Souk's Notes - Unofficial Summaries

2026-01-24 · open the original ↗

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

2026-01-23 · open the original ↗

• Section 4 of the 25th Amendment exists but is deliberately built to be nearly unusable. It requires the Vice President and a majority of the Cabinet (or a congressionally-created alternate body) to declare the President unable to serve, followed by a two-thirds vote of both the House and Senate — a higher bar than impeachment, which needs only a simple House majority and two-thirds in the Senate alone.
• The amendment's drafters intentionally left "inability" undefined and distinct from unpopularity. Written after the JFK assassination to solve a continuity-of-government gap dating to the 1841 "Tyler Precedent," Section 4 was meant to address incapacity — a stroke, kidnapping — not disagreement with a President's performance or policy choices, and it has never been invoked or interpreted by the Supreme Court.
• Applying Section 4 to Trump today is legally live but practically near-impossible. It would require Vice President Vance to risk his political future by leading a Cabinet revolt he'd likely lose, or a Congress controlled by Republicans to create a new commission under Section 4's "other body" clause — both scenarios the author calls effectively zero-probability under current political conditions.
• Even the attempt would carry severe market consequences. The author cites the S&P 500's roughly 23% decline across Watergate and the NYSE's early closure after the 1981 Reagan shooting as precedent for how markets react to presidential-transition uncertainty, arguing an actual Section 4 invocation would likely trigger a flash crash, a VIX spike, and a gold rally as markets questioned who held authority.

2026-01-23 · open the original ↗

• Every abstraction you lean on was written by someone. One speaker recalls interviewing a database developer at Microsoft who would generate a unique integer by marking a column as a unique identifier; asked what he would do if the database lacked that feature, he said he had never considered that somebody had to write that code. The speaker's answer: "you can be that guy". The clip matches Eric Lippert's story in the Hanselminutes interview about his data structures book.

2026-01-23 · open the original ↗

• They set out to build a fast-turnaround, fully reusable orbital rocket. The founders say the industry norm is to build and launch a rocket once, lose it in the ocean, and never see it again; they wanted a vehicle that reaches orbit, flies back, and repeats the cycle "again and again," a problem they felt nobody else was attacking with rigor.
• Launching the company felt like an irresponsible, highest-stakes bet. One of the two co-founders calls starting from zero, with no one else in the field beside them, the hardest personal decision they've made, reached only once they concluded the idea "had to be tried" regardless of whether it succeeded.

2026-01-23 · open the original ↗

• Silver hit $100 and he uses it to restate the original recommendation rather than take a victory lap. The relative he mentions bought around $11.50; the call covered gold, silver and the commodity complex broadly, with an explicit instruction to find a vehicle giving direct commodity exposure rather than buying the companies that consume commodities. He passes along the Bank of America view that in 2026 the charts for all commodities look like gold and silver did in 2025, with the caveat that he does not know if it is true.
• Intel down 15% on earnings gets the same treatment: this was never a stock argument. He quotes his own first video on it from two years ago — as a stock there is no reason to own it, none, zero. The reason to own it is that roughly 95 to 97% of high-end chips are made sixty miles off the coast of China, and the morning after Taiwan is taken, Intel is the most valuable company in America while Nvidia and AMD trade below $5. The Davos quote he cites makes the same point independently: the single biggest point of failure in the world economy is that 97% of high-end chips come from Taiwan. On the fundamentals he offers a comparison rather than a model — Intel at $13 billion of quarterly revenue against a $500 billion valuation on a company expected to lose $40 billion on $20 billion of revenue — plus the ramp from 18A to 14A and the big tech packaging commitments already signed.
• He spends the rest correcting stories he thinks are wrong, including ones that flatter his own thesis. China did not just cut anyone off: the export licensing laws went in in 2019 and no silver application has been refused in the years since, so the cutoff story is false. Three weeks ago the internet had JP Morgan bankrupt on $500 billion of losses, along with Bank of America, Goldman and COMEX itself; none of them are. There is real stress, but he refuses the suppression narrative outright — you cannot suppress the price of a global commodity — and locates the paper discount instead in genuine uncertainty about whether COMEX can deliver, which persists until miners and refiners physically add supply on a timeline he says he has no ability to estimate. On Europe he notes that Trump's tariff and Greenland threats did in a week what no one else managed: 80% now want a European armed force, 61% favour federalising against 27% who would dissolve, and the right-wing parties tied to Putin are polling badly for it. And he flags what is buried in the $2.5 billion critical minerals agency bill — the US will set an elevated global price for critical minerals so producers can profit, meaning it pays $110 for what China sells at $30, and subsidises the rest of the world while doing it.

2026-01-23 · open the original ↗

• The FIRE early-retirement movement is "economic delusion". The speaker rejects the idea of retiring within about seven years, arguing that most people who reach the end of a long career find they do not want to stop working. The clip offers no figures, and it cuts off just as it starts giving "the Lossdog take" on how hard success is.

2026-01-23 · open the original ↗

• Let the LLM handle ambiguity, and fetch the facts with code. Hanselman, a type 1 diabetic, asks GitHub Copilot CLI whether 90 days of his work meetings correlate with 90 days of his blood sugar. He calls an LLM "an ambiguity loop": good at the open-ended question, but his calendar and glucose must not be hallucinated, so glucose comes from a database through a Python skill he wrote and meetings, Teams and email through Work IQ.
• He steers the agent mid-run rather than waiting for the answer. While it works he adds context, such as checking who attends the meetings; he calls this steering.
• The agent's verdict: Monday afternoons, not Satya meetings. As he reads it out, the agent found Satya meetings on Thursdays at 1:00 and Thursday the worst glucose weekday, but concluded those meetings were not spiking him; it flagged XLT and pre-Satya prep meetings on Thursday mornings, rated Tuesdays with James best and Amanda neutral, and put the problem at Mondays at 4:00, possibly from skipping lunch, which he says is what he does. He says he did not know the answers beforehand and "could go either way"; he works on Copilot CLI, and jokes that it is the fun of that work that shows.

2026-01-23 · open the original ↗

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

2026-01-23 · open the original ↗

• The host's UCG-Fiber failed in software while still passing traffic. Starting in November 2025 it dropped off the UniFi cloud site manager (unifi.ui.com), and firmware updates, power cycles and toggling cloud access did not fix it. Later the local interface slowed, froze, and restarted UniFi Network on its own. He does not think heat was the cause.
• He links the failure to an early-release review unit. Ubiquiti sent him the original free for review, and it died less than a year in, which he thinks the early-release build may explain. He bought the replacement himself for just over $300 with shipping, choosing the model with no onboard storage because a separate UNVR handles Protect.
• A backup taken from the failing gateway restored nothing useful. It reported success after about 5 minutes but left the default 192.168.1.0/24 network instead of his 192.168.200.0/24, with no networks, Wi-Fi or adopted devices. It also showed an impossible backup date of March 6, 2121.
• The older cloud backup from November 11, 2025 restored correctly. That brought back the right subnet, the WAN IP and all his UniFi devices. The cost is that any changes he made since November must be remembered and re-applied. He then updated UniFi OS and UniFi Network before moving the gateway into the rack.
• A new gateway brings a new MAC address, which may need an ISP modem reboot. Where the WAN uses DHCP, the ISP's lease can be tied to the old MAC. He did not need to reboot his Comcast modem, probably because he has a static IP. He says a like-for-like swap is easy, but moving to a different gateway, such as a rack-mount model, may need extra steps.

2026-01-23 · open the original ↗

• Aim AI at hard ROI metrics, and it exposes worthless work. Dell's global CTO John Roese (episode 953) says that in the fiscal year Dell reported in March, revenue grew $10 billion while absolute cost fell 4%, which he says had not happened in Dell's 41 years. He credits people, process and technology. He cites a consulting report, he thinks McKinsey's, that calls generative AI a "BS work detector": if AI is applied to a task and revenue and cost do not move, the work may not be worth doing. The host frames this against MIT NANDA's 95% failure figure for enterprise AI, which he calls contentious.
• Laughs per minute gives comedy a measurable target. Software engineer and podcast host Joel Beasley (episode 945) says four to five laughs per minute is the standard for a full-time professional comedian. He records every set, about 10 to 11 months into stand-up, and has Gemini count laughs, since at recording it was the only browser chatbot that took video and GPT and Grok did not. He logs results in a spreadsheet, and he also uses a saved writing-coach prompt modeled on how professionals gave feedback at about 10 comedy writers' groups.
• A niche intersection of skills beats application tactics. A Netflix senior data scientist (episode 947) says he got into DoorDash with a "briefcase" document, which took at least a couple of days, setting out the company's pain points and how he would solve them, sent to the hiring manager. The host doubts it would stand out now that he gets 30-page AI-made pitches weekly. The guest says he landed Netflix, after multiple earlier rejections, because his ads and forecasting experience at Spotify matched what Netflix needed when it launched ads.
• Systems that ignore how humans decide will break on edge cases. Sandy Pentland (episode 949) argues today's AI-driven systems could fall apart the way the Soviet Union's did, through bad data, fast change and ignoring human nature. He says people are "social foragers" and cites Daniel Kahneman's estimate that 95% of what we do comes from culture. He says his group has shown polarization comes from a few loud voices, that traders handling $10 million a day do reliably better when they watch other traders, and that equation-followers who missed tail risk "got creamed" in 2008.
• Workslop comes from vague AI mandates and poor work context. Dropbox's VP of Engineering (episode 951) cites almost three-quarters of CEOs feeling competitive pressure to adopt AI and about 55% of employees not knowing how to use it. His remedies are to treat AI output as a first draft, and to build better context than slow, token-hungry MCP connectors provide; Dropbox's Dash pairs a BM25 lexical index with a vector index. He also recommends internal eval sets defining what good looks like, and training led by high performers.

2026-01-23 · open the original ↗

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

2026-01-23 · open the original ↗

• Gold functions as a dollar alternative, not a commodity metal. Tom Lee argues price-to-sales logic fails for gold because total industrial and jewelry demand is about $120 billion against a roughly $30 trillion network value; viewed this way, other assets are effectively debased against gold.
• Most portfolios are underweight gold relative to recommended allocations. He notes Ray Dalio has suggested up to 10% and that he himself has previously floated 15%, while most investors currently hold zero.

2026-01-23 · open the original ↗

• Take-maximal-risk is offered as the best business advice, over caution. One host says never holding back — modeling the acquisitive habits of firms like Google, which buys roughly one company a day knowing most fail — matters more than any single piece of advice; the other credits watching his entrepreneur father repeatedly fail and rebuild, distilling the lesson to sheer resiliency instead.
• Across-the-board "peanut butter" raises are called a lazy shortcut, not fairness. With 55% of Fortune 500 companies planning uniform percentage raises in 2026, hosts say the practice avoids the harder work of individual evaluation; they favor a base raise plus a separate bonus pool sized to performance, while cautioning that nearly every employee believes they personally outperformed.
• Gold near $4,900 is called bad pot odds even by its own bulls. With gold up from roughly $2,000 about a year and a half ago, hosts say the setup now risks roughly 10% downside to capture perhaps 3% more upside toward $5,000, and neither would buy it here, though one holds gold only as a hedge against a separate short-silver position.
• Large banks are called priced for perfection; fintech looks cheaper. Hosts say traditional banks like Goldman Sachs and JPMorgan trade in the "stratosphere" on legacy technology and fee-driven "annuitized returns," while post-selloff fintech and crypto-linked names like Coinbase (down over 50%) and Robinhood (down about 40%) offer better relative pot odds, despite carrying the same disruption risk.

2026-01-23 · open the original ↗

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

2026-01-22 · open the original ↗

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

2026-01-22 · open the original ↗

• "AI" is branding over statistics; the intelligence is human. One speaker relays a joke of unknown origin: the public is told it is artificial intelligence, investors machine learning, developers stochastic programming, and it is actually linear interpolation. The two voices then agree the intelligent part is the people behind it, including the underpaid workers "running the mechanical turks" that train the systems. The clip matches a passage of Hanselman's interview with Eric Lippert, where Lippert tells the joke.

2026-01-22 · open the original ↗

• Outsource the syntax to AI, never the problem solving. Lippert, asked why write a data structures book at "the peak of AI-assisted coding", says turning a real business problem into a clear solution and then into working code is "the whole game"; the syntax step can go to any tool. The book, Fabulous Adventures in Data Structures and Algorithms, was in Manning early access with 7 of 19 chapters out (chapter 7, on category theory, released the day of recording).
• The book is the unusual material school skips, for curious coders. It is not a tutorial and does not start with Hello World: part one gives traditional structures (lists, queues, stacks, combinatorics, trees, memoized immutable quad trees) a twist, part two covers developer-tools algorithms like unification and anti-unification, and part three stochastic programming from Lippert's Facebook work. A coming chapter explains how code formatters decide what is "pretty".
• Stochastic programming is principled Bayesian updating at scale. Lippert's example: from a bag of 1,000 coins with one double-headed, ten heads in a row (odds 1 in 1,024) moves you to roughly a 50/50 belief. At Facebook they built language tools for data scientists to apply that updating across billions of observations: is this new account a real person when 999 in 1,000 are bots, is this the broken router, is this reviewer classifying spam correctly.
• Bot spam persists because the platforms stopped fighting it. Pressed by Hanselman on Instagram crypto-group spam about 40 times a week, Lippert gives four reasons: the problem is hard, the research division getting good results was laid off entirely, attackers keep getting more sophisticated, and site owners "no longer particularly care about the user protection problem". Lippert adds the feed still shows radio-station content farms after more than a thousand blocks, because the models optimize for outcomes other than user enjoyment.
• Some algorithms feel discovered rather than invented. Lippert includes Bill Gosper's Game of Life algorithm, which computes the future of boards too big for memory and gets faster as the board grows, purely because it changed how they think about functional programming. Hanselman ties this to Douglas Crockford saying he discovered JSON rather than invented it.

2026-01-22 · open the original ↗

• Five rate cuts have moved the ten-year the wrong way, which is the fact the whole episode turns on. Short rates came down 1.75%, from 5.25% to 3.75%, trading around 3.64%. The ten-year was about 3.82% when cutting began and is 4.27% now — half a point higher after all of it. And the supply argument cuts against the administration too: Treasury has stopped issuing at the long end, so with supply withdrawn the yield should be compressing, and it is rising anyway. Whatever the theory was, it is not working out that way.
• The maturity wall is why control of the Fed matters so much to them. The debt is heavily front-loaded, with $10 to $11 trillion to roll in the next year, moving from a 0% funds rate at issuance to 3.75% — roughly $375 billion of additional annual interest on a $10 trillion roll. Last week alone, $654 billion was sold across nine auctions, $500 billion of it bills at four to twenty-six weeks, meaning it rolls again by June and July; of the $154 billion in coupons, only $50 billion was the ten-year and the rest were notes of three years or less. Around 83% of the debt now sits inside a two-to-three-year roll, with another $3 to $4 trillion to issue this year. His read is that the administration expected to control the Fed by now and to be rolling at 2% rather than 3.75% — Miran having voted for 1% — and notes Miran is heading back to the Council of Economic Advisers, by his reading with no credibility left on the Board.
• The composition of who holds the debt has quietly changed, and not in a reassuring direction. Foreign central banks hold around $4 trillion; the category labelled foreign private holders is above $9 trillion, and those are hedge funds — roughly $4 trillion domestic and $3 trillion foreign — running basis trades, gilt trades and yen carry trades. Central banks are strong hands; leveraged funds forced to sell on a margin call are the weakest hands there are, and they are now the largest block. He also flags the measurement problem underneath all of it: PCE is reported at 2.8% and he thinks it is at least 3.5 to 3.8%; the BLS head was fired after the numbers came out and the Fed has since questioned the data quality; and GDP has not been published for two months pending a reconfiguration. For someone who has watched these series for fifty years, that is not a good place for a country to be. The closing observation is that Takaichi's administration is using the same language about the BOJ that Washington uses about the Fed — more debt, at lower rates, with restrictive policy rejected — because every indebted government has reached the same conclusion and is crowding into the short end, which is the only end a central bank controls.

2026-01-22 · open the original ↗

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

2026-01-22 · open the original ↗

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

2026-01-22 · open the original ↗

• Diffusion is a general-purpose way to learn any data distribution by training a model to reverse noise, and it now spans far more than images. Visiting partner Francois Chaubard cites DeepMind's Nobel Prize-winning protein-folding work, the "diffusion policy" approach used for robotics and self-driving, and Google's GenCast, which he calls the most accurate weather forecasting system in the world, as the same core mechanism applied to different domains.
• Progress came from making the training objective easier to learn, not from a new core idea. Chaubard traces a line of papers that moved from predicting the noised data itself, to predicting the added error, to predicting a "velocity," each step making the math simpler and driving the Fréchet Inception Distance quality metric down from about 222 in the original 2015 paper to roughly 8-10 in modern models.
• Flow matching, from Meta researcher Yaron Lipman, collapsed diffusion training into a straight line between noise and data. Chaubard demonstrates it as roughly five lines of training-loop code that work unmodified whether the underlying model is generating images, proteins, DNA, or robot trajectories, calling it one of the more profound simplifications in recent machine learning.
• Chaubard's advice splits by role: model-trainers should treat diffusion as a default tool, and everyone else should update on how fast it's improving. He argues anyone training their own models should seriously evaluate diffusion regardless of application, and that non-trainers should expect roughly the same ~1,000x quality jump seen in image generation over five years (Midjourney to Sora/Veo/Flux) to extend into proteins, robotics policies, and self-driving as the technique keeps getting simpler.

2026-01-22 · open the original ↗

• Bigger models are more general, not better at everything. Ozdemir points to the Qwen family, which he believes is Alibaba's, spanning about half a billion to over a trillion parameters, with over a trillion being GPT-5 or Opus-level size and Kimi among open-weight models up there. A big general model is built to serve very different users, so it can do more horizontal things, including ones you do not need.
• Size models in rough tiers by what the task demands. Ozdemir calls under 10 billion parameters small, enough for lookups, RAG or web search and runnable compressed on a laptop CPU, but not long-horizon tasks with several tool calls. He puts 10 to 100 billion as medium, able to handle agents, especially fine-tuned on task data, and 100 billion plus as large, for enterprise-wide multilingual, multi-task use. He frames the choice as an experiment, testing for example half a billion against eight billion parameters.

2026-01-22 · open the original ↗

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

2026-01-22 · open the original ↗

• Social-media spam persists because platforms stopped fighting it. Asked why obvious bots still get through, one speaker, who no longer works at Facebook, gives four reasons: it is a hard problem; the research team getting good results was laid off along with its whole division; attackers are sophisticated and getting more so; and the people running social media sites "no longer particularly care about the user protection problem". The other speaker sums it up as less about math than an ongoing battle they stopped fighting. The clip matches Eric Lippert's answer in his Hanselminutes interview.

2026-01-21 · open the original ↗

• The Greenland "framework" amounts to Trump backing down. He notes it surfaced within minutes of the EU suspending its trade deal with the US. In his reading, a framework just means assigning officials such as JD Vance and Marco Rubio to talk with Europe. He adds that Republicans had threatened impeachment over any move into Greenland, so he says the idea was dead from the start.
• He expects a new distraction to replace Greenland soon. He treats Greenland as one more diversion from the Epstein files and predicts something will "fill that void" any day. His own guess is Iran; he notes others speculate Cuba, and says a major winter storm might buy time. He concedes this is guesswork given how unpredictable Trump is.
• The administration will claim victory for powers it already had. He predicts any agreement will cover things the US was already allowed to do, such as placing military assets on Greenland. He says claiming a win even in defeat is this administration's trademark, because its supporters will believe it.

2026-01-21 · open the original ↗

• A 2% pullback isn't a real buying opportunity, one host argues. He says he didn't buy the dip and still favors "pot odds to the downside," mocking a Bloomberg guest who called the session "the best buying opportunity so far in 2026" after just one down day.
• He wants a genuine washout — at least 5% — before buying. With the S&P near 6,900, he puts that threshold at roughly 6,650, calling anything less, including yesterday's selloff, not a washout at all.

2026-01-21 · open the original ↗

• A hard concurrency problem the AI twice failed may now work. Mark reports that gRPC shared-memory transport, wanted for Dapr's Go sidecar, stalled on concurrency in C++ and again in Go last fall, but a weekend retry, perhaps reflecting a jump in Opus, passes the gRPC tests modified for shared memory. He stresses he has not verified it independently, and both agree it needs real code review before any PR to the gRPC Go project.
• Faster turns produced more polish, not worse code. Mark built ZoomIt's new video trimmer with AI, choosing it as a self-contained test of classic Win32 C++; Sonnet 4.5 in GitHub Copilot produced a working clipper in about 10 prompts, and later a volume control plus jump-to-start and jump-to-end buttons came from one prompt. Dark mode took under half an hour; fractional-DPI support took about a day and some 20 iterations, and dark mode is what bumped ZoomIt to version
• Let the agent see the problem instead of being its eyes. Mark's agent kept declaring DPI bugs "fixed" until he gave it screenshots and had it add debug output. Hanselman's rule: if you catch yourself copy-pasting into an agent, give it a way to see that output itself, such as an MCP server for debug output, Playwright or regular screenshots.
• Human judgment still catches what the agent's confidence hides. Live, a dark-mode registry override had regressed; the agent repeatedly announced it understood the issue before landing a one-line fix, and a test at 125% DPI found newly cut-off text. Mark's collection of agent failures includes a project that did not build after Spec Kit marked tasks done and Playwright tests reporting a broken UI as fine, though he still calls AI a 10x accelerant for him.
• Git worktrees let several agents work in parallel. Hanselman explains each worktree as a separate folder on its own branch that Git tracks, so three agents can work on unrelated parts of ZoomIt and merge at the end; he uses a small tool, Worktrunk, to make switching as easy as changing branches. He also flags best-practice skills, such as React ones, and expects tools to enable them automatically within months.

2026-01-21 · open the original ↗

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

2026-01-21 · open the original ↗

• The historical record says currency moves alone have never broken this trade. Since 1995 the yen has run from 160 to 80, back to 150, down to 80, and back up again, through the whole ZIRP era — and the carry trade never came close to breaking, not once. He walks the chart back to 1960 through the Plaza Accord iterations to make the point stick. So the popular version, in which a yen rally kills the trade, has no precedent behind it. The trade is a multi-leg structure — buy the yen, buy the JGB, borrow against it at 25 basis points against a 1.5% coupon, take the carry overseas into higher-yielding assets — and the question is which leg actually fails.
• A regulatory change has removed the natural buyers of the long end, and that is new. Rules running April 2025 to March 2026 compel Japanese life insurers to match 30- and 40-year policy liabilities against long-duration assets using market-consistent discounting on current curves — sensible in itself — but with duration-risk penalties, solvency hits of 20 to 30% for mismatches, and targets that extend duration from 12-15 years to 20. The effect is that insurers are now saturated: they cannot buy more 40-year JGBs without taking capital charges they will not take. City banks and trust banks are stepping back on the same charts. So a country at 230-240% debt is cutting taxes and running a stimulus while its principal domestic buyers of long debt have been regulated out of the market.
• Who replaced them is the question he says he cannot answer, and he asks the audience directly. Overseas buyers are now 53% of ultra-long JGB issuance, a shift that started around 2019. He wants to know who they are and, more to the point, why — and says outright that he is fascinated and may be missing something. What he is confident about is the break mechanism, and it is a price mechanism rather than a currency one: the 40-year was at par six or seven months ago and is 83.55 now. A bank that borrowed the full value of bonds bought at par gets a margin call, and the only way to meet it is to sell US securities and bring the money home. He opens by setting Greenland aside as irrelevant to any of this, and notes that Takaichi called a snap election promising to abolish food taxes with no offsetting rise — which is what sent the 30-year, 10-year and 40-year yields vertical the day before.

2026-01-21 · open the original ↗

• An agent chooses its path; a workflow has it chosen. Ozdemir defines an agent as an LLM with access to tools in the function-calling sense, such as searching or reading a file. Most production AI deployments are workflows, he says: retrieval, a human-engineered prompt, an LLM call and display, with the LLM never deciding to use a tool. Take away the LLM's choice of next step and you are in workflow territory; he says he renamed his book from something like "Building Applied AI" to "Building Agentic AI" partly to make this distinction honestly.
• Agents' unpredictability grows with the number of tools. Ozdemir says modern deployments often start at five to 10 tools, raising questions of which tool goes first, misuse and inefficient repeated searches. The developer ends up hoping the agent navigates that self-chosen path efficiently and correctly.

2026-01-21 · open the original ↗

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

2026-01-21 · open the original ↗

• Nvidia's valuation gap with Cisco rests on the CUDA ecosystem, not just chip demand. At ~46x P/E backed by record net income, Nvidia is far below Cisco's 200x dot-com peak, and unlike Cisco's unused "dark fiber," Nvidia's hardware is actively used — 4.5 million developers now build on CUDA (up from 1.8 million in 2020), creating switching costs Cisco never had and that AMD and Intel haven't matched.
• Export controls function as a regulatory moat, not just a risk. The U.S. government lets Nvidia sell H200 chips into China in exchange for a 25% cut of that revenue — a state-capitalism arrangement the author has criticized on principle but argues gives Nvidia government-backed skin-in-the-game support against Chinese alternatives like Huawei, which remain several generations behind on frontier AI workloads.
• The author's base case (50% probability) projects a $3.5 trillion market cap by 2028, below today's price. That scenario assumes the AI market reaches $400B with Nvidia holding 60% share and margins compressing to 55%; the bull case (25%) reaches $8T on 70% share and 60% margins, while the bear case (25%) falls to $1.35T if the AI market stalls at $250B and share drops to 50%.
• The author names concrete sell triggers rather than holding unconditionally. He will consider selling if any two or more of the following deteriorate: hyperscaler capex growth decelerates, gross margins fall below 58%, data-center GPU share drops below 80%, China revenue weakens, or AMD/custom chips (Google TPU, Amazon Trainium, Microsoft Maia) show real production traction — and would buy aggressively on a "Liberation Day"-style drop to $80–90/share.

2026-01-21 · open the original ↗

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

2026-01-21 · open the original ↗

• BI tools solve a two-decade-old problem: getting data into the business. Olly Hughes, co-founder and CEO of Count, says organizations now have too much information rather than too little. His hypothesis is that the BI tool is among the most expensive and lowest-ROI software a company buys, and that dashboards bake in a support-function role for the data team.
• Most data-team work only maintains the status quo. Hughes says he interviewed 400 data leaders, who told him about 90% of their teams' work produces no incremental value, though the best teams reach 50/50 or 40/60. He names four "service traps": drowning the business in information, acting as an answering machine (which AI could turn into a "Cambrian chart explosion"), retreating from stakeholders, and building things the business cannot see, such as dbt tweaks or Snowflake cost tuning. His simplest test is whether the exec team can name three examples of the data team directly driving value last quarter.
• The data team's purpose is to visualize the business, not data. Hughes argues the way out is "operational clarity" (making the business look simple, with a high signal-to-noise ratio) and being the organization's problem solvers, working in agile iterations rather than waterfall. A co-host adds that the quality of stakeholders' questions decays when they only ever get charts back; Hughes says purpose is what lets a team say no to low-value questions, and names time to decision as a metric to track.
• The proposed answer is one shared canvas for analysts and stakeholders. Hughes describes Count as an infinite collaborative whiteboard, like Miro or Figma, running on live data. The demo showed a metric tree (revenue broken into MRR, churn and upsell), a metric map of about 60 onboarding-funnel metrics, and SQL cells whose downstream metrics are traced and that can be pushed to dbt or a semantic layer. He says customers also use it as a dbt IDE.
• His agent works on the canvas, where every output stays editable. In the demo, an ambiguous question about Chicago led the agent to break the problem down and report that Chicago ranks seventh in sales but loses money through excessive discounting, mainly in office supplies, while a narrow query ran in parallel. Hughes says AI amplifies whatever paradigm it is given and that teams can work this way without Count; Shane adds his view that nondeterministic AI products make data science more important, not less.

2026-01-21 · open the original ↗

• The bond market, not the Fed, is called the one setting rates. With bond prices sliding to a 113 handle and the 10-year yield near 4.3% even as Trump pushes for faster cuts, hosts argue the Fed always plays catch-up to the bond market rather than the reverse, calling any attempt to predict what moves bonds long-term essentially unknowable.
• Prediction markets are called worse than legal sportsbooks on price. Despite covering far more markets, hosts say prediction-market fees run 1-2% per side, effectively making a binary market roughly 10% wide once costs are included — no better than DraftKings or FanDuel — and argue only a shift to true exchange-based pricing would let them compete on execution quality.
• Naked puts' record is credited to years without a real crash. Hosts call selling naked puts the historically most consistent options strategy, outperforming passive long positions with little pain since 2009 — but they caveat that every sell-off in that stretch, including last April's near-20% drop, snapped back quickly, and the strategy's edge would erode if that pattern breaks.
• A short-silver position was "annihilated" even as scalp trades kept winning. With silver near $94 and the gold-silver ratio near 50, a multi-decade low, one host describes scalping the short side profitably on every trade since Sunday night while his core short-silver, long-gold position took heavy losses — a correct short-term read alongside a badly timed core position.

2026-01-21 · open the original ↗

• Outsource the syntax to AI, never the problem solving. Lippert, asked why write a data structures book at "the peak of AI-assisted coding", says turning a real business problem into a clear solution and then into working code is "the whole game"; the syntax step can go to any tool. The book, Fabulous Adventures in Data Structures and Algorithms, was in Manning early access with 7 of 19 chapters out (chapter 7, on category theory, released the day of recording).
• The book is the unusual material school skips, for curious coders. It is not a tutorial and does not start with Hello World: part one gives traditional structures (lists, queues, stacks, combinatorics, trees, memoized immutable quad trees) a twist, part two covers developer-tools algorithms like unification and anti-unification, and part three stochastic programming from Lippert's Facebook work. A coming chapter explains how code formatters decide what is "pretty".
• Stochastic programming is principled Bayesian updating at scale. Lippert's example: from a bag of 1,000 coins with one double-headed, ten heads in a row (odds 1 in 1,024) moves you to roughly a 50/50 belief. At Facebook they built language tools for data scientists to apply that updating across billions of observations: is this new account a real person when 999 in 1,000 are bots, is this the broken router, is this reviewer classifying spam correctly.
• Bot spam persists because the platforms stopped fighting it. Pressed by Hanselman on Instagram crypto-group spam about 40 times a week, Lippert gives four reasons: the problem is hard, the research division getting good results was laid off entirely, attackers keep getting more sophisticated, and site owners "no longer particularly care about the user protection problem". Lippert adds the feed still shows radio-station content farms after more than a thousand blocks, because the models optimize for outcomes other than user enjoyment.
• Some algorithms feel discovered rather than invented. Lippert includes Bill Gosper's Game of Life algorithm, which computes the future of boards too big for memory and gets faster as the board grows, purely because it changed how they think about functional programming. Hanselman ties this to Douglas Crockford saying he discovered JSON rather than invented it.

2026-01-21 · open the original ↗

• The Articles of Confederation (1777–1789) deliberately built a government too weak to govern. Article 8 barred federal taxation, Article 6 barred a federal standing army, and Article 13 required all 13 states to agree before any amendment could pass — far stricter than today's two-thirds-Senate, three-quarters-of-states threshold.
• Shays' Rebellion in Massachusetts exposed that weakness in practice. Between 1,500 and 4,000 people stormed foreclosure courts and tried to seize an armory over roughly six months starting in 1786, and Congress had neither an army nor the taxing power to respond — a privately paid militia, not a federal one, ultimately put it down.
• That crisis directly pushed the Continental Congress to scrap the Articles. By 1787 Congress moved to replace the Articles of Confederation, resulting in the Constitution that was ratified in 1789.

2026-01-20 · open the original ↗

• His central claim is a stated hypothesis, not a report: the US delayed Iran strikes by two-to-three weeks to let the regime exhaust its own opposition first. He speculates this would let the IRGC kill and demoralize genuine pro-democracy fighters before Washington intervenes, leaving whoever eventually topples the regime too weakened to resist US-favored outcomes.
• He argues the underlying US motive is oil, gas, and pressure on Russia and China, not Iranian democracy. He states plainly that the US doesn't care about freeing the Iranian people, and frames control of Iran's energy resources and leverage over Russia and China as the real objective — a claim of motive he presents as fact rather than inference.
• He draws an explicit historical comparison to the Soviets halting outside Warsaw in 1944. He says Soviet forces stopped at the edge of the city and let German forces crush the Warsaw Uprising before advancing, and offers this as the template for his theory while saying he hopes to be proven wrong.

2026-01-20 · open the original ↗

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

2026-01-20 · open the original ↗

• Count the conditionals to decide between a workflow and an agent. Ozdemir defines an agent as an LLM with tools that chooses its next step, and a workflow as a deterministic code path, which is what most production deployments are. His first question to clients is to describe the existing process, which nine times out of 10 settles it. A plain keyword search is a workflow; a few branches suggest a hybrid; a conditional at every step points to an agent. He adds that almost any agent's work can be done by a workflow, and the question is how big and maintainable that workflow would be.
• Match model type and size to the task, then test. Ozdemir says autoencoding models like BERT cannot generate text but do embeddings and classification almost as well as huge GPT models, at around 100 million parameters. For generative models he uses rough tiers: under 10 billion parameters for retrieval and search, 10 to 100 billion for multi-turn agents (especially fine-tuned), and 100 billion plus for enterprise-wide multilingual use. Context length is not enough either: in his needle-in-a-haystack test models missed his inserted birthday depending on position, and he calls out Grok as the worst.
• Choose evaluation metrics by task and by the cost of errors. Ozdemir splits evaluation into generative (free text, multiple choice) and understanding (embeddings, classification) buckets with separate metrics. He uses precision when false positives are expensive and recall when false negatives are, and says the question comes before the math: what is expensive to you.
• Forced workflow steps patch agents' unreliability. Ozdemir's example is OpenAI's deep research, which he says always asks a follow-up question even when told not to, because a skipped clarification costs more than an annoying one; Cursor's planning mode is the same pattern. In one case study GPT-4.1, told in its system prompt to always look up an Airbnb policy article first, skipped the tool and made up an answer 5% of the time.
• Extra reasoning and compression both trade quality unpredictably. Ozdemir found no obvious correlation between reasoning level and performance on MathQA and Humanity's Last Exam, and elsewhere gains of maybe 1 or 2% accuracy for 2x the cost, which in his consulting almost never pays. He compares a quantized or distilled model to "the cousin" of the original: in his newsletter's quantized-Llama test, it scored a little worse on benchmarks and its top-10 candidate tokens often diverged from the original's.

2026-01-20 · open the original ↗

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

2026-01-20 · open the original ↗

• He expects 2026 to extend the bull market that began in 2022, with resilience more visible. But he flags three transitions the market still has to absorb: a new Fed, a more aggressive White House picking sector and country winners and losers, and lingering doubts over the AI narrative.
• A newly confirmed Fed is typically "tested" by markets, which can itself cause a correction. The identification-and-confirmation process is, in his view, enough on its own to trigger a pullback.
• The White House is set to pick winners and losers across many more sectors and countries in 2026. He contrasts this with 2025, when disruption was concentrated in IT consulting and healthcare, and reads gold's rally as reflecting that broader uncertainty.
• He guesses the combination of these transitions could produce a drawdown that feels like a bear market. This is his own forecast, weighed against offsetting positives he sees, such as a turning ISM and a possible housing recovery as rates are cut.

2026-01-19 · open the original ↗

• The structural change since 2008 is who holds the debt, and it is a move from strong hands to weak. Before the financial crisis the largest block of US Treasuries sat with central banks — unlevered, not trading, no forced sellers. Today foreign central banks hold around $4 trillion while the hedge fund category is above $9 trillion, and the growth traces the Fed balance sheet: from about $3 trillion in 2014 to $4 trillion by 2020, with M2 going from roughly $10 trillion to $22.4 trillion. The expansion supplied the collateral and the funds bought it. The Cayman Islands is the largest single domicile, and the Fed's own count there went from about $423 billion to over $1.8 trillion once it recognised it had been counting wrong.
• The Bank of England is now saying the same thing about gilts, with sharper numbers. Hedge fund gilt positions have reached £100 billion, about $125 to $130 billion, and Bailey has warned of potential severe disruption. Funds are a third of gilt transactions, up from 15%, and — the line he tells listeners to read twice — are leveraging gilts to amplify bets sometimes by as much as 100 times on small differences between current and future bond values. A Liberal Democrat MP on the select committee put it more precisely still: a small group of mainly US-based hedge funds domiciled in the Cayman Islands account for 90% of all net borrowing in that market. The proposed remedy is haircuts on repo, where borrowers currently get 99.99 cents on the dollar; the industry's objection is that tighter rules would raise government borrowing costs, which he reframes as a choice between higher borrowing costs and blowing up the banking system. The deputy governor's response is that the status quo is not an option.
• March 2020 is the precedent, and it shows exactly how the failure propagates. Hedge funds sold over $200 billion of Treasuries within days and reduced total exposure by $430 billion — possible only because leverage means selling some is selling much more — which spiked yields while equities crashed and forced the Fed to buy $1 trillion of government debt to restore order. And that was not even a financial crisis. His conclusion is that the same eleven funds run the same crowded relative-value structure in both markets, funded ultra-short in repo against long assets, violating the first rule of never borrowing short to invest long. So the next shock does not need to start in Treasuries: losses anywhere force deleveraging everywhere, the largest holders of US and UK government debt become sellers simultaneously, and yields go up — before anything happens to the carry trade at all.

2026-01-19 · open the original ↗

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

2026-01-19 · open the original ↗

• Remote coding interviews no longer work, so his company holds them in person. Rajeeva says that from this year it stopped remote coding interviews, because cheating tools are too many and too hard to detect, and hunting for evidence of cheating spoiled the conversation. In person, candidates get a practical problem resembling daily work, such as failing tests and missing implementations, and may Google APIs under a "gentleman's agreement" not to use an LLM. He concedes that, as more code is LLM-generated, this may not scale, even though staff already get Cursor and ChatGPT licenses.

2026-01-19 · open the original ↗

• Gabbard's through-line is anti-war, not pro-Russia, in the host's reading. The host ties it to her service as a US Army medic in Iraq in 2004–2005, and argues her opposition to the military-industrial complex is the goal she pursues "no matter who she has to meet with" — a double-edged sword. The host says openly they are unsure, and a year earlier would have called her an asset.
• Her record is contradictory and her 2017 Assad trip looks shady. The host cites her 2015 push for tighter refugee limits from Syria and Iraq, her doubts that Assad was behind chemical attacks, her use of "radical Islam", and the undisclosed funding of her 2017 Assad meeting with two companions she called peace advocates, which the host says they were not. The host reads the trip as a failed any-means attempt to end the Syrian war.
• Her bad press may stem from crossing party power, the host suggests. As a Democrat she endorsed Bernie Sanders in 2016 and says she was warned Hillary Clinton's allies would destroy her; in 2018 she attacked the president over Saudi Arabia and voted "present" on impeachment, calling it partisan. The backlash link is posed as a possibility, not a finding.
• Her shift dates to the Russia–Ukraine war, the host argues. The host sees that war as another forever war, and the theory is she now echoes Kremlin talking points and bashes Ukraine as long as it keeps US troops out, citing a March 18, 2025 tweet. Two readings of her mid-2025 nuclear-annihilation video before the Iran strikes are offered — protecting Russian interests, or fear of another Middle East war — without choosing.
• Reports say she is excluded from White House inner circles. The host speculates the tension exists because most officials favor expanding the US military machine and she does not, while stressing they are not certain.

2026-01-18 · open the original ↗

• The US dollar still dominates global reserves, but well off its historical peak. It holds 58% of roughly $11.5 trillion in global reserve currency, has held reserve status since 1945, and has ranged between about 41% (early 1990s) and 70% (2000s) since then.
• The euro and yen are the closest challengers, but both have faded from their highs. The euro, launched in 1999, holds 20% of reserves after peaking near 28% in 2009; the yen holds 6% after peaking near 9% around 1995–96 during Japan's post-bubble turmoil.
• China's currency share is far smaller than its economic weight would suggest, by design. Despite being the world's second-largest economy and largest manufacturer, the renminbi makes up only about 2% of global reserves because Beijing restricts convertibility — less than 1% is allowed to leave the country — to keep it cheap and protect exports, per a Visual Capitalist chart cited in the piece.

2026-01-18 · open the original ↗

• Bessent said the US will not outsource its national security, and the rebuttal was immediate and documentary. The response he shows lists the parts sourcing for four US weapons systems, where the semiconductor suppliers are overwhelmingly Chinese — the outsourcing happened, and by this account has been happening since 1971. The practical consequence surfaced the same week: the stated reason for not striking Iran is that the US lacks the munitions and cannot rebuild them fast enough. His reference point is the twelve-day war with Israel, in which eight years of missile production was expended on defence alone.
• He is honest about a claim he can no longer source, and corrects the record around it. He has repeatedly cited a DoD report putting the rare earth buildout at $1.5 to $2 trillion over twenty years; commenters challenged him, he went looking, and he cannot find it. What he can verify is Japan's example: about $1.5 billion invested publicly since 2010, real progress on light rare earths, essentially none on heavy, and still 70% dependent on China after fifteen years. Against that, the US commitment is $2.5 billion for a strategic reserve and a promise of hundreds of billions into mining — which returns him to the distinction he will not let go of: mining is not refining.
• His objection is to the ownership model, not the spending, and MP Materials is the case study. China's method was for the state to declare the buildout necessary and hire companies to execute — the same way it built its grid and its data centres, and the same way it made DeepSeek the national default so every industry's usage fed back into it. The US method is to buy stock and preferred equity: $400 million into MP alongside Apple's $500 million, plus a guaranteed price of $110 for material available from China at $30. Trump has now invoked Section 232 to push for price floors so that US mining is, in the bill's phrase, economically viable. His argument is that if the state is paying anyway, it should contract the work and attach conditions — no dividends, no buybacks — rather than underwrite a margin. Underneath it is the cultural claim he keeps returning to: America treats anything without a profit as having no value, healthcare included, and that premise cannot survive the amount of money this requires. He is careful to note he flagged MP at 11 when it ran to 96 and is now somewhere around 40 or 50, that the guaranteed price was the whole reason for the interest, and that the silver smelter funded at $2 billion takes five to seven years to build. He asks the audience whether the approach is right — as a real question, not a rhetorical one.

2026-01-18 · open the original ↗

• The DOJ's grand jury subpoena of Chair Powell over Fed building-renovation costs triggered a defiant video response calling it political interference, yet markets barely reacted. US stocks hit new record highs and investment-grade credit spreads compressed to their lowest since 2007, with realized volatility in Treasuries and the dollar staying notably low despite the institutional confrontation.
• China posted a $1.2 trillion trade surplus for the year, successfully offsetting a 20% drop in shipments to the US by redirecting exports elsewhere. Commodities otherwise dominated price action, with gold, silver and copper all hitting new record highs while Bitcoin traded near $95,000.
• The week ahead brings a new 10% "Greenland" tariff on eight European nations — Denmark, Norway, Sweden, France, Germany, the UK, the Netherlands and Finland — meant to incentivize a deal for the purchase of Greenland. This lands alongside a possible Fed chair nomination announcement and a heavy Davos speaking schedule themed "The Spirit of Dialogue."

2026-01-18 · open the original ↗

• Well-paid engineers quit when their work lacks a mission. Rajeeva cites developer-productivity studies that put engineers' productive time at maybe 3 or 4 hours a day, and says creative knowledge workers who can finish their assignments in 2 hours start asking what they are doing. His remedy is making R&D believe it is in the business of innovation, free from being dictated by quarterly plans, even in "boring" fields like enterprise content management. He points to engineer-built Acceldata capabilities such as ODP and a pulse monitoring system, and a next platform version due in May. The host adds a happiness-economics lecture from about 15 years ago putting the plateau around $80,000 to $100,000 of US household income.
• Hire senior engineers on whether they can build. Rajeeva believes technical management requires people who have been in the trenches. He tests senior candidates on turning any business problem, such as designing an e-commerce warehouse or logistics system, into something built in Go, Python or Java. For new graduates he bets on potential, partly scores and background and partly interest, and decides after a few months.

2026-01-17 · open the original ↗

• A skill should give the agent tools, not claimed expertise. Hanselman says starting skill files with "you're an expert" is almost gaslighting the agent; he is the expert in his own diabetes. His open-source Nightscout skill is a skill.md instruction manual plus a Python script (cgm.py) and PowerShell scripts, cloned into the skills folder, and he shows it working in GitHub Copilot CLI, a product he works on, with the Claude Opus 4.5 model.
• Code does the math; the LLM handles the ambiguity. Glucose readings, uploaded every 5 minutes from his Dexcom G7 to Nightscout (hosted on T1Pal), are cached in a local SQLite database, and cgm.py's commands do the analysis so the model never estimates blood sugar. He says the data stays on his machine and the scripts can also run without AI, or with a fully local model in LM Studio or Ollama.
• Watch for ad hoc queries and fold them into the tool. Asked for his worst lunch, the agent wrote its own query (over 200 on 10 days; the worst, the day before, rose 200 points in two hours). Hanselman then had it propose generalized commands and added "day" and "worst" to cgm.py, treating the exchange as rubber-ducking.
• Nothing in a diabetes tool should be hard-coded. He rejected the agent's built-in meal-time presets because "everybody's diabetes is different" and people eat at different times in other countries, replacing them with hour-start and hour-end parameters and an instruction for the skill to ask the user when their meals are.
• This is AI-augmented coding, not vibe coding. Hanselman says a professional will spend about half the time designing and talking and half reviewing code; he steers the agent mid-task, approves its tool calls rather than running in "yolo" mode, and checks the diffs once the agent commits and pushes, saying he still needs to go through it all properly. He dates the current wave to October and Opus 4.5.

2026-01-17 · open the original ↗

• He was suspended from X, and the timing lines up with one video. The European piece was shadow banned on TikTok immediately — views on his last three videos fell 75%, and the European video specifically became ineligible for promotion. He appealed, TikTok cleared it, and all three videos immediately picked up 200 to 300 views, which is what convinces him the suppression was real rather than algorithmic noise. The X appeal is pending and he does not expect it to succeed. His observation is that the platforms are owned by people who decide what can be discussed, and his puzzlement is genuine: why bother with someone with an audience his size.
• He works out what backwardation in silver actually means and admits he had been reading it backwards. A futures price normally sits above spot, with the premium covering storage, insurance and physical delivery — and, he argues, an embedded expectation of delivery that is never quite 100%, whether it prices at 98.87 or 93. He had been thinking in terms of premium over spot. Here spot is the premium, and the correct reading is a discount applied to the future because delivery is genuinely uncertain. The corroboration is behavioural: COMEX participants going directly to miners offering 10% over spot on forward contracts, because they need metal the exchange cannot supply. The $10 to $15 paper-physical spread is one historic event; the second, which worries him more, is how long it has persisted.
• How it closes tells you whether the market is broken, and his answer is that it cannot close quickly. If a day's worth of silver appeared and every contract could be filled, the gap would shut immediately and futures would reprice to near-certain delivery. That is not available. High prices pull more mining and refining, but nobody currently holding is selling anything, so supply arrives slowly and the gap narrows over months — six or seven, he guesses, without pretending to know. He then asks the audience to look back at the news cycle they just lived through: JP Morgan, COMEX, Goldman, Bank of America and Deutsche Bank were all declared bankrupt, and days later it emerged that the American banks are long silver. His read is that people saw genuine stress, saw repo activity that was ordinary year-end balance sheet dressing, and built a bailout narrative on top of a facility that would never be used for that purpose. Recognising stress is good; understanding why it is there is the part that matters. He closes by saying he is not sure it is feasible to keep doing this in the current environment, and will let people know.

2026-01-17 · open the original ↗

• A CTO must think in customers' shoes, because engineers won't. Rajeeva, who grew Acceldata's engineering from four to 100 people through COVID, remote work and the return to office, says technologists assume "if I build it, they'll come". Customers act on guidance with two axes, a technology mandate ("we need to do AI") and a business problem, and need wins they can show their leadership to renew a proof of value into a contract. Product managers focus on features and bugs, so if not the CTO, he asks, then who.
• Early-stage engineers stay when they get room to innovate. Rajeeva says money stops mattering to people past a point, and startups lose context and speed when people churn. He keeps them with engineer-led projects rather than a fixed annual roadmap: an internal data platform (ODP) built to avoid license fees and vulnerabilities, an AI agent that fixes vulnerabilities in shipped enterprise software, and an in-house data-center stack because VMware is expensive and OpenStack too complex.

2026-01-17 · open the original ↗

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

2026-01-17 · open the original ↗

• He argues the Abraham Lincoln carrier strike group, four days from Iran, is effectively unthreatened by Iran's arsenal. The group reportedly includes a missile cruiser, four to six destroyers, a Virginia-class attack submarine, a replenishment ship, and 50–60 aircraft; he notes 60% of Iran's missile capacity was already destroyed in the 12-day war, and expects F/A-18 Growler electronic warfare to blunt any drone swarm.
• He cautions the deployment may not signal imminent mass strikes. He points out the US kept two carrier strike groups near Iran during the 12-day war without striking Iranian territory itself, suggesting this could be a continuation of the US's long-standing routine Middle East presence.
• If strikes do come, he expects them aimed at military and leadership targets rather than mass civilian ones. He compares this to the Maduro extraction and notes that while Gulf states are publicly urging restraint, they have quietly stood up a joint air-defense cell with the US in Qatar as a backstop.

2026-01-16 · open the original ↗

• Paying everyone in a role the same ignores real differences in skill. One host argues that because individuals bring different strengths, uniform pay for a given role "makes absolutely no sense," pushing back on the idea that equal pay for equal roles is inherently fair.
• Pay transparency's wage-gap benefit is real but comes with a feared tradeoff. The hosts call the idea that transparency narrows the gap for women and other underpaid groups "a very valid argument," but one worries it won't lift the bottom up — instead compressing pay at the top, meaning higher earners (described as men) end up making less.

2026-01-16 · open the original ↗

• Silver's supply is tightening as China curbs exports of a metal the US now treats as a national-security asset. He notes 70% of mined silver goes to industrial use and is largely unrecyclable, while China has started heavily restricting exports in response to high prices and the US has designated silver a crucial metal used extensively in military technology.
• At $92 an ounce, a sustained silver price would only add a modest bump to inflation — but a rise to $200–300 would hit every American's wallet. He contrasts today's price with the far higher levels "some crazy people have predicted," arguing that scenario, not current prices, is what would meaningfully affect consumers.

2026-01-16 · open the original ↗

• Silver's price surge will not collapse the banking system — at worst, it feeds more inflation. He argues silver is deeply embedded in the economy, powering the grid, refining oil into gasoline, and running high-tech manufacturing thanks to its conductivity, but rejects the idea that this translates into systemic banking risk.

2026-01-16 · open the original ↗

• Repeated attempts to short silver as a "bubble" have been crushed as the price broke through one high after another. Shorts piled in above $40, then again above $50, and again at $70, each time expecting a top; silver instead kept climbing, pushing through $90 to its current $92.15 an ounce.
• An unconfirmed theory holds that major global banks are short silver for hundreds of billions, possibly over a trillion, dollars. He stresses there are no confirmed reports backing this figure and that it remains unverified.

2026-01-16 · open the original ↗

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