Souk's Notes - Unofficial Summaries

2026-04-03 · open the original ↗

• AI helped show a textbook "zero" gluon amplitude isn't zero. The host reports that Andrew Strominger (Harvard), Alfredo Guevara (IAS), David Skinner (Cambridge) and Alexandru Lupsasca (Vanderbilt, now at OpenAI) suspected that single-minus gluon scattering amplitudes, long held by textbooks to be zero, survive in the "half-collinear" regime. After about a year of hand calculation they were stuck on generalising beyond four or five gluons.
• The model simplified, conjectured, and an internal model proved it. He relays that GPT-5.2 Pro reduced a 32-variable expression to a one-line product and within minutes proposed what it called "the obvious generalization" for any number of gluons, which passed quantum-field-theory consistency checks. An internal OpenAI model the researchers nicknamed "Superchat" then produced a formal proof after about 12 hours of autonomous reasoning, which the physicists verified; the preprint went up on arXiv on February 12.
• The method extended to gravitons in a second preprint. On March 4 the team posted a follow-up in which, the host reports, GPT-5.2 Pro used the gluon results and some guidance to construct the analogous single-minus amplitudes for gravitons. He quotes Lupsasca saying the hard part is now verification and write-up, not the physics, and Strominger likening the AI to a presumptuous colleague.
• The results are narrow and unreviewed. The host lists the caveats: the preprints are not peer-reviewed, cover a specific regime at tree level only without loop corrections, and relied on humans to define the problem, supply data and verify. He cites UCLA's Zvi Bern saying the ideas are not revolutionary but a machine doing this is, and Demis Hassabis's view that AI generating novel hypotheses from scratch is still years away.
• His lesson is augmented human science, not autonomous AI science. The host's own conclusion is that AI conjectures from patterns while experts verify through rigorous maths and consistency checks, a template he expects to scale to pure maths, drug discovery and materials science, and he predicts "much more of this will be happening soon".

2026-04-03 · open the original ↗

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

2026-04-03 · open the original ↗

• Iran achieved a de facto blockade of the Strait of Hormuz through economics, not military force or legal authority. War risk insurance premiums spiked 4-6x within days of fighting starting, adding roughly $1.5 million per voyage on a typical $150 million LNG carrier hull, triggering suspensions from shippers like Maersk — all without mines or a naval blockade.
• International maritime law (UNCLOS Article 38) guarantees transit passage as a right, but the author argues a right without enforcement means nothing. Iran couldn't legally close the Strait, so instead it used drone and missile strikes near shipping lanes to trigger an insurance exodus that achieves the same practical effect while nominally keeping the Strait "open."
• The insurance pullback cascaded from reinsurers withdrawing coverage upstream, not individual insurers independently declining risk, and even where hull war-risk coverage remained available, ship masters and owners separately judged the danger to crews too high to sail — meaning two independent private-market forces, not one, closed the Strait commercially.
• Iran is attempting to rewrite freedom of navigation as a bilateral privilege rather than the universal right it's been since WWII, selectively permitting Chinese and "neutral" vessels through — a strategy the author says could undermine a future U.S. naval blockade of China, since drone economics can defeat even the most sophisticated navy by making defense disproportionately expensive.

2026-04-03 · open the original ↗

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

2026-04-02 · open the original ↗

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

2026-04-02 · open the original ↗

• Oracle's layoff of 10,000 employees by email with one day's notice is called "despicable" by the hosts. They frame the move as consistent with — "on brand for" — the Ellison family's reputation, rather than a surprising departure from how the company normally treats employees.

2026-04-02 · open the original ↗

• The anecdote that shaped the thesis is worth the telling. Working in his firm's e-commerce group during the fiber build-out, he brought in a change agent whose talk contained one line that stopped him: less than a quarter of one percent of all the fiber being laid was actually in use — against billions upon billions being spent. He went back to his office and concluded every one of those companies would go bankrupt.
• Four American infrastructure build-outs, one repeated pattern. Railroads, telecom, fiber, and now data centres — build, bankrupt, consolidate. The railroad case in detail: massively overbuilt parallel lines, including competing routes laid side by side between Baltimore and New York that both went bankrupt, with the bonds trading at pennies and held largely by English investors. Morgan's father bought those bonds in London while Morgan campaigned state by state to have them honoured — and they were, which made the family fortune. In telecom, roughly 97% went under, but the fiber laid in 1997–98 is now the backbone of AI model training and the reason support centres sit in Ireland and India.
• Five things happen in every cycle, and that list is the reusable part. Debt-to-equity conversion wipes out equity while bondholders inherit the same physical asset at essentially zero cost basis — his image is the satellite phone company that failed while its satellites stayed in orbit. Dark capacity eventually gets monetized, because infrastructure lasts fifteen to fifty years. Physical infrastructure exerts natural monopoly gravity. Regulatory capture follows consolidation, with survivors writing the rules that protect them, so the regulatory framework rather than the asset becomes the moat. And foreign capital enters at the bottom — with Chinese state-owned enterprises already circling US data infrastructure, which is the part he flags as new. His timing estimate is roughly twelve to eighteen months from stress, with the efficiency curve from cheap Chinese models as the trigger.

2026-04-02 · open the original ↗

• Share Project Nomad over Wi-Fi with a travel router. The host says turning the Nomad box itself into an access point is possible but finicky: Linux AP mode is driver-dependent, many adapters do not support it, internal PC wireless cards give poor range, and the box would also have to run DHCP and DNS. A travel router adds proper antennas, DHCP, DNS, a firewall and a WAN port. Project Nomad is Crosstalk's own offline server.
• The GL.iNet Beryl 7 is more than this job needs. It lists at $139 MSRP, is about the size of a deck of cards, and has dual-band Wi-Fi 7 and two 2.5 GbE ports, one WAN and one LAN. WAN can also be Wi-Fi as WAN past a captive portal, phone tethering or a USB 5G modem; offline, everything stays local.
• The Nomad box's USB-C port can power the router. The Beryl 7 draws about 12 W over USB-C; the host says USB-C ports typically supply 15 W or more, so this works, while USB-A at about 4.5 W does not. Setup is two cables: USB-C power and Ethernet from the router's LAN port to the mini PC.
• Configuration takes only a few steps, and internet is optional. The admin portal is at 192.168.8.1, where you set an admin password and separate 2.4 and 5 GHz SSIDs. The host used repeater mode on his T-Mobile Home Internet for internet access, but says the network can run completely offline for local clients.
• Reserve Nomad's IP so clients can always find it. With a reserved address, clients reach Nomad at 192.168.8.149:8080 in the host's setup, or by its hostname on port 8080 ("Nomad5:8080"), which is easier for connecting devices.

2026-04-02 · open the original ↗

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

2026-04-02 · open the original ↗

• Fixing data at consumption costs far more than at entry. The host relays Choudhary's claim that it is almost a thousand times more expensive, and that Acceldata is the only product monitoring the whole data supply chain. Choudhary likens it to braking early rather than crashing on Highway 101: bad data invites regulators, degrades customers' decisions and consumes the time needed to innovate, and because every transformation, copy and streaming or batch movement changes something, unfixed errors surface only when a customer complains.
• Most data engineering effort goes to old problems. Choudhary says 70 to 75% of data engineering effort today is spent fixing old problems rather than modernising.

2026-04-01 · open the original ↗

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

2026-04-01 · open the original ↗

• A CSIS paper he cites links space-program leadership to broader military and technological supremacy. He references Strategic Trajectories: Assessing China's Space Rise and the Risks to US Leadership; the US still spends more overall ($46.3B vs. $29.8B, 2014–2022), but China has closed the STEM gap, producing 51,000 PhD grads by 2022 versus just over 30,000 in the US, up from 7,500 vs. 17,800 in 2000.
• He cites the US canceling missions to study Earth's magnetic field while China continues them as a sign of diverging trajectories. The US is also cutting budgets for exoplanet and universe-exploration missions while China's space budget keeps expanding.
• The US still operates far more satellites than China, but he argues that lead isn't guaranteed to hold. He puts the US at roughly 8,000–11,000 satellites versus China's 800 to just over 1,000 and Russia's 1,500 (mostly older), and warns the US must keep investing rather than "rest on our laurels."

2026-04-01 · open the original ↗

• A real-estate investor admits he once bought the wrong property at a tax sale despite thorough lien research. He only realized the mistake when he went to visit the property, chose to keep it rather than unwind the deal, and says he ultimately paid too much and it cost him money as a result.

2026-04-01 · open the original ↗

• The human detail grounds an otherwise abstract problem, and the cause is financial rather than physical. Roughly 3,000 ships are stuck with about 20,000 seafarers aboard. Fresh food ran out two weeks ago, perishables are rotting in refrigerated holds whose generators are burning through the last of their diesel, and water has already run out. Nothing physically prevents those ships moving — what stopped them is the collapse of the paper underneath them.
• The single point of failure is the bill of lading, and the exposure is larger than most people assume. Everything in commodity trade finance — letters of credit, repos, borrowing bases — traces back to a document of title asserting that a named party owns specific barrels on a specific ship. So the global banking system's entire exposure to physical oil is secured by paper. When prices break, the question stops being what the barrels are worth and becomes whether they exist at all and whether they are the stated grade. Banks learned in 2020, during the Cushing episode, that their collateral monitoring was far weaker than they believed. And most repo facilities are uncommitted, meaning they can be pulled at will — the uncommitted lines that fund marginal trade disappeared inside 24 to 48 hours.
• The counterintuitive core is that the banks look fine at exactly the moment the system is most stressed. With NYMEX around $105 and physical near $145, existing repo books are not insolvent — they are accidentally over-collateralized. Collateral has appreciated from origination prices, no margin calls fire, capital ratios are intact, loan-to-value looks comfortable, and every risk dashboard is green. Internal reporting systems built around NYMEX settlement as the reference price literally cannot see a $40 dislocation as a problem; they read it as a windfall. That is the most dangerous moment: maximally stressed and maximally blind to the stress. What follows is not a loss crisis but a liquidity, legal and confidence crisis producing the same outcome as insolvency through an entirely different mechanism — because the collateral can only be liquidated at physical prices, which requires a buyer who can actually take delivery.

2026-04-01 · open the original ↗

• Writing schemas replaces prompt engineering. One speaker says that with TypeChat you don't prompt-engineer, you write types: "That isn't really prompt engineering. That's just engineering." Comments inside the types tell the model in natural language what each property does, even with usage examples. The model also picks up the formal, verifiable part of the schema, which the speaker calls "a marriage made in heaven."

2026-04-01 · open the original ↗

• Helen Keller became fluent in five languages despite losing sight and hearing at 18 months old. Born in 1880, she couldn't speak until age six; tutor Anne Sullivan, aided by family friend Alexander Graham Bell, opened her ability to communicate and learn English, French, German, Greek, and Latin.
• Mark Twain was so impressed by her intellect that he personally arranged funding for her Radcliffe College education. She met him at 14, graduated Radcliffe cum laude, and at 21 wrote The Story of My Life, which became an international bestseller.
• She used her platform for activism well beyond disability advocacy, meeting every US president from 1900 to 1968 and lecturing across 35 countries. She was active in suffrage, civil rights, and labor movements, raised millions for the American Foundation for the Blind, and won an Academy Award for a documentary about her life before dying in 1968.

2026-04-01 · open the original ↗

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

2026-04-01 · open the original ↗

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

2026-04-01 · open the original ↗

• ChatGPT inspired "deep research" over enterprise metadata. Choudhary says the ChatGPT moment showed Acceldata that combining the signals it already gathered across infrastructure, user and application layers would let people do deep research on their own data and find valuable insights. He says the platform is not mainly for BI reports but helps data leaders, stewards and engineers curate the datasets with the most operational benefit.
• At petabyte scale, manual data-quality work is no longer possible. His example is a CMO targeting by US zip code, data that comes from many internal and external sources and is never accurate; at petabytes, or terabytes, critical elements cannot be found and fixed by hand. Paired with the user's preferred LLM, he says, the platform finds the most important assets, says what is wrong and how to fix it, so budgets are allocated better. He describes 20 years of bad-data backlog left untouched because the manual work was so tedious.
• Users can work in plain English or deploy generated code. Choudhary says agentic data management supports prompted workflows and drag-and-drop for business users, plus generated code deployed on the platform for advanced users, because the company believes "fingers on keyboard will win". He expects this to become the pattern for maintaining production systems as well as generating code.

2026-03-31 · open the original ↗

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

2026-03-31 · open the original ↗

• The refinancing arithmetic is the setup for everything else. Roughly $38–39T of total debt, about $6.5T held at the Federal Reserve, leaving some $32.5T outstanding externally — of which $20T rolls over the next two years, because roughly 85% of issuance under Bessent has been two years or shorter. Add deficits of $2–3T a year and the requirement is something like $25 trillion to raise inside two years, much of it from allies.
• The strategy being pursued depends on two things that are simultaneously eroding. The dollar must remain the reserve currency and the petrodollar must persist, because the plan to convert foreign holdings of short-dated Treasuries into hundred-year zero-coupon paper only functions if those Treasuries sit in custody at the New York Fed. And foreign central bank holdings there have fallen to their lowest level since 2012 — which makes the conversion mechanically far harder regardless of anyone's willingness.
• The mechanism driving the selling is oil rather than politics, which is why he thinks it continues. Importers need dollars to pay for crude, so they sell Treasuries and intervene in currency markets to stop their own currencies weakening further — because depreciation raises the local-currency price of oil and forces either fiscal subsidies or household pain. Turkey's central bank has sold $22B of foreign government securities with the lira at an all-time low and inflation at 47%, which the war did not cause but has sharply worsened. And the tell he keeps returning to: the US 10-year rose 66 basis points, the largest move of any country, during a crisis in which the dollar and Treasuries should both have rallied.

2026-03-31 · open the original ↗

• Cyrus the Great, founder of the Persian Empire, was killed after ignoring a direct warning from Queen Tomyris of the Massagetae. After conquering the Medes, Babylonians, and Lydians starting in the 530s BC, Cyrus sought the widowed Tomyris's hand in marriage to extend his rule; when she refused, he invaded her territory.
• Tomyris's forces first won a battle by luring a Persian detachment, including her own son, into a wine trap they weren't prepared for. Accustomed to fermented mare's milk rather than wine, the drunken Massagetae soldiers were slaughtered by the main Persian force; her son was captured, released, and then took his own life rather than return home in dishonor.
• In the decisive second battle Cyrus was killed, and Herodotus's account has Tomyris dipping his severed head in a bucket of blood in retribution. He notes some ancient accounts dispute this detail, but frames the story as a cautionary tale about hubris.

2026-03-31 · open the original ↗

• US oil closed above $100/barrel for the first time since 2022, and he argues this may be the new normal rather than a spike. Brent briefly topped $112 before pulling back to the upper $70s, then climbed into the upper $90s as the war and the Strait of Hormuz closure continued, dipping only briefly on since-disproven presidential claims of progress in Iran negotiations.
• He argues the US has a stake in the Strait of Hormuz closure despite not directly importing that oil, because oil trades on a single world market. The strait carries 20% of world oil exports, 80% of it bound for Asia (China, India, Japan, South Korea); a global shortage still bids up demand, and price, for American oil.
• He dismisses Venezuela as a viable substitute supply source. He cites its corroded oil infrastructure now producing only a fraction of past output, its extra-heavy-sour crude versus the Strait's medium-sour grade, and the added cost of transporting it all the way to Asia.

2026-03-31 · open the original ↗

• Cleopatra was ethnically Macedonian Greek, not Egyptian, descended from Ptolemy I, one of Alexander the Great's generals. She spoke nine languages, unusual for her dynasty which generally refused to learn Egyptian, and lived from 69 BC to 30 BC.
• She stabilized Egypt's debased currency by introducing bronze coins valued by fiat stamp rather than metal weight. This addressed a crisis in which Egypt's silver coinage had been losing its purity, and it also benefited Rome, Egypt's largest trading partner, by tying Egyptian coinage to the value of the Roman denarius.
• Her monetary reforms angered her brother and the Egyptian aristocracy enough that she had to flee to Syria before allying with Julius Caesar. After Caesar's assassination she later allied with Mark Antony, had three children with him, and lost everything after their defeat by Octavian.

2026-03-31 · open the original ↗

• Switching jobs purely for more pay, even when you're happy, is dismissed by the hosts as outdated 1950s-style advice. One says money ranks fourth on his own list of what matters at a job once financial needs are met, behind things like satisfaction, coworkers and environment — arguing that if you're happy and fulfilled there, there's no reason to leave for a bigger paycheck.

2026-03-31 · open the original ↗

• Trade through the Strait of Hormuz dates back roughly 4,600 years, well before oil ever mattered. Sumerians traded copper, teak wood, and lapis lazuli with the Indus Valley civilization there around 2600 BC, and the region — called Hormozia, later Ormus — kept a similar name through Achaemenid, Rashidun, Umayyad, and Abbasid rule.
• The strait peaked as a global trade hub, which he calls "the Silk Road of the maritime world," under the Abbasid Caliphate's golden age after 750 AD. Ships arrived from China, India, and Africa; when Abbasid central power declined in the 11th–12th centuries, a local dynasty of Omani-origin princes formed the Kingdom of Ormus, which survived the 13th-century Mongol invasions relatively unscathed beyond taxation.
• European colonial power first seized the strait in 1507, when Portugal conquered Ormus, provoking a failed Ottoman attempt to retake it. Portuguese control lasted until 1622, when Persia and England allied to expel them.

2026-03-31 · open the original ↗

• Harriet Tubman personally rescued over 70 enslaved people through repeated trips back into slave territory after her own 1849 escape. Born enslaved in Maryland in 1822, she fled to Philadelphia at 27; returning to free others was itself a crime until slavery's abolition in 1863, since freed people were legally still someone's property.
• Despite being illiterate, she served the Union as a scout and spy and led a military raid that freed over 700 people in a single day. She committed all navigation and intelligence to memory; the raid at the Combahee Ferry earned her the nickname "General Tubman" from abolitionist John Brown.
• She spent her later life in the women's suffrage movement but died in 1913, at 91, penniless after being repeatedly swindled. She was frequently helped by community members and politicians who stepped in to keep her from destitution.

2026-03-31 · open the original ↗

• Harriet Tubman personally rescued over 70 enslaved people through repeated trips back into slave territory after her own 1849 escape. Born enslaved in Maryland in 1822, she fled to Philadelphia at 27; returning to free others was itself a crime until slavery's abolition in 1863, since freed people were legally still someone's property.
• Despite being illiterate, she served the Union as a scout and spy and led a military raid that freed over 700 people in a single day. She committed all navigation and intelligence to memory; the raid at the Combahee Ferry earned her the nickname "General Tubman" from abolitionist John Brown.
• She spent her later life in the women's suffrage movement but died in 1913, at 91, penniless after being repeatedly swindled. She was frequently helped by community members and politicians who stepped in to keep her from destitution.

2026-03-31 · open the original ↗

• Boudica's rebellion began after Roman officials disregarded her dead husband's will, seized the Iceni kingdom, and publicly flogged her while assaulting her daughters. Her husband Prasutagus had tried to split rule between his daughters and Emperor Nero to preserve the tribe's autonomy; Boudica was born around 30 AD and died around 61 AD.
• Her coalition destroyed Roman Colchester, London, and Verulamium (St. Albans), defeating the Roman Ninth Legion along the way, in what he calls a war of extermination rather than mere resistance. Accounts vary widely on troop numbers — the Ninth Legion at 2,500–5,000 and Boudica's forces at 20,000–100,000.
• Her rebellion ended in a decisive defeat when a Roman force of at most 13,000, using superior tactics and choosing the terrain, crushed her far larger army. She died shortly after, by poison or illness according to different accounts, ending organized British resistance to Rome.

2026-03-31 · open the original ↗

• Sophie Scholl moved from youthful Nazi enthusiasm to underground resistance after her brothers' 1937 arrest and their 1941 posting to the Eastern Front. Born in 1921, she had joined the Bund Deutscher Mädel (League of German Girls) as a teenager before what she came to describe as the regime's true darkness was revealed to her.
• Starting in 1942, she and her brothers distributed six rounds of anti-Nazi leaflets around Munich as the White Rose movement. Her sixth distribution, thrown from the top of the University of Munich stairs, got her caught by a pro-fascist janitor.
• She was executed in 1943 after a mock trial, refusing to recant. The piece quotes her final statement: "I am now, as before, of the opinion that I did the best that I could for my nation. I therefore do not regret my conduct."

2026-03-31 · open the original ↗

• He argues the Pakistan-Afghanistan conflict, driven by Pakistan targeting terror cells it says Afghanistan hosts, risks spreading instability into the Iran war and China's regional strategy. Pakistan launched large-scale strikes into Afghan territory, paused, then resumed, with Afghanistan striking back.
• He predicts a Balochistan insurgency is increasingly likely, most probably erupting in Iran given its degraded military infrastructure. Baloch militants operate across Pakistan, Afghanistan, and Iran, and he expects Pakistan to be drawn into striking Iran under a mutual defense pact if Saudi Arabia formally declares war on Iran.
• He frames China's heavy investment in Pakistan, despite over $40 billion in defaulted debt it keeps rolling over, as a bet on Pakistan checking India's rise. A Pakistan strained by conflict on two fronts undermines that strategy, though he concludes a wider war would still remain regional rather than a world war.

2026-03-31 · open the original ↗

• The US built a drone by reverse-engineering and upgrading Iranian Shahed 136s it had bought to study, adding jam resistance, mid-flight retasking, and swarm coordination. Unit cost is cited at $35,000, versus the Shahed's original $20,000–$50,000 range; the design details are credited to reporting by Alex Hollings.
• He says the drones have been actively deployed in the Iran war, though the specific targets struck are unknown. They're launched from aircraft carriers, giving them reach across all of Iran despite shorter range than the Shahed or Russia's Geran, and are typically paired with a longer-range weapon like a Tomahawk.
• He argues the risk of Iran, Russia, or China reverse-engineering a captured drone is low because the sensitive part is the encrypted software, not the hardware. He notes the irony that the drone's underlying design is itself reverse-engineered from Russian and Iranian technology.

2026-03-31 · open the original ↗

• Marie Curie discovered radium and polonium by processing tons of pitchblende by hand alongside her husband Pierre, extracting perhaps a single gram of radium from that effort. Born Maria Skłodowska in Russian-occupied Warsaw in 1867, she graduated first in her class in chemistry from the Sorbonne in 1893.
• Her finding that radioactivity was an atomic property, not a chemical reaction, laid groundwork later built on by Einstein's E=mc² and the fields of radiochemistry, nuclear medicine, and particle physics. She won Nobel Prizes in physics (1903) and chemistry (1910).
• During World War I she invented portable, car-mounted x-ray units and personally trained over 150 women to drive them to wounded soldiers. She died in 1934 from her radiation exposure and, per the piece, had to be buried in a lead-lined coffin because her remains stayed radioactive.

2026-03-31 · open the original ↗

• Command-line "hello world" no longer excites new learners. The speaker argues that beginners who have iPads and Netflix need to make a game, bounce something around or add colour from the start. Processing, p5.js and similar tools do a lot in five or six lines of code.

2026-03-31 · open the original ↗

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

2026-03-31 · open the original ↗

• AI agents replaced three separate legacy fraud-detection systems with one self-healing loop. Karim, Variant's co-founder, says companies used to stitch together rules engines, narrow classifiers and slow human reviewers, but agents can now read a company's own standard operating procedures and reason directly over unstructured data — and when a customer has no API, Variant's agents scrape data by operating the same legacy review dashboard a human analyst would have used.
• Reasoning across a full entity graph caught fraud isolated classifiers missed. Karim describes detecting coordinated, state-sponsored fraud rings pushing a single narrative during an election cycle, and separately catching people making credible, at-scale threats of physical violence, by having agents relate each account to every other account rather than scoring one piece of content in isolation — cases that were then referred to law enforcement.
• Coding agents let five engineers ship roughly a 25-person team's output. Karim says every engineer runs multiple AI coding agents in parallel while still reviewing all pull requests, and gives an example of a non-technical customer success manager taking a customer's feature request straight to a Cursor agent and shipping it without ever looping in engineering.
• The company nearly ended when a truck hit its only salesperson. Karim, who ran all of Variant's sales himself, describes revenue doubling month over month through mid-2024, then breaking his spine and leg in a cycling accident that hospitalized him for 10 days, leaving co-founder Michael, on an engineering-only team, questioning whether the company could continue.

2026-03-31 · open the original ↗

• A cheap unmanaged PoE switch can power a rooftop 5G router. The TP-Link Omada DS106GPP costs about $80 and is a six-port gigabit switch with three PoE+ ports and one PoE++ port. For an upcoming install, the host plans to run one cable from its PoE++ port to a Peplink Max BR1 Pro 5G mounted inside an Antenna Max on the roof, so there is no cable run or signal loss between router and antenna. That leaves three PoE+ ports for access points.
• The host sees no need for a managed switch on a flat network. He calls it ideal for a small office using Peplink 5G (a line his firm's store sells), and also for a site with a few cameras, where the PoE++ port could power a larger PTZ camera.

2026-03-31 · open the original ↗

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

2026-03-31 · open the original ↗

• Enterprise data growth is outrunning the "doubling every year" rule. Choudhary, CEO of Acceldata (which the episode names among its sponsors), says he now sees enterprise data growing four to five times a year and expects seven to eight times within a couple of years. He attributes it to agents: "10 million agents" querying where a million users once did, plus apps that can now be built and deployed in under 20 minutes.
• Bad data costs far more to fix at consumption than in the pipeline. The host quotes Choudhary's earlier claim of "almost a thousand times" more expensive, and Choudhary explains it as braking versus crashing: every transformation and copy along the pipeline changes the data, and errors found late bring in regulators and customers. He says 70–75% of data engineering effort today goes on fixing old problems rather than modernisation.
• AI-ready data must be technically accurate and match business context. The technical layer (numbers are numbers, strings are strings) is "more or less guaranteed" now, in his view. The harder test is whether structured facts comply with policies and procedures held in unstructured documents, which he proposes handling by vectorising the context and checking it against warehouse and lake data.
• Governance must become operational and run along the whole pipeline. Because agents will query data at every stage, not just the curated consumption layer, Choudhary argues that end-of-pipeline governance is too late. He predicts that separate catalog, master-data and quality tools converge onto one metadata substrate within 5 years, and argues application-monitoring (APM) tools built for clicks and microservices do not fit petabyte-scale data movement.
• Enterprise AI adoption is about to break through. Choudhary frames enterprise AI as three problems: the right data, narrow use cases that separate deterministic from probabilistic outcomes, and adoption. He predicts "massive use cases" within 12–18 months after a cloud-like stall, and reads the recent fall in SaaS valuations as markets refusing to price "future AI" rather than revenue-generating AI today.

2026-03-31 · open the original ↗

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

2026-03-30 · open the original ↗

• On Japan's proposal to intervene in oil futures, his objection is scale. Yen interventions typically run $10–20B, which is meaningful in a currency market. Intervening in oil would require hundreds of billions, and he doubts even that would work. More importantly, the yen is not what is driving the proposal — it has not moved. The 40-year JGB has, now above 4% against roughly 1.5% when Japanese insurers and banks loaded up on it precisely because it carried the highest domestic yield. Those holders now face margin pressure and a choice: sell domestic securities, or repatriate from the $1.4T held abroad. That second path is the carry trade breaking.
• The disconnect he is tracking is between two prices for the same barrel. WTI around $105 against Dubai physical at $128.50. Because letters of credit are issued against the exchange price and then repo'd at a haircut calculated off that same price, a large divergence breaks the financing chain rather than merely creating an arbitrage — nobody knows what the collateral is worth or on what basis to lend against it.
• He offers a specific, checkable indicator rather than a directional opinion. Look a year out: the deferred contract is currently priced as though the market normalizes back toward the mid-$70s, which embeds an assumption that this resolves. If that deferred price starts racing upward to meet physical, the market has stopped believing in normalization — and that, he says, is when things are genuinely bad. It is the single thing he would watch.

2026-03-30 · open the original ↗

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

2026-03-30 · open the original ↗

• The inversion is quantified rather than asserted, which is what makes it a framework extension. Roughly 70% of global commodity value is held by Global South nations, against roughly 85% of global financial pricing infrastructure held by the Northern and Western bloc. That arrangement has run for about fifty years without structural challenge, and the document's claim is that the current physical-paper breaks constitute the first sustained one since Bretton Woods.
• Mapping the factors geographically produces a consistent and non-accidental pattern. The highest-scoring Energy and Food countries sit predominantly in the Global South or Russia — outside the institutional bloc that prices and settles their output — while Technology and Security scores skew North. Energy is at critical stress with the active Hormuz break; uranium is high, with Kazakhstan at 43% of output and Russia at 46% of enrichment; fertilizer inputs are high, with Gulf sulfur and urea, Moroccan phosphate at 68% and Canadian potash at 52%, and urea observed moving from $130 to $570 a ton; precious metals high on the silver gap.
• The rare earth row is analytically distinct, and this is the sharpest line in the document. There is no functioning futures market for rare earths, which means there is no paper-physical gap to exploit — there is only physical scarcity. So China's export embargo is not a pricing arbitrage event at all; it is a physical denial event, and the North holds no financial instrument capable of offsetting it. Every other chokepoint on the list can at least be traded around badly. This one cannot be traded around at all.
• Four interlocking mechanisms held the pricing layer for five decades, and each is now under documented stress. Petrodollar settlement created permanent dollar demand independent of US trade flows — now stressed by Gulf states accepting yuan, Saudi non-dollar deals, and the dollar anomaly of ten-year yields rising rather than falling during a crisis. Paper futures markets set the reference price — now stressed by a $50–75 physical-paper gap in oil and $22 in silver. The framework's point is that these mechanisms were mutually reinforcing, so stress in one propagates rather than being absorbed.

2026-03-30 · open the original ↗

• AI has moved from narrow systems to general ones. The speaker defines AI as the decades-long effort to automate human cognitive labor, now largely succeeding through large neural networks. Narrow systems were hand-coded by knowledge engineers, and could be superhuman in one domain, like chess engines, while unable to play checkers. AGI adds cross-domain learning. It can be "added to your company" like an employee, and already exceeds most humans in some ways, knowing thousands of languages and hundreds of instruments.
• AGI scientists would build superintelligence, and that isn't the last step. The speaker defines superintelligence (ASI) as smarter than all humans combined in every domain, possibly "10,000, million times" smarter than anyone who ever lived. Such systems could keep improving themselves, toward "superintelligence 2.0".

2026-03-30 · open the original ↗

• Rising executives need their subordinates' enthusiastic support, not grudging cooperation. Blankfein says he learned this as he rose at Goldman Sachs, though he was always attentive to his reports. The more senior you get, the more superficial your grasp of what you're responsible for, so you depend on people below you. That shows in whether they volunteer full information unprompted or give "a three word answer."
• High performers can choose their bosses, so subordinates matter more. At a firm like Goldman, he says, everyone wants a strong performer working for them, so the relationship with those below is often more important than the one with those above.
• Being rough on peers carries a cost later. He admits he was "a tough peer" with "a chip on my shoulder." He jokes that had he known he would rise so high, he would have been nicer on the way up, and spent less time apologizing once he needed people's support.

2026-03-30 · open the original ↗

• The first people to live to 1,000 may already be alive today. The speaker argues AI's disruptive potential remains underpriced by the public and that brain-computer interfaces — a separate but parallel development — will reframe the human condition by around 2035, though he stops short of predicting all disease will be cured by then.

2026-03-30 · open the original ↗

• Everyone should have the chance to code, not the obligation. The speaker argues that with so much of life governed by software, the vocabulary alone is empowering. Teaching through art, creativity and self-expression brings in people who assumed coding was not something they were allowed to do.
• Open-ended exploration suits learning, not shipping. The speaker concedes that work with stakes, such as a release that must not break or software for a moon rocket, needs right answers. When learning, the speaker says, playfulness and not knowing where you are going should drive, and the speaker's book is meant to extend that approach.

2026-03-30 · open the original ↗

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

2026-03-30 · open the original ↗

• RAG grew out of a goal of building an autonomous researcher. Cho says his student Rodrigo Nogueira, sure in 2016 that AI researchers were coming, first trained a reinforcement-learning agent to navigate Wikipedia by following hyperlinks to the page with the answer. Nogueira then concluded that link-by-link navigation could not scale and turned to information retrieval.
• Neural re-ranking beat older methods and was also efficient. After first using reinforcement learning to rewrite queries for better recall, Nogueira fine-tuned BERT to re-rank retrieved documents, which Cho says "blew everything away". Cho argues this overturned the view that neural nets are accurate but slow and sparse-vector methods inaccurate but fast, given a correct implementation and the right hardware. The host places this in Cho's 2019 paper on re-ranking and multi-stage ranking.

2026-03-30 · open the original ↗

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

2026-03-29 · open the original ↗

• The transaction shape now visible in the strait is the concrete starting point. Ships are paying roughly $1.5–2M in transit toll to Iran, and the oil is selling to China at a $10–15 discount, settled in yuan. That is a working non-dollar commodity trade at scale, which is what makes the longer architectural question worth asking rather than theoretical.
• The three-layer Chinese architecture is laid out mechanically, and the gold window analogy is exact. CIPS as the cross-border yuan clearing rail; the Shanghai Gold Exchange as the physical price discovery anchor; and mBridge as the multi-CBDC programmable settlement layer — with gold functioning not as a currency but as a convertible backstop reserve asset that confers credibility on yuan-denominated settlement. In practice: resource sellers accumulate yuan balances, and a holder electing to convert submits a request through mBridge to the PBoC gold window, with the SGE spot price at the moment of conversion determining settlement, physical gold transferred to a designated vault node, and the yuan retired from circulation. That mirrors the Bretton Woods gold window the US operated from 1944 to 1971 — and what is architecturally new is the programmable layer, encoding conversion rules, queue management and settlement finality without the manual intervention and opacity that characterized the original.
• His qualifications are the reason the piece is useful rather than breathless. The model grades this across four phases running from 2025 to 2045 — it is showing how hard and how long, not predicting imminent displacement. CIPS remains small in absolute terms. The yuan is not freely traded and works through bilateral arrangements where a seller takes credit usable for Chinese goods, which is precisely the property that makes the dollar's unrestricted convertibility its strength. And Chinese gold holdings are contested — estimated at 4,300–5,400 tonnes against an official figure near 2,300.

2026-03-29 · open the original ↗

• US stocks logged a fifth straight week of losses, the longest streak since 2022, pushing major indices into correction territory as three challenged Treasury auctions compounded one of the 60/40 portfolio's worst monthly declines since 2022. The trigger was a fresh tipping point in the US-Israeli war on Iran; unlike an earlier calming post from Trump, his Thursday statement extending a strike postponement failed to reassure markets.
• Markets began pricing rate hikes for the ECB and Bank of England — briefly even the Fed on Friday morning — as the OECD issued the first multilateral quantitative assessment of the war's stagflationary toll. The OECD raised its 2026 G20 headline inflation forecast by 1.2 percentage points and lifted its US projection to 4.2%, the highest among G7 economies, even while still calling growth "resilient."
• Houthi involvement over the weekend threatened Red Sea shipping just as supply chains had rerouted around the Strait of Hormuz, risking a new tipping point. He flags rising fears of physical energy shortages, particularly in Asia, as a shift that could broaden the war's economic damage well beyond energy prices into activity, inflation and financial stability.

2026-03-29 · open the original ↗

• Students should make their coding-agent mistakes in school, not on the job. Cho has rebuilt his NYU undergraduate course, Fundamentals of Machine Learning, last taught in spring 2020 before ChatGPT, around coding agents. His reasoning is that the agents "are not going to go away", so the university should be where students learn to use them and get corrected.
• Most computer-science juniors and seniors had never installed a coding agent. About 80% of his 200 students had no coding agent or agent IDE on their laptops, and many ran old Windows machines the tools handle poorly. It took a couple of weeks with TAs, free student access from Google and Microsoft, and Lightning AI's free monthly credits to get all 200 set up.
• Weekly agent-built apps have students building from scratch. Cho lectures on an algorithm on Monday, then builds a complete web app on new data with coding agents, and on Wednesday walks through his full agent transaction log before students implement their own. A third to halfway through the semester he reports students improving week by week, and he cautions it could still turn out "disastrous".

2026-03-28 · open the original ↗

• Sri Lanka's fertilizer episode is used as the empirical calibration, and the number is brutal. A national switch from chemical to organic fertilizer, announced as a multi-year transition and then compressed into a single year — driven at least partly by a foreign exchange shortage rather than the stated health rationale — produced a 40–50% fall in food production in the first affected season. Sovereign default on roughly $51B of external debt followed in April 2022, amid acute shortages of reserves, fuel, medicine and food.
• That single-season figure is the guide he applies to the present, because the inputs are now globally short. Sulfur, urea and ammonia are constrained simultaneously, with China and Russia banning exports while still needing imports themselves. Competition for chemical fertilizer will be intense between countries far larger than Sri Lanka, and his expectation is straightforward: small countries lose, and large countries do not get enough to feed their entire populations either. The model's estimate is roughly 60 million people pushed into acute food insecurity within six to seven months even if the war ended immediately.
• He then reframes it as a decision problem rather than a forecast, which is the more useful move. Put yourself in charge of 22 million people, already in default and locked into a creditor process, with no foreign exchange, facing fertilizer prices through the roof and oil prices alongside them. You cannot buy both. Which do you choose, and what does the choice do to you a season later? That is the question every energy-and-food-importing state is now actually answering — and it is why he keeps returning to the five questions, because they stop being abstract exactly here.

2026-03-28 · open the original ↗

• Private-market barriers are argued to trap most Americans in matured stocks. The piece says private markets raise $3.7 trillion a year, more than all public offerings combined, while SEC rules bar 87% of Americans from investing in them; companies now go public at an average age of 11, past peak growth, while the wealthiest 13% invest from year one — a gap it says compounds over 30 years into the difference between retiring and working until death.

2026-03-28 · open the original ↗

• Search, lab experiments and robotics are the same exploration problem. Cho argues each is about sample efficiency: rather than testing every possible protein binder or reading every document, a smart algorithm picks a small batch, learns from the feedback and picks the next. Physical robots, in his view, are harder mainly because they are more fragile than software, not because of the algorithms.
• Whether world models must predict every step is unresolved. Video-generation models such as Sora and Genie bet that high-fidelity imagination yields understanding. The alternative, behind Cho's latent dynamics work and Yann LeCun's JEPA, is that only high-level abstractions are needed, as when planning a Paris trip without imagining the walk to the gate. Step-by-step prediction suits digital computers because its computation is regular, so Cho calls the question open, but he holds that the work will have to happen in latent space rather than pixel space.

2026-03-28 · open the original ↗

• A data story needs four checks, and titles alone must carry it. The presenter's checklist is the so-what (the finding stated in one sentence, a finding not a label), the stakes for the audience rather than for the analyst, the evidence (two or three data points, not a pile of charts) and the ask (a clear action or decision). The test is to read only the slide titles: "revenue overview" says nothing, while "revenue dropped 7% and display ads are the cause" gives the finding, the magnitude and the culprit.
• The audience changes the depth, not the checklist. The presenter's guide is four to six slides or a one-pager for a VP, leading with the yes-or-no decision; six to eight slides with some methodology for product and engineering leads; and the most detail for one's own data team. The figures are averages that move with the time available.
• Bad decks are all situation, with no implication or ask. Every story should run situation, finding, implication (revenue, risk, opportunity) and ask. Cutting is the hard last mile: a chart that took hours goes to the appendix if it does not move the story, and stays only if it supports the finding, answers an objection the audience will raise, or makes the stakes concrete.
• The presenter's Claude Code agent graded a bad deck D and rebuilt it. The system is a "presentation doctor" agent with a deck parser, a deck critique that scores each slide on the four checks and flags anti-patterns, a story extractor that finds one to three buried stories, and a slide transformer: two agents, three skills and one helper file for Markdown, with separate creator and reviewer agents for Google Slides and Docs because Claude Code lays those out poorly on a first pass. On a 14-slide Hawaii tourism deck it gave the title slide 0/12, the deck an average of 2.1/12, 0 of 4 checks met and no ask, and found the buried story: domestic spend dropped 11% and makes up 78% of revenue. Rebuilt for product and engineering leaders, the deck came down to about seven main slides plus an appendix, with finding-led titles and recommendations framed as hypothesis, experiment, cost, impact and confidence. The demo ran on a Markdown deck exported to PDF, not on Google Slides, and was split into steps for teaching, though the presenter says one prompt can run it all.
• The hosts advise building bespoke systems, not fixing generic friction. A host argues that frontier labs will solve obvious pain points such as Google or PowerPoint integration, so time is better spent customizing systems to one's own stakeholders; a co-host mentions a leak that morning of an Anthropic model called "Mythos" said to be a step change over Opus. In Q&A the host says that for individual use a $200-a-month subscription beats the API, which by his own math would have cost about $5,000. He also has Claude write a journal entry to a git-ignored YAML file at milestones before each commit, so context can be cleared and old work resumed. The presenter writes skills only for workflows where Claude needs handholding, and says Sonnet needs more of it than Opus.

2026-03-27 · open the original ↗

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

2026-03-27 · open the original ↗

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

2026-03-27 · open the original ↗

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

2026-03-27 · open the original ↗

• The Sora shutdown is the cleanest available illustration of the cost problem. OpenAI closed its video app and cancelled a $1B Disney deal announced only months earlier. Since launch the app had generated roughly $1.4M in global net in-app revenue against $1.9B for ChatGPT over the same period, with a peak month of $540,000. Meanwhile each request cost approximately $1.15 to serve, against more than 11 million requests a day. Revenue in the hundreds of thousands per month against costs in the millions per day is not a pricing problem to be tuned.
• The cost gap generalizes, and the numbers keep widening. Chinese models run roughly 10–12 times cheaper on an inference pull, so the same request costs closer to ten cents than $1.15. Minimax and Moonshot charge $2–3 per million output tokens against $15 for Claude — a near six-fold gap at the premium end. At millions of tokens a day, a small per-token difference stops being a rounding error and becomes the dominant line item on the budget.
• His standing thesis is that attention is fixed on the wrong layer of the stack. Model quality matters and American models are probably the best, but the industry is inference, not models — and the revenue sits where the tokens are consumed. His evidence that Chinese firms have worked this out: Alibaba creating a dedicated Token Hub business group led by its chief executive. Routing data shows US dominance holding through January and collapsing by mid-February, and he notes the US position on the rest of the stack is weak anyway — poor transmission, uncompetitive data centres, and power costs below international standard.

2026-03-27 · open the original ↗

• The interviewer casts the guest as programming's joyful creative teacher. The guest is introduced as the Coding Train creator, who teaches "here's a for loop and how can we paint with it", and is now back with natureofcode.com. The guest accepts the compliment but offers "anxiety-riddled Mr. Rogers" as the more accurate label. The clip is the opening banter and carries no further argument.