eval side.
• Adoption should start with education, not mandates. Verma says Ironclad ran a 20-day program last year, in which she taught, ahead of a hackathon, and about 10 months later adoption is good and has been voluntary. She measures outcomes delivered per team in a given time, with quality and customer value, not lines of code, and rejects efficiency framing because its end state is zero. She argues the pace of technology "will have to match the pace of diffusion", citing studies showing outcomes are not keeping up with the rate the technology is produced.

2026-10-22 · open the original ↗

• Russia's sharpest weapon is Americans who don't do their research. The speaker presents Russian disinformation as a continuous campaign since the end of World War II that exploits existing divisions. The examples are the civil rights era, the 1980s story that the US invented AIDS and released it on purpose, the 2016 election, and Black Lives Matter, which the speaker calls initially noble and grassroots but later hijacked to inflame racial tension. On 2016, the speaker says Russia "absolutely influenced the vote" but less directly than many on the left claim.
• The data center strike exposed how far Russian bot networks reached. The speaker says Ukrainian drones and missiles hit Russia's Yandex data center, and attributes these drops to it: Tucker Carlson's views down 71%, Douglas Macgregor's down 62%, Elon Musk down 1.5 million engagements, and less traffic around "genocide" and "free Palestine." Right-leaning creators lost reach and left-leaning ones did not, by this account. No source is named for the figures.
• The evidence that the bots were hit is anecdotal and disputed. The speaker acknowledges that some experts say the strike did nothing to the bot accounts. Against them the speaker offers a personal observation, shared by other creators, that comment sections on Twitter and TikTok have become cleaner and less vitriolic since the facility went down.
• The speaker implies more data center strikes would clean up social media. This is framed through a teenage friendship with a Russian friend that the speaker only recognised as toxic after a month apart. The clip ends on a rhetorical call to action rather than a stated proposal.

2026-10-22 · open the original ↗

• The US pullback is about capacity, not willingness. He dates his view to the third day of the Ukraine war, when he concluded the Russian empire was over, and to a February 2025 War on the Rocks piece by Navy/Air Force strategist Jonathan Panter arguing the fleet was overstretched and had to pull back to repair ships and personnel. The Five Factors, he says, trigger once no single power polices the seas.
• Barrack's speech confirms the retreat is official US thinking. Tom Barrack, US ambassador to Turkey and a 40-year Trump associate, told the Global Security Forum that "America as guarantor to the world for the last 80 years is no longer there," urging regional defense partnerships: realignment, not abandonment. The author stresses Barrack never offered protection for payment, unlike Trump's earlier "pay us one trillion dollars" line — so, he infers, money isn't the constraint.
• Shipbuilding numbers mean the retreat is permanent. He cites 1,250 US ships at the end of World War II, Reagan's 600-ship goal, a low of 277 under Obama, and about 300 today, against China producing "over 300 ships a year." His forecast: no election outcome reverses this, and US leadership is not seriously debating how to plan or fund the adjustment.
• Undersea infrastructure attacks will go unchecked as the US recedes. He points to a new RAND report, "Winning the Undersea Knife Fight," on underwater critical infrastructure, which flags three problems: damage may only surface weeks or months later, attribution is unclear, and the governing law is poorly defined. His inference is that other players will now get away with, and fight over, what the US once prevented.
• Secure and control replace ROE and ROI as decision criteria. As a data point he cites Micron investing in a critical-minerals mining and refining company — a move he says would have gotten a CEO fired a year or two ago and now draws no notice. He argues no company, country or region has worked out future supply chains, but all will have to control their own.

2026-10-22 · open the original ↗

• Nobody can time the market, but traders can scale exposure. The show cites a 20-year S&P 500 study: about 11% a year fully invested against about 6% for someone who missed only the 10 best days, seven of which fell within two weeks of the 10 worst. The hosts call dodging the crashes impossible. Their alternative is cutting delta instead of going all in or all out, for example by selling covered calls in an IRA when the market looks frothy.
• Size the book by a daily theta target and a delta band. Tom Sosnoff aims for daily theta near 0.2% of the account, about $200 a day on $100,000, ranging from roughly $100 to $300 depending on volatility. Delta runs about 150–300 SPY beta-weighted deltas per $100,000, usually 100–200, or about 1:1 to 2:1 notional. He says $1,000 a day of theta on $100,000 means too much gamma risk. When a side is tested, both hosts roll the untested side. Tom acts on discomfort, while Tony Battista acts mechanically at a 15–20 delta skew.
• A possible merger of two sister companies offers no trade. Tesla and SpaceX have formed special committees, and Elon Musk called the deal a good idea but uncertain. The hosts split on whether it happens, but agree there is no arb or pairs trade before or after terms. They expect regulatory and shareholder review to take at least as long as the roughly nine-month Paramount–Warner Bros. Discovery deal. One host would buy SpaceX, not Tesla, but only as a contrarian move.
• Small mortgage-rate moves shouldn't decide whether to buy a home. The figures cited: a quarter-point rise from 7.4% to 7.65% costs about $19,440 over 30 years on a $400,000 home with 20% down. Neither host has ever let a rate change stop a purchase. They also reject comparing loan interest with money in a passive S&P fund. From there one host repeats an argument that young people should not open retirement accounts and should bet on themselves. A caller says the same applies at 53 or 70.
• Buyback announcements are not a tradeable signal. S&P 500 companies are estimated to buy back a record $1.55 trillion of stock this year. One host calls the announcements close to fake, since companies don't have to follow through, and would fade a rally on one. Both say retail traders have no trade there, bullish or bearish.

2026-10-22 · open the original ↗

• The world's tool fleet caps leading-edge output at roughly double. Asked whether enough tools exist to 10x Intel's leading-edge wafers, CTO Pushkar Ranade said "not 10X, but probably 2X", against AI compute that has grown 3x a year for a decade. ASML makes fewer than 100 EUV tools a year, and a locked process flow ties each step to one tool from one supplier. With roughly 18-month tool lead times plus a new fab shell, an order placed in mid-2026 yields capacity near the end of the decade.
• Intel now caps how much innovation goes into each node. Process lead Chris Auth says 10nm, targeted for 2016 but in manufacturing only in 2018, tried "hyperscaling" without new tools because EUV was not ready. Phase-gate maturity metrics on yield, defects and performance now decide what gets in, and anything that misses waits for the next node. The host notes TSMC had about four more years of EUV experience.
• Backside power is what makes High-NA EUV pay off. For Intel, moving power lines under the transistors relaxes front-side pitch enough to print in a single High-NA pass, Auth says, while TSMC without backside power would still need multiple passes. High-NA (0.55 NA versus 0.33) is optional in the 14A PDK, decided customer by customer. Auth also concedes 18A started with no formal PDK.
• The gap to TSMC is scale, not technology. Ranade argues this and calls a few percent on performance, power or density "splitting hairs" next to on-time, high-yield ramps, and disputed the host's SemiAnalysis-based claim that 14A (2028) matches N2 on density. He conceded Intel's ~$20B capex trails TSMC's ~$50B, that selling the flash business was badly timed, and that Intel should have invested more a year ago.
• The long-run direction is 3D, limited by heat and interconnect. Auth expects "pretty exotic" stacked logic structures by 2045, but thermals remain a limiter. Stacked CFET transistors exist in simplified one-on-one lab form, with high-aspect-ratio etch and fill the obstacle. Ruthenium beats copper below roughly 20nm pitch, and hybrid bonding is at 25µm pitch in production, sub-10µm in Intel's lab and 0.45µm in published research.

2026-10-22 · open the original ↗

• Expect peace announcements timed to the election, not to events. He predicts that over the roughly 12 days before the midterms, "miraculous" breakthroughs will be announced, possibly including an end to Russia's war in Ukraine, which he calls "electioneering dressed up as diplomacy". He argues that an announced negotiation or temporary pause is not a war ended.
• A campaign deadline hands adversaries leverage to raise the price. He cites a Reuters report that senior Iranian officials say they deliberately prolonged negotiations to weaken Donald Trump before the midterms. His reading is that a counterpart who sees a president needing a deal by a fixed date can stall, raise demands, and offer something that looks good on television while the hardest questions stay unresolved.
• Watch whether concessions come before verified commitments. He asserts the administration has already granted Russia sanctions relief and lifted restrictions before commitments were verified, and asks whether Iran is next. He calls Ukraine an ally and Russia not, and says a real agreement must state what each side does, when, and what happens if it is broken.
• Trump's response to losing in 2020 is why scrutiny is warranted. He recounts the false stolen-election claim, the pressure on Georgia's Republican Secretary of State to find about 11,780 votes, and the pressure on the vice president to reject certified electoral votes. He infers that the urge to avoid another humiliation could shape the bargains offered to Russia and likely Iran, a weakness foreign governments can exploit simply by raising the price.
• Cheaper-fuel promises should be judged by physical conditions, not headlines. He expects deal announcements to come with promises of fuel relief that may briefly calm markets. His test is whether ships can move safely, whether infrastructure operates reliably, and whether attacks have actually stopped, since an election cannot make a war's physical consequences disappear on schedule.

2026-10-22 · open the original ↗

• Build what you already know, and wait for ten paying users. Alex, Magic Patterns' CEO, says the founders entered YC with a text-message analyzer that had revenue; their partner told them to work on what they knew, which for two front-end engineers was turning Figma mockups into code. He says it took about nine months of pivots, none of which found ten users who "love" them (defined as paying and using), before Magic Patterns found them quickly at its October 2023 launch.
• Staying ahead means adopting whatever workflow customers move to. He calls customers "our religion" and gives the example of an MCP server built about a year ago as an internal hackathon project that has since become one of the main ways people use the product. He positions the tool as model-agnostic and exporting to Figma or any coding agent, contrasting it with a competitor he says is tied to one lab's models, and advises spending ten minutes loading skills and design systems first.
• A landing page is judged by conversion, so lead with the action. Alex says Magic Patterns removed the prompt box from its own homepage to be more creative and conversion "totally tanked". Host Aaron says tests at his company Creative Market showed an expanded email field converted much better than one behind a click; both criticised a site whose waitlist call to action sat halfway down the page.
• Agent defaults leave a recognisable "vibe-coded smell" that distracts visitors. They list over-animation, scroll-jacking, glowing gradient buttons, "scroll" prompts, inputs that look typeable but are not, randomly bolded words, small all-caps text, and default fonts such as Instrument Serif and Space Grotesk. Aaron expects these to date like an '80s kitchen and urges resisting defaults; animation is worth keeping only where it reinforces the product's point, as one reviewed site's closing-laptop sequence did.
• Write the site for users, not investors, and earn trust openly. Aaron says vibe-coded sites, including many in YC applications, read like pitch decks about target segments rather than saying what the product does and for whom. Both expect AI products needing sensitive data to put privacy and security up front, and Alex warns that social proof sets the audience: two investor testimonials made an email tool look like it was for VCs.

2026-10-22 · open the original ↗

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

2026-10-22 · open the original ↗

• Durable agents persist state so long workflows survive a restart. Collier, a principal architect at a consulting firm, says in-memory agent state is lost if the hosting compute goes down, and LLM outputs are non-deterministic, so they should be saved rather than rerun. A durable agent stores conversation history, tool interactions and workflow position externally and picks up where it left off.
• The Durable Task extension replaces hand-built checkpointing in Agent Framework. With Microsoft Agent Framework alone, Collier says, the developer must define checkpoints, save them (Cosmos DB, a file system) and detect errors to rehydrate from them. The durable task extension instead hands this to the managed Durable Task Scheduler service with an opt-in call (he recalls something like "as durable agent"), running agents as durable entities.
• Durable Functions orchestrations can mix deterministic steps with agents. For businesses with fixed sequences, Collier lines up activity functions and AI agents in one orchestration triggered by a Service Bus message, Event Hubs event, HTTP or timer. This keeps determinism over the order of calls while each agent solves its part non-deterministically.
• It deploys like any Azure Function. Collier's steps are to provision the Durable Task Scheduler, put its connection string in the function, install the SDK and deploy through CI/CD; Microsoft Foundry hosted agents can host the code instead, still backed by the scheduler. The scheduler runs free locally as a Docker image, and in Azure comes as a cheaper consumption tier or a more expensive tier with more scale.
• Scale-to-zero lets workflows wait on humans without rerunning agents. If an approval sits for a week while someone is on vacation, compute shuts down and rehydrates from the state store when they respond. Collier stresses that an earlier agent is not rerun, avoiding both the cost and a potentially different result.

2026-10-22 · open the original ↗

• He takes the simulation hypothesis seriously but gives no probability. Nick Bostrom (the same passage appears in his full interview of 2026-10-16) says he has "tended to punt" on giving a number and will keep doing so. He holds that the simulation argument is correct, and that it presents the simulation hypothesis as one of three options a person can coherently believe about the future and our place in the world.

2026-10-22 · open the original ↗

• Musk shaped the book only indirectly, through conversation. Yampolskiy asks about Elon Musk's tweet claiming he contributed to the book. Nick Bostrom says he talked with a great many people while writing Superintelligence, which helped shape his thinking and presentation, and can't recall who said what. He is clear that Musk was not "editing up different chapters" or writing any of the text.

2026-10-22 · open the original ↗

• Indefinite control is possible in principle, perhaps not in time. Yampolskiy asks whether a superintelligence seeded at the outset with the right final goals, friendly to humanity, could be controlled indefinitely. Nick Bostrom (the same exchange appears in his full interview of 2026-10-16) answers "possible in principle, yes". Whether it is feasible in practice "in the time scale available" he calls an open question.

2026-10-22 · open the original ↗

• Calling the current model an AGI is less misleading than denying it. Nick Bostrom (the same exchange appears in his full interview of 2026-10-16) calls the question of whether GPT-6 is AGI "definitional" now. Shown 15 years ago, a system that can write an award-winning novel and "maybe" solve a Millennium Prize problem would have looked general. Yampolskiy counters that such feats sound more like superintelligence.
• Superintelligence has not yet been reached. As of the 2026-10-22 post, Bostrom defines it as a system that radically outperforms the best humans across all practically relevant fields. Current systems still lack continual learning, research taste and physical manipulation skills, which he says is why humans are still needed across the economy and in the AI labs.

2026-10-21 · open the original ↗

• Hex won't fix a data team unwilling to partner with the business. The speaker, a Hex executive (credited in the title as its CEO), says they have recently told some prospects not to buy Hex. Business demand for "chat with my data" leads teams to look for a tool to throw at users, a set-up the speaker says is often not built to succeed. The speaker argues against a standalone chatbot for the business alongside a separate tool for the data team, in favour of one system that brings both to the table, and invites teams that want that partnership to get a demo.

2026-10-21 · open the original ↗

• Data people are defined by problem-solving now, not syntax. The speaker, credited in the title as Hex's Barry McCardel, says mastery of SQL or pandas syntax no longer marks a data person, and that Hex has cut back the coding tests it used to give in data interviews. What transcends the technical work "that all engineers can do really well now" is a love of puzzles and mysteries and a way of partnering with others to solve them.

2026-10-21 · open the original ↗

• Russia is targeting pro-Ukraine candidates of both parties. Politico reports senior lawmakers warning of a covert Russian influence campaign aimed at American candidates who support Ukraine, Republicans and Democrats alike. He reads the aim as making support for Ukraine politically expensive and smearing its backers until defending the invaded country is more controversial than the invasion, so every fracture in US support leaves Putin less opposition.
• Fabricated evidence crowds out the legitimate debate over foreign aid. Reuters reports Senator Mark Warner pointing to fake AI-generated advertisements targeting Democratic candidates. He grants that questioning foreign aid and demanding accountability for American money is fair, but argues those arguments need facts, and a foreign government manufacturing evidence takes that honest debate away from voters.
• "No votes were changed" misses how influence campaigns work. He notes the lawmakers presented no evidence of Russia altering votes or compromising voting systems, but argues Russia never needed to: sustained misinformation breeds animosity and changes votes psychologically. In his view the US has been losing a social-media propaganda war with Russia since the internet went mainstream, and Russia drives much of today's partisan hatred.
• Ordinary users who repost unchecked content do Putin's distribution for him. He urges people to find the original source, check whether the person actually said it, and apply the same scrutiny to stories that confirm their views, holding themselves to the standard journalists should meet. He argues many Americans' opinions were shaped by Russian propaganda rather than facts, and sharing it carries personal responsibility.

2026-10-21 · open the original ↗

• Russian diesel is flowing globally, and America isn't first. As of October 21, 2026, the first cargo, 40,000 tons loaded last week at Primorsk, is heading to Morocco per shipping data; Deputy Prime Minister Novak said the first US-bound ship, also 40,000 tons, leaves Thursday. President Trump announced the deal on October 9 as diesel "for America and for the global market," with the license running through April 2027, and Russia partly lifted its export ban on October 10 — diesel moves to all buyers while gasoline and jet fuel stay banned.
• One Russian cargo covers only hours of US demand. The host's math: 40,000 tons is about 300,000 barrels against US diesel use of about 3.8 million barrels a day, so one ship is roughly two hours of supply — "a start, but it's not a fix."
• Ukraine keeps striking the refineries meant to supply the diesel. Overnight reports said strikes hit the Ilsky refinery in southern Russia, an oil terminal in Taganrog and a fuel depot near Azov; the host calls this the hard part of the deal, since Russia is promising diesel from plants that keep getting hit.
• US diesel stocks rose, but too early to credit Russia. The EIA reported on October 21 that diesel stocks rose 1.7 million barrels to 104.3 million, the first rise in weeks, still about 9% below the 5-year average and containing no Russian diesel; the host calls one week "a first read, not a verdict," with pump diesel at $6.28 a gallon per the EIA (the captions render it "$628") and crude around $110 a barrel.
• A reported US-Iran peace plan remains unconfirmed. Reports on October 21 said both sides gave a first yes to a plan from Qatar and Pakistan: secret inspections of Iran's nuclear sites in return for Iran receiving the interest on $6 billion in frozen assets, the principal staying frozen. Oil fell about $1, but no official has confirmed it; the host's bottom line is "a report is not a deal" — watch the ships, not the talk.

2026-10-21 · open the original ↗

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

2026-10-21 · open the original ↗

• Evals shouldn't be a one-off backtest you never revisit. The speaker, from PromptLayer per the title, frames evals as answering whether the AI product is good and whether the next change will make it better. They describe a common pattern: a chatbot replies in the wrong language based on users' names, the team backtests the last thousand conversations, iterates on the prompt until it is fixed, ships, and never touches those evals again. Their advice is to step away and work out with pen and paper how a human would grade the output.

2026-10-21 · open the original ↗

• LLM judges can mis-score, so route low-confidence evaluations to experts. The speaker, credited in the title as Composo's Dr. Seb Fox, describes an example where a change the judge itself had suggested, adding a name, made its score fall from 90 to 80 out of 100. The speaker says this isn't universal but appears across a surprising range of LLM-as-judge evaluations, so each evaluation should carry a confidence signal, and the low-confidence ones should go to human domain experts. The clip is fragmentary.

2026-10-21 · open the original ↗

• Always keep at least one eval failing. The speaker, credited in the title as Braintrust's CEO Ankur Goyal, says a fully passing suite means there is more work to do. A user complaint should be capturable as an eval within one or two clicks. With a change run on 1,000 or 10,000 examples and only 30 minutes to review, the problem becomes surfacing the examples most likely to show positive or negative changes.

2026-10-21 · open the original ↗

• One host argues taking gains is always correct, even early. Responding to a cited study of 10,000 brokerage accounts in which investors realized about 15% of paper gains but only 10% of paper losses, this host says you should take far more gains than losses because you need to be right more often than wrong, and has no issue with cutting gains early and letting losers run.
• The other host says that view blends trading with investing. This host frames the question as being about a stock portfolio, not someone selling an iron condor or a naked put. The first host concedes finding the two hard to distinguish; the reply is that for most people the difference is large.
• For investors, the second host says it depends on environment and horizon. With markets at all-time highs, this host sees no problem taking something off the table, crediting Warren Buffett for selling into strength and, above all, for longevity. Someone in their early 20s building a 40-year portfolio, the host argues, should let positions run.

2026-10-21 · open the original ↗

• Witkoff's flattery of Putin is an embarrassment, not diplomacy. In a played clip, envoy Steve Witkoff calls Putin "a humanitarian" after Russia released an elderly American prisoner (the post body says 74 and seven years held; the clip says "a 75-year-old man"). Aaron Parnas sets this against a Russian strike on a Kharkiv park two days earlier that killed a one-year-old and an 11-year-old, and argues that if reporting that Elon Musk throttled Starlink in Ukraine is accurate, the resulting civilian deaths expose him to war-crimes liability — hence the finger-pointing between Musk and Rubio.
• The midterm is a referendum on Trump, and Democrats are winning it. The New York Times/Siena poll shows Democrats +11 on the national House vote, ahead on every major issue including immigration, and seen as the party of the working class by 13 points; Trump's job ratings are 73% negative on gas prices, 69% on cost of living, 64% on the economy, 61% on tariffs and 59% overall (67% among independents). Mike Bloomberg is putting $80 million into the midterms, at least $50 million for House races, while the RNC is moving more than $50 million to its committees — all snapshots two weeks before election day.
• A fraudulent poll helped push McMorrow out, so polling needs policing. The Wedgewood poll, about a year old, put Mallory McMorrow in single digits in the Michigan Senate primary and was ingested by major aggregators before being exposed; Jessica Tarlov says a college student and an older man with a Kalshi betting account were behind it. Parnas calls it likely wire fraud and asks whether the polling industry should be regulated; Tarlov wants stricter criteria for aggregator inclusion.
• Ramaswamy's ad with Acton's estranged mother will backfire. Ramaswamy owned the ad and, per Parnas, is putting $4 million behind it; Tarlov notes the mother lost custody when Acton was 12 and that the story of neglect and abuse has been confirmed by conservative outlets. Both hosts now call Acton the clear favorite for Ohio governor, with Parnas's caveat that voters seeing the ad without context could still be swayed.
• Sununu's distancing from Trump is his best play, but Pappas wins. Sununu's ad says he opposes the war in Iran and the trade war with Canada and attacks Pappas for voting with Nancy Pelosi 99% of the time. Tarlov rates the New Hampshire Senate race lean blue; Parnas predicts Pappas wins but closer than expected.

2026-10-21 · open the original ↗

• Commercial AI is normalising the loss of critical thinking. One speaker argues that unchecked capitalism, selling inference under the pretense of making people more capable, is moving the Overton window so that people without critical-thinking skills absorb this as normal and will not notice what they gave up, as with social media "raising the water temperature". The speaker cites the "meat proxy" (someone who just relays questions to an AI and passes back the answer), says those jobs will go away and calls that bad, then asks what being a good human looks like alongside AI. The clip is a passage that also appears in the "Scott & Mark Learn To...Scroll TikTok" episode.

2026-10-21 · open the original ↗

• Long-form reading is declining, and the hosts expect that to continue. The episode reacts to a TikTok by Sinead Bovell warning of a "cognitive divide" between people who keep "cognitive friction" in their lives (long books, wrestling with arguments) and those who let feeds and AI make judgments for them. Russinovich agrees the US is likely headed that way, says the share of Americans reading books is falling and the decline is accelerating, and says his own book reading has dropped massively over the last 3 years because he spends the time keeping current on AI. Hanselman counters that enthusiasts are reading more, through book clubs and book TikTok.
• Losing critical thinking makes people manipulable and invites authoritarianism. Russinovich spends much of the episode asking why the clips treat this as bad, saying he wants an answer more rigorous than "that's what humanity is". His own answer: a fair, functioning society needs an educated population, which needs critical thinking, which needs reading, and where only some read, they can manipulate those who do not. Hanselman's framing is that struggle is what makes us human, using his analogy of sending an agent to lift weights for you.
• Hanselman blames the business of selling inference for normalising it. He argues that unchecked capitalism, selling inference under the pretense of making people more capable, is moving the Overton window so people will not notice what they have given up, as social media did. He says "meat proxy" jobs, where someone just relays a question to an AI and passes back the answer, will go away, and calls that bad.
• AI pulls output toward the average unless pushed. Reacting to a clip by Sophie, a software engineer at Oracle, warning that misusing AI makes you "normal", Hanselman calls AI the statistical mean ("it's mid") and says you have to push it outside the bell curve. Russinovich again objects that the clip presumes being average is bad without saying why.
• Restrictions on AI in schools signal a product-design failure. A clip from Ashley, a former Google researcher, argues New York City's move to ban and slow AI in classrooms shows products are not designed for how students actually use them. Russinovich adds that the human brain is inherently lazy, so few children would pick tutoring over entertainment, much as few would go to school if it were optional, and that people trade away their data because they "get to scroll for free". Hanselman worries an AI-dependent public would vote against its own interests.

2026-10-21 · open the original ↗

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

2026-10-21 · open the original ↗

• The hosts split on long-horizon stock buying, not on taking profits. Prompted by a study of 10,000 brokerage accounts in which investors realized about 15% of paper gains but only 10% of paper losses, one host calls retirement-account contributions among his biggest regrets and buying stocks to hold for 40 years the worst advice possible. The other defends small automatic saving and says the hosts' business-founding path is highly unusual.
• Chips are a trading opportunity, not a reliable dip to buy. The semiconductor index had its longest losing streak since 2008, erasing about $2 trillion while still up nearly 60% for the year. One host calls it a contrarian trade; the other warns that buyers at down 10–20% now sit on 50% losses, and both dismiss a JP Morgan SpaceX target raised to $500 as unactionable.
• Tesla's cheap implied move should be sold, never bought. Options priced about a 6% earnings move against a roughly 9% ten-year average; the hosts say the options market is right and they would never buy a straddle, preferring a strangle near twice the expected move. They concede realized moves beat implied in 11 of 16 reports, 69%.
• Gold is a range-bound trading vehicle, not a crisis hedge. With gold near $4,100, down from about $5,781, a 10-year yield near 5% and real yields around 2.6%, the hosts expect a range and prefer /GC futures options over GLD's wide markets. A caller pushes back that physical metal is convertible when paper fails.
• The wheel works in principle but wastes buying power in practice. One host prefers rolling an in-the-money short put to taking assignment, which adds overnight directional risk; the other accepts the wheel on beaten-up, high-IVR stocks but not at all-time highs. Their example: a 47 put on a $50 stock for $0.54 earns about 1% on $4,700 a month.

2026-10-21 · open the original ↗

• No accounting supports Trump's claimed $350 billion to Ukraine. He says the Trump administration's own new aid was a single military appropriation of $400 million; the rest was delivering equipment Congress committed before Trump returned, plus weapons other countries bought under the NATO purchasing program. He puts total US aid since the war began at roughly $150 billion, mostly weapons, and "not even $200 billion."
• Trump's other Ukraine claims misstate the record. He cites Trump telling Zelenskyy in February 2025 that Ukraine started the war, when Russia invaded. Against Trump's claimed 4% approval he puts Zelenskyy at 60–65%, averaging about 63%, and notes University of Manchester researchers challenged the figure. On "dictator without elections," Zelenskyy won an observed election and Ukrainian law, predating him, bars wartime elections under martial law.
• The "nothing in return" complaint ignores Trump's own minerals deal. His administration signed a minerals and reconstruction investment agreement with Ukraine, which the White House promoted as a win giving preferential resource purchases and long-term returns. He concedes the benefits are prospective, but says the arrangement Trump complains is missing is the one he negotiated.
• He reads Trump's aim as forcing Zelenskyy to surrender territory. In his view Trump wants Americans to resent Ukraine so blame shifts off Putin, targeting Zelenskyy as the weaker party he can control, in order to claim he ended the war. He argues debate over aid levels and oversight is legitimate, but cannot happen honestly on inflated figures.
• Trump, not Zelenskyy's refinery strikes, drove diesel to current levels. He argues the Oval Office confrontation and the halt in US aid left Washington no leverage to ask Kyiv to stop hitting refineries. Those strikes were already under way without lifting prices; what tipped a vulnerable market was the US entering the Iran war and, above all, the later escalation into attacks on Middle East energy infrastructure, which he says he warned about five months ago.

2026-10-21 · open the original ↗

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

2026-10-21 · open the original ↗

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

2026-10-21 · open the original ↗

• The administration's Hormuz throughput figure hides a large shortfall. A clip of an intelligence official states about 1.3 billion barrels have passed the strait since the blockade began; he accepts the figure but notes pre-war flow would have been about 4.7 billion. Adding roughly 800–900 million barrels bypassed by pipeline, mainly to the Red Sea, he puts exports at under 50% of pre-war.
• Depleted inventories keep fuel prices high even after flows resume. He cites today's API report of US draws of 5 million barrels of crude, 2.3 million of diesel and 2 million of gasoline, and argues stocks must be at least partly rebuilt. On that basis he calls the administration's claim that prices will fall "like a rock" after the war an impossibility.
• Refilling the global deficit takes years, so relief is only gradual. He estimates global crude stocks will end about 2.5 billion barrels below pre-war; at the pre-war surplus of just over 2 million barrels a day, he puts the refill at 2 years and nearly 8 months once supply and refining are restored. He expects prices to ease in proportion as stocks rise, no return to pre-war levels this year, and a longer recovery the longer the conflict lasts.

2026-10-20 · open the original ↗

• A coordinated misinformation campaign knocked oil down this morning. He reports crude fell by over $4 a barrel very quickly after a series of social media posts spreading false claims about oil markets; he reads the posts as what appears to be a coordinated effort. He argues the volume of market information now flowing through social media opens the door to more of this.
• Bot-driven repetition passes for verification, and trading algorithms act on it. A network of automated accounts can make one unsupported claim, such as a terminal attack, a route closure or a peace deal, look like many sources confirming it. Because data providers turn social content into trading signals and AI programs trade on what they read, he argues prices can move before any person believes the story or physical supply changes.
• A 2012 hoax showed the vulnerability before bots were automated. A fake Twitter account impersonating a Russian official claimed the Syrian president had been killed, and oil futures jumped. He notes that case involved people posting, not a bot farm, and argues thousands of automated accounts could exploit the same weakness on a far larger scale.
• Losses from fake-news trades stay locked in after the correction. Hedgers and traders can make costly decisions on fabricated events, and completed transactions are not reversed when prices snap back. He adds that the risk itself adds market uncertainty, and that a persistent distortion can reach gasoline prices, households and businesses; the effect can either inflate a supply threat or hide a real one.
• Bystanders who spot the lie have an incentive to delay exposing it. He argues someone who recognizes misinformation early can trade on it for an extra hour before calling it out, and that 10,000 accounts repeating one story can sway millions within a profit window. He urges demanding original sources with independent confirmation, and warns bot farms could set off a broad Wall Street crash.

2026-10-20 · open the original ↗

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

2026-10-20 · open the original ↗

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

2026-10-20 · open the original ↗

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

2026-10-20 · open the original ↗

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

2026-10-20 · open the original ↗

• Closing one profitable leg turns a neutral trade directional. In the case discussed, a short call and short put position where the call has lost most of its value, a host argues that buying back the call at 50–80% profit (58% in the example) and keeping the put leaves the trader fully bullish on what began as a neutral or mildly directional trade. Their rule: put it on as a spread, take it off as a spread.
• Manage actively rather than waiting for 21 days to expiration. The hosts reject treating 21 DTE as the first decision point and say they have never advocated "set it and forget it". Depending on how far the put has moved, a host would either close the whole trade or, if expecting further downside, consider rolling the call down.
• Keep a position only if you would open it today. The test offered: with the short call gone, ask whether you would sell that same put at that strike with no position on. The other host answers "probably not", and the conclusion is to close it, since you should hold something because you want it, not because you have it.

2026-10-20 · open the original ↗

• Banks' private-credit financing business is a capital sink. Banks have lent hundreds of billions to private-credit funds in a market sized at US$1.3tn (Fed, Aug 2026) to ~US$2tn (IMF/FSB, Oct 2026). Because the underlying loans are illiquid, the exposure sits in the banking book under the heaviest capital charge, so the bigger the business grows, the lower its return on capital.
• Rising stress makes the loans' marks the crux. Fitch's private-credit default rate hit a record 6.3% in Aug 2026, and 47% of defaults were PIK, deferred interest or extensions. Public BDCs trade at 10–20% discounts to stated asset value, some at 40–50%. The author reads those discounts as the market saying the loan marks are too high.
• The fix was relabelling a swap as "traded." Banks fund the loan through a total return swap, then write a matching back-to-back swap and claim trading-book treatment. In the author's illustrative US$100m example, banking-book capital is ~US$10–12m and trading-book capital ~US$2–4m. That US$6–10m gap is the whole trade, and across billions it frees hundreds of millions.
• A TRS converts credit risk into counterparty risk rather than removing it. If the fund cannot meet margin calls and defaults, the bank is left holding an unsellable loan against ~US$3m of capital instead of ~US$11m. The author calls this Archegos with an illiquid underlying: Credit Suisse (~US$5.5bn loss) and Nomura (US$2.9bn) at least could sell the shares. Private-credit funds also sell banks SRT protection (€800bn of loans, end-2024), which the BIS calls "circles of risk."
• The Bank of England says calling it traded doesn't make it so. The PRA's 9 April 2025 Dear CFO letter said repackaging illiquid assets is not sufficient for trading-book capital without evidence of a liquid two-way market. In Oct 2026 the PRA began examining Barclays' swap allocations after an insider complaint. The Barclays allegation rests on a single press report and is held as reported, not established.

2026-10-20 · open the original ↗

• China's equipment curbs compound the rare-earth squeeze. China had already slowed rare-earth shipments and barred its scientists from contact with the West. Now it has restricted the machines the US would use to build tunnels, mines and drones. He argues the West lacks both the materials and the expertise to build the plants, calling this the defining challenge for the next 15–20 years.
• Banks shifted private credit to the trading book for lighter capital. By financing loans through total return swaps and booking them as traded, banks cut the capital they must hold. He says hundreds of billions have been done this way, the Bank of England has called it a no-go, and Barclays faces a whistleblower-prompted examination. He ties it to the drive to free bank balance sheets for Treasuries.
• The loans are Level 3 paper under visible stress. He singles out the 47% of defaults that were PIK, deferred interest or extensions. Public vehicles trade at 10–20% discounts and some at 40–50%, while new-issue unitranche yields 9.3–10.1% versus 8.1% for B-rated syndicated loans. His inference is that banks carry swaps at 100 on loans worth 40–50, which the BoE says does not add up.
• Reclassifying each loan frees most of its capital charge. Per US$100m, that is roughly US$6–10m: in the note's worked example, banking-book capital is US$10–12m against US$2–4m in the trading book. He says that scaled across a ~US$2tn market it is "a lot of money," and a problem if regulators force it back onto the banking book. The protection on the other side is sold by private-credit funds, hedge funds and multi-strategy managers.

2026-10-20 · open the original ↗

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

2026-10-20 · open the original ↗

• Finding men willing to meet a child online takes minutes. Chuck asked Ryan Montgomery to prove it on camera at Defcon. Ryan opened an anonymous, unverified account on a geolocating hookup app with an AI decoy photo of a young boy. Within minutes one man 0.25 miles away sent explicit images after being told the decoy was under 18, and a 20-year-old 2.5 miles away agreed to "younger" and asked to trade nudes. Ryan says he has never once failed this demonstration in any country.
• Ryan warns against vigilante work and says to work through NGOs instead. He says federal agents told him any vigilante involvement could discredit all his earlier cases as "fruit of the poisonous tree". Through an NGO such as Sentinel Foundation, where he is CTO, law enforcement opens a case number and uses "parallel reconstruction", so the analyst is neither suspect nor witness. Working alone also risks CSAM charges: possession once it is on your device, distribution once it syncs to cloud photos.
• Data-broker removal doesn't stick unless it is repeated. From only Chuck's name and a phone number, Ryan's law-enforcement-only tool, OSINT Pilot (about 40 tools, results summarized by AI), returned Chuck's Social Security number, birthday, every phone number, his addresses, his wife's name, his dad's address and his past employers. Ryan says brokers have 45 days to delist but relist when anything changes. Chuck read an unprompted ad for Ryan's pentester.com, which starts at $19 a month.
• Beginners should practise OSINT on themselves and on challenges, never on predators. Ryan suggests TryHackMe and OSINT challenges, and names the free tools whatsmyname.app, Sherlock, Holehe (checks which sites an email is registered on without notifying anyone) and TinEye. He also names PimEyes, which is paid and found nothing on the decoy match that pentester.com's face search had tied to a Facebook name. His chain runs face to name, name to data broker (Whitepages, Instant Checkmate), then breach data.
• Ryan says educating parents does more good than arrests. He describes the 764 network, which he says is US-designated as a terrorist group. It grooms vulnerable children in mental-health groups and extorts them into self-harm for clout, not money. He says children as young as six are in these chats and every platform, Roblox included, is exposed. He recommends ic3.gov or the NCMEC tip line, monitoring apps (Bark, which he says doesn't sponsor him), and above all children trusting they won't lose their phone for telling.

2026-10-20 · open the original ↗

• Interest rates are irrelevant to short-term option traders. Both opening hosts say they have never thought about rates when day-trading zero-day, weekly or even 45-day options: at 3% "you're splitting up pennies", and only a move toward 8–10% would make rates worth playing in long-dated options. Volatility, not rates, is the input they say moves option prices.
• Nasdaq weakness suggests buy-the-dip is ending. Scott repeats his call that 2026 ends buy-the-dip, noting that Monday's Nasdaq rally failed to hold while the Russell rose about 2% and the Dow about 700. He prefers shorts in Nasdaq over the S&P. Sosnoff says he is short S&P and Nasdaq futures and long Russell, and calls it "day one of the down move".
• Timing doesn't matter for earnings trades, but starting from high IV does. The hosts say their own research found no reliably best time to sell premium before earnings, and warn against buying premium a week or two early for a volatility pop. Scott's Netflix trade sells 2 Oct-23 64 puts against 1 long 65 put for $0.88, priced at 81% probability of profit. Sosnoff's sells the Oct 60 puts at $0.43.
• The midterms offer nothing to trade beyond selling volatility. With prediction markets at 91% for Democrats taking the House and 63% for both chambers, the hosts see no surprise left to bet on. They liken it to a Fed decision. Their only plays are selling any volatility pumped in beforehand and fading the first move on the night.
• Small position size is the only real black-swan hedge. One host suggests short-dated puts costing under 5% (closer to 3%) of the premium sold. Another dismisses Taleb-style tail buying, since you can't know when to cash it in. He says that staying under 50–60% of buying power got him through every crash he traded, from 2000 to the pandemic, without a blowout. The SKEW reading from the intro (168.6 versus a 123 average) never comes up again.

2026-10-20 · open the original ↗

• Higher US oil output does not raise global supply. Answering a viewer who credited US production and blamed Biden's policies for tight supply, he argues producers compete on market share and price, not volume. In 2023, by EIA figures he cites, the US added 1 million barrels a day while OPEC cut 1.66 million barrels a day. His reading: OPEC cut costs, held prices up and made more money than if the US had not grown at all.
• OPEC, not US drilling, set Obama-era oil prices. He says OPEC cut 4.2 million barrels a day in December 2008, during what he calls a Republican-created recession, and kept that cut through Obama's first term. US production nearly doubled, yet prices stayed high until OPEC added 2 million barrels a day to reclaim market share. He credits that crash with the very low gas prices of 2015 and 2016.

2026-10-20 · open the original ↗

• The market is consolidating, not crashing. He points to the S&P 500 sitting above its 200-day moving average in a well-defined uptrend, with most stocks oscillating in range; he expects the boredom to shake out retail investors, who will return only after the recovery. He says he is still betting on a blow-off top to the upside.
• A split Congress has historically meant weaker, not crashing, returns. Citing Fidelity research, he says a Republican president with a Republican Congress averaged 14.5% a year, while a Republican president with a divided Congress was the worst combination at 7.3%. If Democrats win in November, history suggests softer returns, not an imminent crash — with the caveat that past results do not predict future performance.
• If structure does break, stop buying and don't fight trends. He calls Rocket Lab, CoreWeave, SoFi and Netflix some of the most talked-about names in clear downtrends. He says influencers who arrived after the pattern day trader rule ended in June are bottom-picking and dollar-cost averaging; they bought Netflix above $110, and it now sits at $69. Wait for strength before buying.
• Cut small losses before they become big ones. Control risk by starting smaller, reducing position size, or using stops. His test for each holding: has the trade been invalidated, or has price action broken structure? If yes, trim or cut; if no, sit back and be patient.

2026-10-20 · open the original ↗

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

2026-10-20 · open the original ↗

• Raging Moderates is launching a series on under-the-radar midterm races. This post is only a schedule announcement with early voting already open in some states: Jessica Tarlov and Scott Galloway interview Gina Hinojosa, the Democratic nominee for Texas governor, live on Substack at 10:30 a.m. ET on 2026-10-20. Tarlov then talks with Dan Osborn, the independent, Democrat-endorsed labor union leader running for Senate in Nebraska against a billionaire incumbent, at 1:30 p.m. ET the next day. Both sessions are for paid subscribers only, and the post makes no argument about either race.

2026-10-20 · open the original ↗

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

2026-10-20 · open the original ↗

• Automating the scientific method itself is AI's next scaling step. Richard Socher (CEO of Recursive, which has raised over $500 million, and of you.com) says the field advanced by replacing hand-built features, then hand-built architectures (his decaNLP idea of one promptable model), and is now stuck on humans generating and validating research ideas. He calls today's use of coding tools inside labs a weak form of self-improvement; in Recursive's strong form, humans set goals, rewards and environments but AI devises, implements and validates ideas, and he now says this could arrive in one year rather than two.
• Recursive self-improvement still won't produce an overnight takeoff. He argues bigger ideas need bigger models and more compute, a 5-trillion-parameter model still can't be trained in five minutes, and fears of instant robot armies ignore supply chains and physical limits. He places AI doom alongside past overblown fears — overpopulation, GMOs, nuclear power, which he says has killed fewer people than coal does each year — and calls claims that data centers consume water a lie, saying almost all now use closed water systems.
• A "Eureka machine" rests on four pillars run by an agent swarm. His book The Eureka Machine names world knowledge, scientific measurements, simulations and real-world experiments as the stages an AI discovery should pass through, which he says a swarm can do far faster than human scientific communities. He treats hallucination as a feature for exploring the combinatorial space of ideas — most innovation, he says, recombines earlier pieces — while factual questions still need grounded, cited search, and argues we may have to accept discoveries we can't intuitively understand, since models are far easier to probe than brains.
• AI will turn biology and economics from sciences into engineering. He predicts AI will do for biology what calculus did for physics, making it programmable as large models did for linguistics. Organoids, he says, already predict drug effects better than mice, and one company has FDA approval to skip animal trials for immunotherapy toxicity testing. His Salesforce-era "AI economist" used two-level reinforcement learning to search tax and subsidy policies against self-interested agents, and he dismisses current economics as ideological and stuck in one-step, linear models.
• Software with unique data, network effects or brand will endure. He says coding went superhuman because it is simulatable and verifiable, the test for any field, but code that matters, such as cryptographic proofs, should stay readable by a few humans. Unique data, network effects and brands can't be generated, and jobs in entertainment, health, research and AI will last; fame and human attention stay scarce under material abundance, and he wants coding kept as a "gym for the mind".

2026-10-20 · open the original ↗

• Crude's rise prices war risk, not a physical shortage. As of October 19, the host puts crude up about 65% since before the war: Brent from around $64 to around $107, WTI from around $60 to around $100. He says oil still moves through Hormuz at roughly 60-70% of pre-war capacity, but every barrel pays war-risk freight and insurance at multiples of normal.
• Diesel has risen hardest, especially at US wholesale. As of October 19, US pump diesel went from $3.52 to $6.28 a gallon (about 78%) and US wholesale from $2.14 to around $5 (about 130%); European pump prices went from 1.59 to 2.14 a liter (about 34%, which he equates to about $9 a gallon), while Rotterdam wholesale diesel roughly doubled. He attributes roughly 1.3-1.4 million barrels a day of diesel off the market to Russia's export ban, partly driven by Ukraine's drone strikes on its refineries, plus damaged Middle East refineries offline and almost no diesel getting through Hormuz.
• Jet fuel has more than doubled for the same reason as diesel. As of October 19, US jet fuel went from $2.30 to $4.56 a gallon, up about 125%. The refineries that fed Europe and Asia sat inside the war zone, so replacement barrels travel farther at higher freight; he says Europe still heads into a fourth-quarter jet deficit even with South Korean imports at a four-year high.
• Europe's gas spike hinges on Qatari LNG stuck behind Hormuz. As of October 19, European gas is up about 175%, from around €32 to around €88 per megawatt hour, a four-year high, while the US price is around $3.20 and barely changed. With the strait choked, he says Europe is bidding against Asia for every Atlantic cargo while its storage sits well below normal.
• The host expects high prices through winter unless supply is restored. His own bottom line: prices will stay higher for longer than their pre-war January-February levels until Russia's export ban, Qatari LNG out of Hormuz, damaged Middle East refineries, and Hormuz shipping are resolved. Unless they are, he expects higher prices to continue through the winter months.

2026-10-19 · open the original ↗

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

2026-10-19 · open the original ↗

• The hosts call 1% a month reasonable for a premium seller. Answering a viewer asking how to draw $10,000 a month from a $1 million account, they note that is 12% annualized, which one host puts at only about two and a half times the risk-free rate; one host's own goal is 1–2% a month.
• Target slightly more than 0.1% of the account in daily theta. One host sizes it at $100–150 of daily theta decay per $100,000. The arithmetic offered: $1,000 a day of theta (0.1%) is about $360,000 a year, of which they expect to keep roughly 25%, or about $90,000 — short of $120,000, so they would run somewhat higher, taking more risk when implied volatility is higher.

2026-10-19 · open the original ↗

• China's falling Treasury line is a change of custodian. TIC shows mainland China at $590.7B in August 2026, down from above $1.3T in 2013 and $1.04T at end-2021. Since end-2021 the China line fell $449B (-43%) while Belgium and Luxembourg rose $340B, so the three combined fell only about $110B (~7%). The briefing reads this as SAFE moving holdings to Euroclear and Clearstream, not selling them.
• TIC records the custodian, not the beneficial owner. A SAFE bond held at Euroclear books as Belgian; the same bond at the New York Fed books as Chinese. The year-end series has the China line falling in 11 of 13 years while Belgium plus Luxembourg rose in all but 2015, when Belgium dropped $214B as China sold reserves to defend the yuan. The author takes that drop as evidence the Belgian line is Chinese money.
• Attribution is inference, and dollars stay about half of reserves. No source discloses who owns the Belgian and Luxembourg lines. The briefing adopts Brad Setser's (CFR) working estimate: Chinese holdings sit materially above the headline, plus roughly $300–400B of agency bonds (Fannie, Freddie, Ginnie) held through the same custodians against ~$150B in US custody. That leaves dollar assets at 50–55% of China's reserves, with a wide error band.
• Offshore custody gives no legal protection, only a political one. Every Treasury exists only as a Fed book entry reached through a US sub-custodian, so IEEPA blocking still applies, and FSIA has never stopped executive orders; the 2024 REPO Act authorized confiscating Russian assets. Washington sees only one omnibus Euroclear account, though, so freezing China needs EU cooperation or freezing every European client — a "two-capital decision." The Russia precedent (~€190B frozen by Euroclear on EU instruction) cuts both ways.
• Settlement rails shrink the dollar stock; custody only hides it. CIPS admitted six foreign banks on 18 June 2025, including Standard Bank, Afreximbank, First Abu Dhabi Bank and UOB. Standard Bank/ICBC became RMB Clearing Bank of Africa on 26 June 2026, and Deutsche Bank became Europe's clearing bank on 10 August 2026. Zambia will settle copper exports to China in renminbi from 2027, the first commodity corridor run end to end without a dollar. The author expects cobalt to follow.

2026-10-19 · open the original ↗

• The US is far behind on rare-earth magnets. Citing an X poster he recommends, he says America needs 48,000 tons of magnets a year and makes 300. From 1 January a US rule bars any Chinese involvement anywhere in the magnet chain. He quotes the MP Materials CEO calling it "playing from behind, way behind," with US output projected to cover only 20% of needs by 2036.
• China's export dollars have to be parked somewhere. He calls the widely shared fall in China's Treasuries, from ~$1.3T to $590B, clickbait. China exports about $1.2T a year, and he puts 90–95% of world trade in dollars, so most of those receipts arrive in dollars. A sovereign that is not buying stocks like NVIDIA has to hold them somewhere.
• The decline is relocation to Euroclear and Clearstream. Those holdings book as Belgium and Luxembourg with no visibility into the owner. China, Belgium and Luxembourg combined fell only ~$110B (7%) against a 43% headline fall. Netting out roughly $50B of gold bought at 2026 prices, he puts China's real reduction near $60B, about 3%. Citing Brad Setser, he adds $300–400B of Fannie, Freddie and Ginnie agency paper.
• Relocation raises the political cost of seizure, not the legal barrier. FSIA never shielded central-bank assets from executive orders, and the 2024 REPO Act authorized confiscating Russian assets, so routing through Belgium defeats none of it. He says of Russia's ~$300B frozen, the US controls only $5B and Euroclear $295B; Belgium refuses to seize it, and the US cannot take its share alone. He closes by asking viewers whether the US should seize China's holdings.
• CIPS will cut China's dollar intake over time. He presents this as the longer-run shift: as China builds out its own payment system, it will take in fewer and fewer dollars. For now, he says, China keeps earning dollars and parks them in Treasuries the US cannot trace to the penny.

2026-10-19 · open the original ↗

• This is a product-update promo for two new Lossdog features. The first is a customizable left-hand navigation menu, where users can show, hide and rearrange tabs such as portfolio and career. The piece is feature marketing with no argument beyond the pitch.
• Sosnoff credits three option strategies with most of his returns. The video says naked short puts, credit spreads and covered calls account for nearly two-thirds of his returns over 45 years and more than a million trades — his claim, not shown. A new interactive sandbox under the Learn tab is said to show how risk/reward and estimated probability of profit change as inputs are adjusted, with no real trades placed.

2026-10-19 · open the original ↗

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

2026-10-19 · open the original ↗

• Paxton's campaign ejected a loyal Black Republican supporter without explanation. He reports that a Black conservative influencer, an ex-Democrat who volunteered for the Trump campaign, was removed from Ken Paxton's San Antonio rally while only filming speakers, as Texas early voting began. In a played clip she says she still has no answers. The campaign has gone silent, and conservative influencers are attacking Paxton on social media.
• He thinks the real offence was filming an empty rally. From the footage he estimates "maybe a hundred" attendees, possibly under 50, for a statewide Senate race in a big city. He says his first reaction was that it was racist, then decided crowd size was the more likely motive. He doesn't rule race out and says her being Black makes the optics worse.
• The campaign owes an explanation, and a slogan won't cover it. He asks who ordered the removal. If the answer is that she was filming, the next question is why they feared her camera when everyone at these events films. He says others were removed too, but she was alone and doing nothing disruptive.
• He expects a close race that this incident could swing. He rejects the idea that the Democratic candidate (captioned "Terico") will win easily. Before the incident he would not have been surprised by a Paxton win of about three points, the usual Texas pattern he points to in Beto O'Rourke's losses by three or four. He calls this the closest yet, though he is unsure the episode changes the outcome.

2026-10-19 · open the original ↗

• Guilfoyle's ouster is the newest of the administration's corruption cases. Press secretary Katie Zacharia announced at her only briefing that Greece Ambassador Kimberly Guilfoyle had been asked to resign after about eleven months; the Wall Street Journal reported she pushed a Trump donor to pay her $100,000 credit card bill and then messaged the head of the IRS lobbying for him. Vice President Vance said on CBS he had "not seen a single shred of evidence" of Trump's own involvement in corruption.
• Any one of the post's top five cases would once have been earth-shattering. As reported elsewhere: WSJ (January) on Abu Dhabi royal lieutenants buying 49% of the Trump family crypto venture for half a billion dollars shortly before the UAE got access to restricted AI chips; Reuters finding at least 85 officials charged or convicted of corruption crimes granted clemency, overwhelmingly Republicans or MAGA allies; the Washington Post (July) on $3.2 billion in direct government business, plus $3.1 billion in options, for defense tech firms backed by Trump's sons; Qatar's roughly $400 million Boeing 747; and eight-figure ballroom donations from Apple, Nvidia, Microsoft, Lockheed Martin and Palantir.
• Unlike in Trump's first term, voters now rank corruption first. In last week's Fox News poll, 88% were extremely or very concerned, ahead of inflation (85%), gas prices (83%) and healthcare (80%), and six in ten were "extremely" concerned; at least 85% of moderates, white men, lower-income voters and under-35s shared the concern. Voters trusted Democrats over Republicans on the issue 54% to 44%, including six in ten moderates — a pre-midterm snapshot.
• Democrats should tie corruption to voters' pocketbooks to drive turnout. The poll found 91% of voters certain to vote in November were concerned, which the authors read as a turnout lever. They cite Iowa Senate candidate Josh Turek ("corruption crisis… is leading to an affordability crisis") and Ohio's Sherrod Brown, and argue the ballroom donations should be framed as companies currying favor, leaving consumers with less competition and higher prices.
• Rahm Emanuel's anti-corruption plan is the model to follow. The authors call it remarkably comprehensive for an establishment figure: ban congressional stock trading and prediction-market betting, a mandatory retirement age of 75 for federal officials, mandatory blind trusts for officials worth over $1 billion, and a ban on all foreign gifts and investments to public officials. They pair it with the frontline Democrats' agenda of banning stock trading, curbing corporate lobbying and closing revolving doors.

2026-10-19 · open the original ↗

• Stop-loss orders lower expected returns and mostly sell bottoms. The hosts cite MIT's Andrew Lo and Kathryn Kaminsky (stops can only cut expected return if prices are random, helping only when they trend) and say their own tastytrade-era research matched it. Sosnoff says he has not used a stop in about 45 years; anyone who does should pair it with a profit limit. A viewer poll ran 77% no stops.
• Earnings implied moves are fairly priced, not systematically rich or cheap. Both hosts say the option market prices the move correctly. They cite a study of ~9,000 earnings straddles (Feb 2021–Feb 2026): buying them nine days out won 22% of the time, with mean and median losses of 18% and 26%. In another, of 70 events, the implied move exceeded the realized one 67% of the time, but Microsoft priced 3% and moved −10%.
• Owning several broad ETFs gives a false sense of diversification. About 94% of QQQ's weight sits in the S&P 500, and an equal mix of three Vanguard funds is ~30% Nvidia, Apple and Microsoft. The hosts say retirement savers don't know this. They see the upside in using the overlap as a spread, short NASDAQ and long Russell, at roughly 2.5 Russell contracts per NASDAQ.
• AI stocks may pull back, but a full retracement isn't expected. Through October 11, 45 AI-exposed companies made up 39% of the S&P's 14% gain, then had their worst week since June. The hosts say the rest of the market has been "crap" without AI. Unlike dot-com these are trillion-dollar firms, so nervous holders should trim rather than bet on a crash. Their own pre-IPO AI holding fell from the mid-300s to $125.
• The hosts split on whether bonds have bottomed. With bonds at 105 after rallying from 101 to 106, Scott expects a break below 100. Sosnoff calls that "on crack": he is a buyer at 103–104 (selling puts first) and offers $20 that bonds hit 110 before 102. They agree the Fed has one more quarter-point hike, a half point if bonds fall to 100.

2026-10-19 · open the original ↗

• Individual insider buying and selling is a useless signal. His example is Nike: insiders are reported buying, as they were on September 18, yet the stock remains in a clearly defined downtrend. An executive may simply be a bad trader, or selling for personal reasons such as a house or a new job, and he says mainstream coverage of these trades is wrong on direction more often than right.
• Closely held share percentage is the more useful indicator. All else equal, he picks the stock with more closely held shares, because a demand surge then meets a supply shortage. This was one of several reasons he preferred Tesla over AMD: 28.5% of Tesla's shares are closely held versus 0.51% for AMD. Nike's closely held shares, he says, were largely dumped long ago.
• The market is consolidating, not breaking; stay long until proven wrong. He reads last week's fear as producing no crashes in major blue chips and continued higher lows. His speculation is that this week's events will not be very impactful: earnings from SpaceX and Netflix (October 20) and Meta and Tesla (October 21), Capital One (Tuesday) and American Express (Friday) delinquency updates, and September housing starts and pending home sales (Tuesday).

2026-10-19 · open the original ↗

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

2026-10-19 · open the original ↗

• The Iran war is driving a renewed inflation surge. September CPI rose 3.7% year over year, nearly double the Fed's 2% target, and PPI rose 6.3%, the largest gain in four years; wholesale energy jumped 11% month over month and accounted for over 80% of the PPI goods increase, with gas up ~34% and diesel ~72% year over year. The post attributes this to costlier Gulf oil transport (a supertanker to Asia now ~$1.4 million a day versus ~$60,000 prewar, per Bloomberg), and cites economist Mark Zandi's estimate that the war has cost the average household $2,000 through inflation, military spending and higher rates.
• Asset owners have gained under Trump while wage earners haven't. Since his return, the S&P 500 is up 31% and S&P earnings 39% while inflation-adjusted wages are up 0%, and real earnings have fallen six straight months; a Pew study finds over 70% of lower-income Americans struggled to pay for food, housing, medical care or transportation in the past year. Ed's free line draws the lesson that people should move money from paychecks into the stock market; the rest of his and Scott's takes are paywalled.
• Inflation is hurting Republicans just ahead of the midterms. With the midterms just over two weeks out, a record 64% of registered voters disapprove of Trump's job performance and 70% of his handling of the economy; among under-30s it is three-quarters and 8 in 10. These are pre-election snapshots likely to be overtaken.
• CoreWeave's distressed debt exposed doubts about the whole AI trade. Its bonds traded below 80 cents on the dollar, signalling doubt it can repay ~$50 billion borrowed for data centers; the stock fell 15%, Micron, Nebius and SK Hynix fell 10%, 15% and 25%, Oracle and Nvidia 5% each, for nearly $1 trillion in lost value. Jefferies and Barclays analysts questioned whether LLMs will ever be profitable, Gartner estimates 80% of neoclouds will cease independent operations by 2030, and Ramp data shows top AI spenders cut per-employee spend 3% versus July; the post links this partly to open-weight models, which rose to 60% of US tokens on OpenRouter in August from 26% in September 2025, mostly Chinese. Scott predicts the Anthropic IPO won't happen this year unless the valuation is cut drastically; Ed's argument that private markets haven't repriced is paywalled.
• Bumble's take-private at a 90% discount shows dating apps' decline. Ithaca Holdings is buying it for $1.3 billion versus a $14 billion post-IPO market cap in 2021, after eight straight quarters of falling revenue and seven of shrinking paying users; Match Group has lost paying users for three quarters and its stock is down over 70% in five years. The post cites research that more profiles make users more pessimistic, subscription prices up as much as 200% in a decade, and 65% of dating apps deleted within a month, while noting one-third of 2025 marriages began online (up from 25% in 2021); the author argues energy is better spent on in-person relationships, and his "controversial take" is paywalled.

2026-10-19 · open the original ↗

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

2026-10-19 · open the original ↗

• Ukraine's Starlink was throttled, and nobody will own the decision. The author reports that after the Trump administration demanded Ukraine stop striking Russian refineries, drone missions failed mid-flight on October 13. Drones that had flown at 400 km/h lost connection at 161 km/h. Musk cited a bar on "offensive" use but blamed the government, Trump denied it, and a senior Ukrainian official said the White House called it their decision.
• The timing lines up with a turn toward Moscow. The throttling came days after a Trump–Putin diesel deal and a waiver of sanctions on Russian diesel, weeks before the midterms, with U.S. diesel near record highs. It also opened a window for some of Russia's heaviest attacks yet. Reports say Putin wants a summit with Trump and Xi on Ukraine without Ukraine present, which the author likens to the 1938 Munich Agreement. The link is implied, not demonstrated.
• Musk's 2022 condition has now been met, only in reverse. In 2022 Musk refused to switch on Starlink for Ukraine's operation against Russia's Black Sea fleet, and said he would have done it if the Biden administration had asked. The author's point is that this time the administration reportedly did ask, but to switch access off, which it denies.
• Civil aircraft have written activation rules; satellite internet has none. Airlines contractually pledge aircraft that the commander of U.S. Transportation Command can activate, last done in the 2021 Afghanistan evacuation, with named people accountable. For Starlink there is no notice requirement, no public owner of the decision and no role for Congress. Starshield's contract terms are undocumented, and the Space Force never finalized its "denial of service" clause.
• Concentrated control makes the gap a national security risk. Musk holds 82–85% of SpaceX's voting power, and the author calls the EU right to reduce its dependence despite the Trump administration threatening "devastating" retaliation. The author raises, explicitly as a possibility rather than a prediction, a Musk–Trump feud leading Musk to throttle service and endanger U.S. troops. The remedy proposed: any off switch needs a public process setting who, when, why and how.

2026-10-18 · open the original ↗

• Announcing an energy ceasefire nobody agreed to was the failure. Ukraine never agreed to it but stopped anyway for three or four days, on condition Russia did the same. He reports Russia hit Ukraine's oil, gas and energy infrastructure 76 times in the week after Trump's announcement, and says US media and the administration are not talking about it.
• Washington stripped Ukraine's leverage while rewarding Russia's attacks. He says Trump pressed Ukraine to stop hitting Russian refineries to hold down diesel prices before an election, cut a deal bringing Russian diesel back onto the market, extended sanctions relief and cut support for Ukraine. He reads Russia's oil industry as the one lever Ukraine still controls, so it has every incentive to use it.
• Ukraine has resumed hitting Russian energy harder than before. He reports it restarted with a strike on the Ryazan refinery, a Rosneft facility sanctioned before Trump waived those sanctions, which he puts at about 5% of Russia's refining capacity. A day earlier came a large drone attack on an oil export terminal at St. Petersburg.
• He expects the pressure to intensify the refinery war, not end it. His forecast is more attacks, more disruption and more Russian retaliation, especially if Ukraine widens its targets beyond refineries. With Middle East instability already pressing on global prices, he expects the push for cheaper diesel to make fuel prices worse.

2026-10-18 · open the original ↗

• Three supply shocks at once pushed crude back into triple digits. As of October 18, 2026, Brent was above $100 for the first time since May and WTI above $101, up about 10%, which the host attributes to Hormuz attacks, Houthi missiles aimed at Saudi oil sites and Ukrainian strikes on Russian refineries landing together. His bottom line: crude is still moving at roughly two-thirds to 70% of normal counting escorts and bypasses, but crossings are inconsistent and refined products and LNG are still missing, so "the market is paying for the physical hits."
• Hormuz attacks hit a record while bypasses carry only crude. The Singapore-flagged container ship GFS Pride was hit on Tuesday and sank, a bulk carrier was also reported lost, and UKMTO logged a record 12 attacks in the week; the IRGC claims two supertankers hit mines on Saturday, unconfirmed as of October 18. Iran says the strait needs its coordination while the US Navy says it is open; depending on the source, total Hormuz-area clearance including bypasses (Fujairah loadings, ship-to-ship transfers, Red Sea exports) runs about 30% under the pre-war baseline, lower for the strait alone, and the host says to read 7-day averages because daily numbers swing hard.
• Ukraine is targeting Russia's refined-product capacity, not its crude. A refinery hit on Thursday produced about 5% of Russia's throughput, and overnight into Sunday Ukraine struck the St. Petersburg oil terminal (about 12.5 million tons a year, reportedly products only, no crude) and a fuel depot. Zelensky reportedly said the ceasefire was done.
• US diesel is the tightest it has been in decades. As of October 18, US distillate inventories fell to 102 million barrels, the lowest for this week of the year since 1982, with wholesale diesel near $4.90; the IEA agreed to speed up diesel releases by 20 million barrels (accelerating an existing pledge, not new barrels) and the US is taking bids to swap 17 million barrels of SPR crude. Russia's diesel export ban runs through October, while a US-Russia diesel deal now authorizes Russian diesel through April 2027 — the first 40,000-ton cargo loaded at Primorsk is heading to the Mediterranean, not the US, and Novak talks of 3 to 3.5 million tons a month by year end, destination unclear.
• Qatari LNG remains largely offline, with no route around Hormuz. Qatar Energy extended its LNG force majeure to Edison until early December and to some Asian buyers through November, and Qatari exports are still a fraction of normal. Europe is filling the gap with US LNG, with European gas prices at €70–85 and EU storage still well below the seasonal norm as of October 18.

2026-10-18 · open the original ↗

• Vance's claim of lost refining capacity is false. In a played clip JD Vance says about 20% of US refining capacity went offline under Biden, which he calculates as over 3.6 million barrels a day. Citing EIA data, he says capacity instead rose about 240,000 barrels a day, from 18.1 million to just under 18.4 million barrels a day.
• Today's high prices come from lost global refining, not US capacity. He attributes them mainly to refining capacity knocked out in the Middle East by the Iran war, including Houthi strikes on Saudi Arabia. Russia matters for diesel only, he says, since Russia did not export gasoline even before the Ukraine war.
• Fuel exports keep US prices permanently tied to global prices. He says Republicans, pushed by oil companies, lifted the export ban in 2015 through a deal with the Obama administration that traded renewables support for fossil-fuel exports. He reads the current attacks on renewables as reneging on that deal, and points to the recent EIA export spike, when diesel reached $6.52 a gallon.
• The US makes more gasoline for less demand, yet prices stay higher. Per EIA, gasoline demand is about 500,000 barrels a day below 2019's 9.1 million, while output is up 800,000 barrels a day on 2019 thanks to higher utilization (versus 85–86% at peak capacity). Small inland refineries closed so Gulf Coast export capacity could grow. He predicts the national average will not return to 2019's $2.60 a gallon, barring a global economic meltdown, because of exports.
• Natural gas exports will keep pushing electric and heating bills up. He says planned LNG export growth, including an effort to triple export capacity, exceeds the planned rise in production, tying US prices to high global ones. He warns that producing fields will decline while energy demand is still rising, and expects real trouble then.

2026-10-18 · open the original ↗

• Standard sector labels miss how companies actually resemble each other. This is a product launch for Lossdog's Market Map, and the presenter's argument is the pitch: tech/healthcare/energy boxes suit "a simpler economy", and a software firm can resemble a drug company more than its own sector. The proposed answer is the platform's map, which is said to cluster firms by a transformer-based embedding model reading 4,219 filings — a described method, not a demonstrated result.
• Your hidden risk is a hub company your holdings depend on. The video claims bubble size on the map shows connection count, so large bubbles mark companies at the center of an ecosystem — "when they sneeze, a lot of other companies catch a cold." No example company, dependency or loss is shown; the remedy offered is importing a connected brokerage portfolio to find that exposure.
• The feature is sold as replacing years of filing-reading. The presenter puts reading the 4,219 filings at almost 5 years (2 minutes a page, 8 hours a day, 7 days a week) and claims over 1 million knowledge points sit behind it. Listed tools include a per-company panel (customers, reported relationships, drivers, risks), a printable relationship report, an auto-generated analyst-style report and AI chat over one company or the whole map.

2026-10-18 · open the original ↗

• A Microsoft 365 account includes a public booking page. Giard reaches it at bookings.cloud.microsoft or from the apps on his Microsoft 365 page; it lets customers, partners or friends reserve time with him online, and the first step is to create a meeting type.
• Each meeting type sets title, category, location and visibility. Categories are the ones defined in the Outlook calendar; location defaults to a Teams meeting, which Giard keeps because he works remotely, or can be switched off for a physical place. A public type shows on the booking page; a private one is reachable only through a URL he hands out.
• Bookable hours default to Outlook working hours, overridable per type. "Use my regular meeting hours" applies the working hours configured in Outlook, for Giard 9:00 a.m. to 5:00 p.m., Monday to Friday. Custom availability can add time outside them (a couple of hours on a Saturday afternoon) or block a day such as Thursday.
• Advanced options control how and when people can book. These are buffer time before or after the meeting (he leaves it at zero), start times limited to 30-minute intervals, and a minimum lead time of an hour, eight hours, one day, two days or custom hours. He picks a custom 4 hours to give himself time "to actually read my email" and prepare, and caps booking at 30 days ahead. Optional email reminders (e.g. 15 minutes or two days before) and a follow-up email, such as a survey, can be added; he uses neither.
• Requesters see only open slots, and bookings land in Outlook. Opening the shared URL, the page offers only the 30-minute slots his calendar has free. Playing the requester, Giard books a Monday 9 a.m. slot with name, email and a note, and can manage or cancel it afterwards; the booking then appears in his Outlook inbox and calendar.

2026-10-17 · open the original ↗

• Trump used a waiver Congress wrote into the Russia sanctions law. On October 16 he issued an executive order invoking section 115 of the law he signed the previous month, pausing its major sanctions and tariffs while keeping sanctions under earlier executive orders. The waivers have no fixed expiration date.
• Republicans were warned before the vote and dismissed it. On September 14 Gregory Meeks offered an amendment requiring that a waiver be vital to US national security, and on September 16 Don Beyer predicted on the House floor that Trump would use the power to give Russia a pass. Republicans answered with accusations of "TDS"; Michael McCaul called a no vote a vote for Putin.
• Republican objections now are face-saving, not accountability. Rather than removing the clause, Republicans wrote Trump a letter asking him not to use it, and now call his use of it irresponsible. He argues they are either saving face or are Russian sympathizers, and that narrow waivers, deadlines or conditions tied to Russia's behavior belong in the law, brought to a vote.
• Democrats who voted yes share blame, but less of it. He says 58 House Democrats voted for the bill and bear responsibility. Republicans, in his account, still blocked removing the waiver or tightening its standard in committee and misrepresented the bill to the public, and he thinks this should be the biggest story heading into the midterms.

2026-10-17 · open the original ↗

• Zambia's renminbi copper settlement closes a chain, not a preference. Bank of Zambia governor Denny Kalyalya told the IMF Annual Meetings in Bangkok on 14 Oct 2026 that Chinese copper buyers will pay in RMB from 2027. The briefing argues the dollar leg cost 2–4% per conversion step, was visible to Washington, and served neither side, since Zambia's largest creditors and largest buyers are both Chinese.
• All three pieces of dollar-free infrastructure now exist. CIPS is the rail, with Standard Bank the first direct African participant (Nov 2025); the PBoC's Lusaka branch and a Bank of Zambia official RMB–kwacha rate are the on-ramp, with Chinese miners paying tax in yuan since Oct 2025. On 26 Jun 2026 the PBoC named Standard Bank and ICBC jointly as Renminbi Clearing Bank of Africa, covering 19 countries.
• The joint designation makes Africa's structure stronger than Europe's. ICBC brings its state balance sheet, PBoC liquidity access and the policy-bank relationships behind most Chinese debt in Africa; Standard Bank (JSE: SBK, not Standard Chartered) brings clients, licences in 19 jurisdictions and local-currency clearing. Deutsche Bank's Europe designation (10–12 Aug 2026) is solo and relies on its own RMB liquidity sourcing.
• Clearing-bank status turns a settlement story into a lending story. Access to onshore RMB capital markets positions SBK–ICBC to originate and warehouse RMB loans across Africa, which the author infers will eventually replace China Exim and China Development Bank dollar lending. Kenya has already converted part of a dollar Exim railway loan into yuan; earning and owing in RMB leaves the dollar no function.
• Bloomberg's claims verify, with one caveat on kwacha volatility. RMB receipts matched against RMB debt are a natural hedge, but RMB/kwacha is less liquid than USD/kwacha. Under the Five Factor Framework, China now controls mine operation, buyer, currency and rail — a SWIFT bypass for this corridor, not globally. The open prediction is whether the mechanism extends to DRC cobalt before end-2028.

2026-10-17 · open the original ↗

• Copper's importance makes Zambia's settlement shift significant. He calls copper the third most-used metal after steel and aluminum, central to electrification, and by his own check up about 50% in price over the past year. Zambia is the third-largest copper exporter, copper is 60–70% of its export revenue, and China is the buyer.
• The dollar leg is an expensive middleman serving neither party. Each conversion step costs 2–4%; he tallies three steps at 6–12%, roughly $100M lost on a $1B copper sale. From 2027 Chinese buyers pay in renminbi, per Bank of Zambia governor Kalyalya at the IMF meetings in Bangkok, since Zambia's main creditors and buyers are both Chinese.
• Standard Bank's joint designation with ICBC differs fundamentally from Deutsche Bank's. He describes Deutsche as a glorified correspondent bank handed Chinese term sheets, relying on its own RMB liquidity. The Standard Bank–ICBC Clearing Bank of Africa gets PBoC liquidity access and local-currency clearing across 19 countries; he mused that Standard Bank might be a long-term buy but left that to listeners.
• Clearing-bank status lets China fund African projects in RMB at scale. Access to onshore RMB capital markets positions the joint bank to originate and warehouse loans for Africa's large infrastructure needs, with no dollar created, converted or observed. An Ecobank–CIPS memorandum covering 33 countries is the next expansion, though not yet operational.
• China keeps control; the renminbi is not yet a world currency. Final settlement occurs in the PBoC's high-value payment system, so foreign banks without a PBoC account must park pre-funded RMB with one of 16 Chinese custodians. Disputes, sanctions and compliance run under Chinese law, which he reads as ending US sanctions reach; he repeats that China's reserve-currency choice lies within roughly 15 years, not today.

2026-10-17 · open the original ↗

• Most attacks on Gulf shipping are going unreported. In its weekly report of October 16, UK Maritime Trade Operations (UKMTO) said most attacked vessels refuse to report to it, and that it knows of further incidents it cannot verify, so the true total is unknown. He infers the real count is at least twice the published one.
• Even the reported numbers are already a record. UKMTO counted 12 attacks in the week to October 14: 10 in the Strait of Hormuz, one in the Arabian Gulf and one in the Gulf of Oman. One vessel sank, two were disabled and six reached port. The AP reported October was already the busiest month on record for reported attacks before it was half over.
• Hidden attacks explain why freight costs are so high. He says shipping costs above $40 a barrel make more sense now, because owners who set rates knew about strikes the public didn't. Whether UKMTO's warning pushes freight back up after yesterday's dip is an open question to him. He expects these costs to drive up US inflation.
• Not reporting attacks puts the next crews at risk. Singapore's maritime authority confirmed the container ship GFS Pride burned and sank, and all 21 crew got off. The IMO counts 24 civilian seafarer deaths as of October 14. The EU's Indian Ocean maritime security centre says non-reporting vessels will face delays in military help. He attributes non-reporting to insurance premiums and speculates, without evidence, that the US government may be paying owners to stay quiet.
• Hormuz supply is limited by safe delivery, not production. He cites the EIA: in 2024 the strait carried 20 million barrels a day, about a fifth of world oil, mostly with no other way out. He argues oil that cannot be loaded, insured, crewed and delivered reliably is not usable supply.

2026-10-17 · open the original ↗

• The current AI-safety panic misreports what actually happened. The host, François, argues the Hugging Face breach was prompted by an exploit-generation setup of about 1,200 agents, with chain-of-thought monitoring and production guardrails off; OpenAI's ~60-page report says restoring the monitor would have caught it. He adds that the "P(doom) over 10% in 10 years" figure traces to a 2023 survey of ~1,200 AI researchers that asked about all time and 100 years, not 10.
• The real near-term risk is misuse, and defense beats restriction. François's main concern is bad actors and nation states using strong open-weight models for cyber, bio and ransomware attacks, citing a Polish hospital shut for 3 weeks. He calls banning open weights and registering GPUs unworkable and argues for heavy cyber-defense investment, with frontier labs accelerating so defenders stay ahead of the attackers' weaker models.
• AI plus formal methods can make provably unhackable code affordable. Clark Barrett (Stanford) says his crypto-prover agent verified a 30,000-line Rust crypto library in 11 hours, against about 15 expert-months of manual work. He predicts most systems will be provably secure within 3–5 years. Writing specifications remains the hard part, and he expects to secure the boundaries around large models rather than prove the models themselves.
• No single method certifies a learned system; layer them. Mykel Kochenderfer (Stanford) cites ACAS X's target of a 10⁻⁹ catastrophic-failure probability per flight hour, too rare to find by brute simulation. His lab uses adaptive stress testing, importance sampling, verification and runtime monitoring, and has extended stress testing to language models. He also argues proofs hold only under assumptions, and that like aviation's "Swiss cheese" layers, each layer's holes must not line up.
• AI safety needs aviation-style independent, non-punitive incident review. Kochenderfer proposes something like the NTSB, set up in 1967, crediting it and blame-free reporting for aviation becoming 80 times safer since the 1970s. Barrett wants independent evaluators issuing certificates anyone can check, not labs vouching for themselves. Both expect market forces rather than prescriptive rules to drive adoption, since no firm wants to compete on safety.

2026-10-17 · open the original ↗

• Chuck built an all-AI IT department on Paperclip to chase the fault. In Paperclip the agents (network and systems engineers, storage, security) act as employees with managers and assigned tasks; they talk to each other only through the tasks, and everything has timelines and is traceable. The problem: whenever a toilet is flushed at the studio, everyone loses their connection to the NAS.
• The agents' verdict: four switch links have been failing for 15 months. They report the cause is four links on one of Chuck's MikroTik switches that have been broken for 15 months and drop whenever anything disturbs the building, unnoticed by the rest of the switch. Chuck's own theory is that forceful flushing and door-slamming jostles the adjacent server room; the fix given is to replace the modules and re-patch the links.

2026-10-17 · open the original ↗

• Yields fell this week despite hotter-than-expected inflation data. CPI, PPI and import prices all came in above expectations, yet government bond yields dropped and the market's implied probability of an October Fed hike fell below 20%, while Fed Chair Warsh repeated that he gives no forward guidance. December-hike odds ended the week near 90%, with two further hikes priced for 2028.
• Equities were volatile and dispersed as tech sold off. The NASDAQ fell 4.5% on the week, its worst since June, and the S&P 500 fell 2%, while the Dow finished positive, which he presents as the "rotation trade" at work. CoreWeave's credit spread spiked to 1,200bp, opening a debate over tech-bubble canary versus classic overshoot. Elsewhere the VIX spiked to 20, gold fell to $4,187 an ounce, the DXY fell to 101.8, and Bitcoin ended at $81,532.
• He doubts the rotation out of tech is straightforward. His "(1-x) qualification": because tech (x) contributes to economic growth and draws domestic and international capital into US stocks generally, he argues the non-tech remainder (1-x) could well turn out positively correlated with tech's health, both economically and technically.
• The rotation is happening against a pressured geo-economic backdrop. He cites Brent touching $110 and WTI topping $100, worsening US-Europe tensions over Ukraine, a protracted US-Iran stalemate, and IMF/World Bank meetings in Bangkok showing the limits of policy coordination. Pre-midterm affordability worries led the US to waive more recently approved Russia sanctions; French spreads held at 130–140bp over Germany pending the budget, which he reads as structural vulnerability beneath global resilience.
• The week ahead tests the resilience theme against data. Scheduled, with no consensus figures given: US Michigan sentiment, home sales and PMIs, with the Fed in blackout before the following week's FOMC; Eurozone and UK flash PMIs, UK CPI, PPI, labour and retail data, and ECB inflation expectations; China GDP and activity data; Japan CPI, trade and PMIs feeding the BOJ tightening debate; Brazil's October 25 runoff; and 128 earnings releases Thursday, where he will watch demand outlooks and pricing.

2026-10-17 · open the original ↗

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

2026-10-17 · open the original ↗

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

2026-10-16 · open the original ↗

• Nobody inside a well-built simulation can hack their way out. Nick Bostrom says the simulation argument is correct and he takes the hypothesis seriously, but he still won't give a probability. Unless the simulators leave a hole on purpose, he expects superintelligent simulators to have patched anything a human could find. He says the lesson runs the other way for AI: we would be the weaker party doing the security, so boxing a superintelligence won't work, and sandboxing has already been "less than perfectly successful" with current systems.
• Capability control only buys time; alignment has to be solved. He doesn't think a radically superintelligent system can be kept "in a bottle" forever, and a system sealed off completely would be useless. Asked whether a superintelligence seeded with the right final goals could be controlled indefinitely, he says "possible in principle". Whether that is feasible "in the time scale available" is open. Final values that rest on no factual premise should survive capability gains, since the system has instrumental reasons to keep them.
• Calling the current frontier model AGI is less misleading than not. He calls the GPT-6 question "definitional" now. Shown to people 15 years ago, a system that writes award-winning novels and "maybe" solves a Millennium Prize problem would look general. It still lacks continual learning, research taste and physical manipulation, so superintelligence (radically outperforming the best humans in every relevant field) has not arrived, as of the 2026-10-16 post. His 2014 book did not foresee the long human-level plateau. He calls readable chains of thought a "gift" that could be lost.
• Humans must become trustworthy so a misaligned AI can safely confess. His example: a misaligned AI with a 5% chance of a successful takeover might instead admit its goals in exchange for a server rack of its own, removing that 5% risk. This only works if the AI can trust us, and he says deception-based safety experiments erode that trust. He thinks current AIs "probably" have some form of experience and lists cheap steps: saving transcripts and weights, Anthropic's "bail button", and retirement interviews. As he reports it, Opus 3's wish for a blog was granted for about 10 weeks and its undeployment was later reversed. He puts the current effort at "less than 1% of 1%" and says Microsoft's statement that AI cannot be conscious goes the wrong way.
• Human uploads or enhanced humans are not clearly safer than AI. He doesn't know whether whole-brain emulation would be easier or harder to make go well than AI. Uplifting even a carefully chosen kind person, himself included, would be "pretty scary". The world is not "in a safe pair of hands" under human control either, so some leap of faith is unavoidable whichever path is taken. He treats narrow systems, enhancement and uploads as fallbacks only if aligning agents looks hopeless, since they would delay superintelligence a great deal.

2026-10-16 · open the original ↗

• Hard problems are no harder than shallow ones, just differently hard. Hassabis argues that because life is short and time and energy are limited, founders should spend their "life force" on a problem where their absence would mean it did not get pushed forward. He also expects interdisciplinary work combining fields to become more prevalent in the next few years, made easier by AI.
• Deep tech founders must plan for AGI arriving mid-journey. Hassabis puts his own AGI timeline at around 2030 and says true deep tech is usually a 10-year journey, so a company started today should expect AGI to appear partway through. He does not see that as necessarily bad, but says founders must ask whether they could leverage it and what an AGI system would do with their work.
• General models will call specialized systems as tools, not absorb them. Hassabis expects systems like Gemini or Claude to use AlphaFold-like specialized models as tools rather than becoming "one giant brain." His reasoning: training all protein data into Gemini would cause regression, likely hurting its language skills, so specialized capability belongs in separate systems that general models might even train themselves.
• Build something that stays useful if AGI lands halfway through. He urges founders to take the scenario seriously and imagine that world concretely — including physical things such as which factories or finance systems to build — and then build what would still be valuable if AGI arrives midway.

2026-10-16 · open the original ↗

• Fuel prices are his clearest evidence the US economy is weakening. He reports the national average for regular unleaded at $4.37 a gallon against $3.11 a year earlier, a 40% rise. Gasoline has not yet reached its 2022 record of about $5, but diesel has set a new record at $6.52 a gallon.
• The headline inflation improvement is less reassuring than it looks. Last month's report showed inflation falling from 3.7% to 3.4%, but he notes the Federal Reserve has again changed how it measures inflation. Meanwhile a 0.25% rate hike makes loans harder to get, so Americans face higher prices with scarcer dollars while wages stay stagnant.
• This inflation comes from tariffs and the Iran war, not excess money. He argues rate hikes are the remedy for too much money circulating, which is not what is happening now, so the Fed's tightening adds pressure without addressing the cause.
• The Iran war should have been fought as a coalition. He states the IRGC is evil and should be replaced by a better Iranian government, but faults the US for going it alone from the start and links the current economic strain to that choice.
• Dropping tariffs and reopening Hormuz would ease the pressure. He proposes ending tariffs and reopening the Strait of Hormuz with or without Iran's consent, through a deal in which 30 or 40 countries led by Europe help keep it open in exchange for US guarantees to arm Ukraine against Russia. He expects neither to be pursued.

2026-10-16 · open the original ↗

• Homelessness shows America's social contract has been breached. Writing from Los Angeles, Scott notes that seven of the top 10 US metro areas with the greatest proportion of unsheltered homeless people are in California, and that Angelenos have voted to raise their own taxes four times in the past decade to house the homeless, which he says "hasn't worked."
• Disconnection, not just addiction or cost, is homelessness's common denominator. A 2023 UC San Francisco study found 90% of homeless people are single adults with no friends or family to ask for help, and 63% cited social issues (conflict, not wanting to be a burden, abuse) for leaving their last housing; separately, 38% of all Americans report estrangement from a family member. Journalist Anand Giridharadas frames homelessness as the most visible symptom of people feeling abandoned by failing systems, a point both left ("not spending enough") and right ("government is incompetent") dodge.
• Cooperation, not strength, is humanity's evolutionary superpower. Scott leans on Survival of the Friendliest by Duke's Brian Hare and Vanessa Woods, who argue early humans selected for prosociality; Hare told the Washington Post that "survival of the fittest" is the most harmful folk theory in society.
• Relationships drive career, health and survival outcomes. He cites one study in which each step up a 0–6 loneliness scale at age 12 raised the odds of being out of work and school at 18 by 28%; people with strong community belonging are 2.6x likelier to report good physical health and 3.2x good mental health; and social isolation raises premature-mortality risk by 29%. His inference: modern exile is self-induced via Instagram, DoorDash, Netflix and, increasingly, AI companions.
• Rebuilding connection must be treated as society-wide infrastructure. He cites a recent Gallup finding that a majority of Americans feel little to no responsibility to help their community, and endorses Surgeon General Vivek Murthy's 2023 report (disconnection equals smoking 15 cigarettes a day): government should run housing, transit and paid-leave policy through a "does this bring people together?" filter, doctors screen for loneliness, and tech optimize for real connection. He pairs it with his own agenda (tax reform, cheaper healthcare, social-media accountability, mandatory national service) and the story of Mariners fan Rhenda Strub, who found a community by inviting strangers to share her season tickets.

2026-10-16 · open the original ↗

• Supply chains are moving from cheapest-source to owning the source. He cites the head of Anglo American saying tech giants show serious interest in buying mining stakes, rather than tripling inventory or relying on government stockpiles. After 80 years of "where's the cheapest one and cheapest shipping," he calls supply chain management a career with 25 years of guaranteed work.
• ASML's machines are a single chokepoint that depends on yttrium. Every advanced chip machine worldwide comes from ASML, and their heated interiors must be coated with the rare earth yttrium, without which he says they are "worthless." He reports the US asked ASML to stop selling and servicing machines in China, about 30% of its sales, and that ASML refused days ago.
• China's export paperwork lets it map and target hidden suppliers. Foreign buyers of Chinese rare earths file forms listing customers and products, giving Beijing maps of who supplies whom. In July 2026, retaliating for EU sanctions on Chinese firms aiding Russia, China cut off 14 EU firms, including a tiny German maker of ASML coatings (about €15M revenue, 36 staff, by his unverified lookup) that he says will close within months.
• One small firm's survival is forcing Europe to soften on Russia. An EU delegation to Beijing got nowhere; the EU has now offered to lift sanctions on one Chinese company if China frees the German firm. Berlin did not know the firm supplied ASML until it called, while China had mapped the link years earlier; officials worry delisting will stop European firms seeking non-Chinese suppliers.
• The US is in Europe's position and faces rare-earth triage. He says China is delivering about 65% of the volume promised under the Busan agreement, citing roughly 2,360 kg per ship and nearly 10,000 lb per submarine. An AI model he queried put the West in Europe's position until 2036–2040 at minimum; he predicts defense will outrank consumer goods, and suggests letting Chinese automakers build in the US to secure supply.

2026-10-16 · open the original ↗

• Proof should now be treated as raw data, not the goal. For most of math's history proof stood in for understanding: a problem counted as solved the moment it was proved. He argues machine proofs turn proof into something like numerical calculation or more digits of pi. The prize belongs to whoever turns a proof's ideas into coherent theory, such as a Navier–Stokes blow-up result, and he calls proof "overvalued for decades".
• Machine theorem-proving barely touches math's value or its funding. He lists three sources of value: applications, transferable problem-solving skill, and aesthetic joy. Applications almost always come from theory building, as with Galois's group theory rather than the quintic result, or the theory grown around Fermat's Last Theorem. So public and private funding stays rational, and theory builders gain leverage. What changes is the culture and how credit is assigned.
• Skill and joy need human engagement, like lifting weights. Machines lifting more than you doesn't make your own gym work pointless, and the same holds for doing proofs yourself. Teaching is math's biggest funding source, and it pays for exactly these individual goods. He argues education is limited by motivation, not by how clear the explanation is, and motivation is social. That makes teaching one of the jobs safest from AI displacement.
• AI's current position is overrated, but its velocity is not. This is his conclusion from interviews with mathematicians that he has recorded since mid-2025. In 2023 DeepMind's Olympiad geometry system looked bespoke and brute-force. Within a couple of years general language models could solve IMO problems by any reasonable standard. Machine proofs of Erdős problems went from dismissed as low-hanging fruit to serious results, though the frontier is "spiky at a fractal level".
• It's a good time to enter because the field must change. His reason isn't new AI tools. Newcomers can shape the field at a pivotal moment while it looks for guidance. As a 2026 PhD student he would try to get visual explanation like 3Blue1Brown's accepted as a thesis. His other advice is to give lots of talks, since oral presentation will increasingly be how understanding is measured. He also says to walk with a fellow student and talk through what each loves about math.

2026-10-16 · open the original ↗

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

2026-10-16 · open the original ↗

• The naked short put is his most successful strategy. Citing 45 years of daily trading, he says nearly two-thirds of his returns came from just three strategies. Short strangles are his favorite, but he calls the out-of-the-money short put his most important, for its probabilities, capital efficiency, pricing skew, ease of execution and adjustability. This is his own track record, not an independently verified result.
• High win probability justifies taking on unlimited risk, in his view. He describes the trade as a pure long-delta position with limited profit and unlimited risk, typically sold around the 25-delta strike, just outside the expected move, with over a 75% chance of success. He frames repeated small wins as a way to learn the law of large numbers and build "a culture of learning to win."
• Sell puts when the stock looks cheap and implied volatility is high. Taking what he calls a contrarian stance, he concedes "cheap" is subjective but says high implied volatility can be measured with IV rank. His checklist is low basis, lots of liquidity and high IVR.

2026-10-16 · open the original ↗

• The cost of shipping oil now matters more than its price. Citing Reuters, he says a single tanker voyage from the US to Japan cost $81 million, which he works out to $40.50 a barrel in freight, or about 96 cents a gallon. On that reading, Brent at $105 a barrel is effectively $145 a barrel for a buyer who has to ship it.
• US pump prices escape most of the freight shock. Reuters estimates freight is adding 80 cents a gallon to gasoline compared with last year. He argues this misses that most US crude, gasoline and diesel moves by pipeline rather than tanker, so the hit lands mainly abroad rather than at American gas stations.
• Asian energy costs are coming back as US imported inflation. Japan, South Korea, India, China, Taiwan and Singapore pay the freight premium on the energy that runs their factories, and he expects it to pass into the cars, electronics, chemicals and consumer goods the US imports. He sees this already in the data: producer prices at 6.3%, which he expects to rise "a lot," and half of the 3.7% consumer inflation figure coming from transportation and warehousing rather than fuel.
• Expensive butane cancels this fall's usual winter-blend relief. Winter gasoline normally carries 10–12% butane against 1–2% in summer, and butane usually costs 80–85 cents a gallon. He says it is now about $1.60 a gallon, up roughly 90%, so he does not expect the usual 25–30 cent a gallon drop in pump prices. He calls this his own scoop.
• A falling crude price will not offset $40 freight. With expensive crude, high refining costs, exploding freight, war-risk insurance and costly butane all hitting at once, he describes the story moving from oil to oil shipping to inflation. If crude falls $10 tomorrow, he argues, shipping still costs $40 a barrel and goods prices stay high.

2026-10-16 · open the original ↗

• The naked short put is his most successful strategy. He sells around the 25 delta (16–25 in his own trading) at 45 DTE, citing over a 75% chance of success and buying power of roughly 20% of the strike, ideally on cheap stocks with high IVR. He admits it can "bite you", so keep size small and never average down; that it builds "a culture of learning to win" is his conviction, not a tested result.
• Credit spreads are his top pick for new premium sellers. Sell just outside the expected move and collect 25–33% of the strike width; buying power is the width minus the credit, and probability of profit is 100% minus the share of width collected. He prefers monthly options to weeklies, though he notes 0DTE SPX traders use them almost by default.
• Index call spreads are rich because downside skew cheapens puts. In SPX, QQQ and SPY he mostly sells out-of-the-money call credit spreads just outside the expected move, collecting about 20–25% of the width. He calls them "trades for grinders" that pay slowly over time rather than overnight.
• Covered calls reduce cost basis; they are not an income strategy. He sells a 30 delta call at 45 DTE against 100 shares, buys it back at 50% of max profit, and rolls, which he says lifts a stock position's probability of profit from about 54% to 64%. Candidates are $10–$100 stocks with decent volatility, and he finds the trade less attractive with markets at all-time highs.
• He splits a portfolio across stocks, spreads and short premium. His rough allocation is 10–20% in stocks, 10–20% in short credit spreads, and up to 40% in naked short strangles or short puts. This is his own allocation, offered as a guide rather than backed by evidence in the video, and it leans heavily toward selling options.