Souk's Notes - Unofficial Summaries

2026-03-20 · open the original ↗

• Useful AI work needs no rigorous academic background. Kiely argues this, having entered the AI industry in January 2022, about 10 months before ChatGPT, without knowing what a perceptron or a convolutional network was, and says that has not stopped the work. With the technology this new, Kiely calls not knowing an advantage, because there are no bad habits to break. The clip is from the Philip Kiely episode, answering Hanselman's question about impostor syndrome.

2026-03-20 · open the original ↗

• Six countries offered to help reopen the Strait of Hormuz, tied to Ukraine. The UK, France, Germany, Italy, the Netherlands and Japan cited UNSC Resolution 2817 on threats to shipping, but conditioned help on the US fully arming Ukraine — a proposal he credits to Finland's President Alexander Stubb.
• He frames the deal as a win for everyone except Russia and Iran. He praises Stubb over the leaders of Germany, France and the UK for understanding that the US president responds mainly to transactional offers rather than moral appeals.
• He predicts the deal would shorten the war and stabilize energy prices. Citing figures that 20% of world oil, 27% of seaborne oil trade, 20% of LNG and 89% of Hormuz crude (mostly bound for Asia) pass through the strait, he argues US forces freed from escort duty could concentrate on IRGC targets, and that Ukraine would gain new room to push back against Russia.

2026-03-20 · open the original ↗

• A pancreatic-cancer enzyme became the key to a new manufacturing process. Co-founder Gora, then in medical school studying pancreatic cancer, found tumor cells locally generating roughly 50% hydrogen peroxide concentrations via a mutated enzyme; pairing that enzyme with a metal catalyst — a combination Solugen calls chemoenzymatic processing — pushed reaction yield from a typical 60% to 96% at scale.
• Solugen deliberately avoided the large-plant-first approach that had sunk earlier bio-chemical startups. Co-founder Sean says they built their first reactor for $10,000 out of PVC pipes from Home Depot and started selling small volumes of peroxide to hot tub owners in Dallas almost immediately, arguing capital constraints forced creative thinking and that deep, direct customer knowledge — not a big plant — determined what they could actually build.
• A hyper-targeted, unconventional sales tactic landed Solugen's first oil-and-gas field trial. Sean describes identifying the one person at a saltwater disposal company who controlled chemical purchasing, then buying billboards along his commute route for roughly $10,000-$15,000 so he would recognize the company by the time they called him.
• Solugen's flagship plant is its original prototype, scaled roughly 1,400x. Sean says the 60-foot bubble-column reactor at their Bioforge 1 facility is functionally identical to their original Y Combinator-era PVC reactor, just 10,000 gallons instead of 7, and that one bottle-sized dose of enzyme yields two to four tanker trucks of finished product; the company builds smaller plants near customers rather than one mega-facility to undercut larger competitors on shipping costs.

2026-03-20 · open the original ↗

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

2026-03-20 · open the original ↗

• The new open model pairs large capacity with small active compute. The host reports that NVIDIA's open-weights Nemotron 3 Super, announced around its GTC conference, has 120 billion parameters with only 12 billion active per token through mixture of experts. It interleaves Mamba state-space layers, which make its 1 million token context practical, with a few attention layers kept for precise retrieval.
• Two new techniques buy more experts and faster decoding. NVIDIA's "latent MoE" compresses tokens before routing, which NVIDIA says lets the model consult four times as many experts for the cost of one. Multi-token prediction heads act as a built-in draft model for speculative decoding, giving up to a three times speedup on code and tool calls. The model was pre-trained natively in 4-bit NVFP4, which NVIDIA says runs up to four times faster on Blackwell than FP8 on Hopper, with no accuracy loss.
• The host frames the model as aimed at multi-agent systems' two bottlenecks. These are "context explosion", where multi-agent workflows generate up to 15 times more tokens than chat and drift from their goal, and the "thinking tax" of reasoning at every step. On an 8,000-in, 64,000-out benchmark he reports up to 2.2 times the throughput of GPT-OSS 120B and 7.5 times that of Qwen 3.5 122B, at matching or better accuracy. He also reports that it put NVIDIA's AI-Q agent first on DeepResearch Bench and DeepResearch Bench II.
• The release includes training data and environments, not just weights. The host lists, from NVIDIA, over 10 trillion tokens of pre- and post-training data, 15 reinforcement learning environments, and full evaluation recipes, all under a permissive commercial license. He names early adopters including Perplexity, CodeRabbit, Greptile, Siemens, Palantir and Cadence. Disclosing that he holds a fellowship at Lightning AI, he cites Artificial Analysis data putting Lightning fastest among inference providers at 480 tokens per second at the time of recording.

2026-03-20 · open the original ↗

• The early network's utopian promise has not been borne out. Allman, who started on the ARPANET when 8-bit host addresses were thought enough, says the builders believed universal communication would end wars, and that the future "doesn't look quite as utopian". International reach was assumed from the start, with UUCP carried early to Europe, Japan and Australia; Allman still has a directory listing every user on the early network, himself included.
• The much-criticised configurability was the glue that spread email. Sendmail predates the internet and was built to join Berkeley's local BerkNet, UUCP and the ARPANET, each with its own address syntax. Allman based the configuration file on production systems from an AI course; it has been proven Turing complete, and he argues email uptake would have been much slower without rewriting rules that bridged the separate pools.
• A logging side tool proved the most durable design. Allman wrote syslog as an aid for delivermail, sendmail's precursor, first using MPX files before interprocess communication existed. He says dozens of implementations still share that core.
• AI assistance is useful, but not conscious or fit for vibe coding. Allman, who worked on neural networks for speech recognition when a megaflop was fast, says the field did not see GPT-style models coming. Using Claude at home he values it for surfacing libraries he had never heard of, but has watched it fix one bug and reinsert another, and cannot see vibe coding working. Hanselman's own framing is a "prose compiler" that fills unclear intent with the statistical mean or randomness.
• Retirement still involves coding, now mostly for the house. Allman builds a wine-cellar database with a GUI, home automation in C++ on Arduino and data passed over MQTT, and now writes Python more than C, though not yet typed Python. He retired during COVID from his third UC Berkeley job and names the students as the best part of it.

2026-03-19 · open the original ↗

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

2026-03-19 · open the original ↗

• He argues the US has deliberately withheld a full assault on the IRGC. Decapitation strikes (the supreme leader killed on day one), massive air and missile campaigns, and multi-domain penetration are all active and unrestrained, he says, while intelligence/electronic warfare and special-forces infiltration are active but held back.
• Four major ground-war components remain entirely unactivated, by his count. He lists close air support coordination, amphibious/multi-axis ground assaults, rapid armor maneuver, and urban and key-node seizures as untouched, concluding the "full force" of the US military has not been unleashed in Iran.
• He predicts the IRGC would flee to Iraq if that full force were applied. This is presented as his own opinion; he notes he debated it at length with a guest on another platform, "The Under Report," who disagreed with his assessment.

2026-03-19 · open the original ↗

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

2026-03-19 · open the original ↗

• AI could not write the book; its drafts were unusable. Kiely wanted to go from idea to hardcovers in six months, against a traditional publisher's 12 to 18 months after manuscript, so tried having AI produce the outline and chapters, and calls the quality "completely unusable". Hanselman adds that none of Hanselman's own books, some almost a thousand pages, used AI, because it "doesn't hold a through line".
• AI earned its place on small, verifiable pieces. A six-line CUDA kernel in figure 4.1 saved Kiely 30 minutes. Kiely also used it for a few bad ideas to throw away against the blank page, and for fact and completeness checks, treated as a to-do list to verify by hand rather than as verdicts.
• Scripts beat direct prompting for mechanical book work. The QR codes and layout came from a script a coding assistant wrote, not from generated output. Pasting all of appendix A into ChatGPT to alphabetise it got it wrong. The clip is from the Philip Kiely episode on Kiely's book Inference Engineering.

2026-03-19 · open the original ↗

• Stored memory closes the loop for long-running AI science jobs. Stewart, a partner director of software engineering innovation at Microsoft, reports a science job that ran 14 days. Packaging its memories and rerunning the same sample on a new machine started the research 150 million tokens lighter, since that was what the first run spent building the plan; the memory store returned about 1,200 relevant memories, and the second run got much further, with many thought channels converging on outputs.
• Even partial thoughts are kept, and relevance scoring decides use. Stewart took memories from 33 incomplete channels too and says the engine judged them relevant, which he calls proof there is "no such thing as a wasted token in science research". Signals between engine and store score each memory 1 to 5; 3 and above are grounded on, lower ones stay stored for other problems. Hanselman's framing is a library where nothing is thrown away and only the pertinent books are pulled.
• The engine is modelled on a brain, with fast and slow channels. It orchestrates fine-tuned scientific models, with language models used for orchestration and summarising, splits work into slow-thinking and fast-thinking channels, and falls back to other tools when one, such as the coding agent, is unavailable. Stewart says it builds missing tools through the GitHub Copilot CLI SDK, and that tests with Microsoft's M365 research team on office tasks went well without any warm-start data.
• Grounding in knowledge graphs is what stops scientific fabrication. Stewart says early language models produced SMILES chemical notation that was flatly wrong; one character changes the drug. The answer was Graph RAG, built with Microsoft Research, whose ontology grows with the data, plus a "scientific bookshelf" of each discipline's materials and tools. He credits memory for the system repeatedly catching itself and regrounding in data.
• Scientists stay in charge, and AI remains their tool. Stewart relays scientists' insistence that "AI is a tool, scientists do science", so they can pick memories themselves as well as the engine selecting them. He describes lab-in-the-loop work sending generated protocols to lab robots for synthesis. The product is Microsoft Discovery; his memory research is not yet published.

2026-03-19 · open the original ↗

• He states his data-point discipline explicitly, which is the most transferable thing in the episode. He gives no weight to anything that does not begin "for the first time in ten years" — and the longer the interval, the more attention it earns. Ten years he looks into; fifty years he studies; never-happened-before demands real work. Both the silver and the oil dislocations now qualify as never-before, which is why he is treating them as structural rather than as a trading opportunity. The opening story makes the same point about leverage: his firm's entire energy trading business, hundreds of billions traded and billions in profit, was legally predicated on owning one small well in Appalachia.
• Three underappreciated stress points, each with a different clock. Fertilizer-to-food linkage on a 90-to-120-day fuse, with Egypt, Pakistan and sub-Saharan Africa most exposed on thin buffers and no foreign exchange. Trade finance seizure as the fast-acting one: letters of credit are priced against exchange benchmarks, so with no valid settlement price the banks stop issuing within 24 to 48 hours and seaborne oil and grain stop moving before the exchange even opens. And clearinghouse capacity — the CME guarantee fund at $8–10B against roughly $200B of notional open interest in WTI alone, before counting silver, gold or grains.
• The tether between paper and physical is a single mechanism, and he judges it broken. Futures and spot stay linked only while the arbitrage of delivery remains credible. With silver at a $17 gap and oil diverging by as much as $75, that credibility is gone. The systemic point follows: the exchanges do not merely price commodities, they provide the reference rate for an enormous shadow system of OTC swaps, supply contracts, ETFs and lending agreements — so when the reference breaks, every downstream contract loses its legal basis at the same moment. That simultaneity is the contagion multiplier.

2026-03-19 · open the original ↗

• Saudi Arabia and Gulf states have warned they may retaliate against Iran directly. He reads this as a threat to send assets into Iran, and notes it undoes China's 2023 diplomatic win of reopening Iranian and Saudi embassies to each other.
• He frames the rivalry as structurally irreconcilable, not just situational. Sunni-Shia divergence, Saudi alignment with the US and West, and Iran's proxy funding (including the Houthi movement against the Saudi-backed Yemeni government) sit alongside an energy rivalry where Saudi Arabia pumps roughly 10 million barrels a day against Iran's roughly 4 million, a gap he attributes to sanctions and weaker Iranian infrastructure.
• He argues Iran's strikes on Gulf neighbors, not Saudi Arabia, backfired strategically. Because Iran launched missiles and drones at the UAE, Kuwait and Qatar rather than Saudi Arabia, he says those states are now aligning more closely with Riyadh, and he believes Saudi Arabia and the Gulf states are "biding their time" on retaliation.

2026-03-19 · open the original ↗

• Japan faces three compounding crises pulling in opposite directions — the author's "Tokyo Trap." The yen has fallen roughly 7% over the past year to around 159, JGB yields have pushed past the 2% danger threshold flagged in January, and an Iran-driven oil shock is hitting an economy that imports virtually all its energy — with Japan's 250% debt-to-GDP ratio (versus ~120% for the U.S.) constraining every response.
• Prime Minister Takaichi faces pressure to commit roughly $550 billion to the U.S. in exchange for a tariff cut from 25% to 15%, on top of a domestic $134 billion stimulus package layered onto a $700+ billion budget — commitments the author argues Japan's fiscal position may not actually be able to sustain.
• The author warns Takaichi risks becoming "Liz Truss 2.0" if Trump extracts too much. Raising rates to defend the weakening yen and control energy-driven inflation could make Japan's debt-servicing costs unsustainable given three decades of near-zero rates, potentially triggering the kind of bond-vigilante confidence collapse that ended Truss's UK premiership.
• Any Japanese fiscal crisis would boomerang directly onto U.S. borrowing costs, since Japan holds over $1 trillion in U.S. Treasuries; if forced to sell Treasuries to defend the yen or cover energy costs, the resulting yield increase would raise U.S. mortgage, credit card, and floating-rate private credit costs — meaning what looks like a bilateral negotiation carries direct consequences for American consumers.

2026-03-19 · open the original ↗

• This clip is a direct sign-up pitch for Lossdog, not an argument. It tells viewers a five-question, 30-second signup at lossdog.com with no credit card gets them a position and a referral link.

2026-03-19 · open the original ↗

• Past technology spread slowly because institutions could block it. Kass says Gutenberg's printing press was illegal by order of the church for a long period, so long that Gutenberg died without knowing it would matter, and global literacy lagged until the 19th century. He argues that people today, though "scrambled by distrust", have more intellectual sovereignty than ever, since earlier generations heard a war's story from a single reporter.
• Removing hardship also removed useful friction, a mark of affluence. Kass's "Idiocracy" argument is that convenience lets people choose to stop thinking, just as nobody needs to drive a stick shift. He welcomes the lost hardship but expects society to restore some "virtuous friction", such as requiring children to read well even though they don't have to.

2026-03-19 · open the original ↗

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

2026-03-18 · open the original ↗

• Persian engineering predated and rivaled Rome's, he argues. He cites the qanat underground aqueduct system, built before Roman aqueducts, and the yakhchal — a desert structure that produced ice — alongside a postal system that Greek rivals praised for covering roughly 2,700 km in nine days regardless of weather.
• Military defeat by Greece and Alexander did not end Persian influence. Despite Greek forces defeating a far larger Persian army and Alexander's conquest lasting only about a century, the Persians re-emerged as the Parthians and then the Sassanids, inflicting some of Rome's most humiliating defeats, including against Crassus around 50 BC.
• He argues Persian culture absorbed Islam rather than the reverse. After the Sassanian Empire fell to the Rashidun Caliphate, he says Persia was not quickly Arabized — instead "Islam became Persianized" — and teases a follow-up video on Persia's later re-Persianization through the 19th and 20th centuries.

2026-03-18 · open the original ↗

• He blames NATO leadership, not citizens, for withholding help against Iran. He excludes Eastern Europe, the Baltics and the Nordics, and faults Western European leaders for staying dependent on Russian oil and gas since the 2014 invasion despite years of opportunity to decouple.
• He argues Europe fears sabotage and refugee spillover from joining the war. A second factor he cites is reluctance to absorb terrorist attacks or a refugee crisis on the continent if European states get involved militarily.
• His most compelling reason: European leadership wants to see America fail. He argues this lets Europe either blame the US for any resulting energy crisis while striking new Gulf oil deals (and potentially resuming Russian purchases), or, if the US succeeds, step in afterward to invest in and buy from a rebuilt Iran.

2026-03-18 · open the original ↗

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

2026-03-18 · open the original ↗

• The claim that net-exporter status shields America is already false. He answers a Fox News host who said the Iran war is the allies' problem, not America's, citing diesel averaging over $5 a gallon nationwide, gasoline he expects to hit $4 a gallon that week, record shipping fuel costs, and wholesale inflation at 3.4% that he predicts will exceed 4% on the next report.
• Fertilizer and food prices are next. He says fertilizer is at record prices and nearing shortages while moving on ships paying record fuel costs, and expects crop impacts to send food prices sharply higher.
• US-Israeli escalation invites Iranian strikes on oil infrastructure. He reports the US and Israel struck the world's largest gas field that day, which he reads as deliberate escalation to force reluctant allies in. He expects Iran to hit oil storage, which he calls worse than closing Hormuz, or else production fields, which Iran has said it will do.

2026-03-18 · open the original ↗

• Understanding how the system works matters more than operating it. Answering Hanselman's learn-to-drive-stick metaphor, Kiely admits to not driving at all and to riding Waymos, but says knowing how the system works explains why the Waymo makes a mistake and gives "a little bit more empathy for the machine". The rest of the clip is zombie-apocalypse banter from the end of the Philip Kiely episode.

2026-03-18 · open the original ↗

• The dislocation is now large enough to be unarguable. Crude trading around $150 in Asia against roughly $100 in America — a $50 gap for the identical commodity — and Dubai printing $173 a barrel that morning against about $98 domestically. He is careful that this has never happened before, which under his own rules makes it the thing to study rather than trade.
• The structural explanation for why paper leads is worth understanding precisely. When a gold price is quoted it is derived from the most active futures contract, not from physical bullion. The total metal represented by outstanding contracts exceeds registered deliverable stock by a wide margin — nominally under 5%, in practice under 1% — with the remainder rolled, closed or cash settled. So paper volume, meaning how many contracts trade, rather than physical scarcity, dominates short-run price formation. That works until conversion demand arrives.
• The evidence that this is structural rather than cyclical is accumulating across metals. Silver has run a deficit for five consecutive years totalling 820 million ounces — nearly a full year of mine production — with COMEX registered inventories down almost 70% since 2020 and London vaults down 40%, alongside a 90-minute exchange outage on a sensitive day that fed manipulation speculation. Copper held a spread under 1% for five years and is now shifting, with the line that matters: global price discovery is no longer purely economic, because access, reliability and alignment with policy goals are becoming embedded in the market. The mechanism that should close these gaps is physical metal moving between exchanges — so wherever that movement is constrained, divergence persists, which is why silver's has now lasted months. And rare earths show no divergence at all, for the simplest possible reason: there is no Western paper market for them to diverge from.

2026-03-18 · open the original ↗

• Abundant, cheap intelligence will free individuals as the Renaissance did. Kass, formerly OpenAI's head of go-to-market, wanted to call his book Unmetered Intelligence, the idea that intelligence will feel like a resource. His comparison is the Renaissance, which followed the late Middle Ages, when he says half the population was lost, and which produced the idea of the sovereign individual.
• Falling model costs mean data-centre owners will not control access. Kass argues models are getting cheap as well as good and already run well on local hardware: he runs Kimi K2 on his Mac Studio. He says this improves privacy and security and widens what individuals can do, while conceding that it also empowers bad actors, as Renaissance-era technology made war more catastrophic.

2026-03-18 · open the original ↗

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

2026-03-17 · open the original ↗

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

2026-03-17 · open the original ↗

• The negotiating problem stated honestly: America wants rare earths and has almost nothing China wants. Blackwell chips were the one candidate, but H200s were already offered and China has not ordered — reportedly declining a $10B purchase. Which is why the April talks slipped by a month, with the stated reason being presidential attention consumed by the war. His blunt assessment is that the US does not have a counterweight and is now weaker than before, having spent munitions it cannot rebuild.
• He separates other people's reasoning about Iran from his own position, which is worth noting. The argument he relays is that removing Venezuela — around 9% of Chinese oil — plus Iran takes out nearly all Chinese imports outside Russia, manufacturing the counterweight that rare earths otherwise deny. His own objection was never political: he reads military history, did not like the logistics, and said in advance it would go badly. He does not claim vindication so much as note that the leverage argument has produced the opposite of leverage.
• The agreements are unravelling faster than they were signed, which is his standing point about how long real change takes. Malaysia passed a law declaring its deal null and void — not suspended, not on hold, never happening — and Europe declared its agreement void days later. His original argument about the 90-day tariff deadline was that countries cannot turn that quickly: rerouting Canadian pipelines and railways toward the coast rather than the US is a three-to-five-year project regardless of how many agreements get signed, which is why almost all Canadian economic activity still runs south. Meanwhile the Supreme Court struck down the reciprocal tariffs imposed under emergency powers, and the administration pivoted the same day to Section 301 — the authority that had already nearly broken the global auto industry once.

2026-03-17 · open the original ↗

• Puddle's founder claims the startup can do recursive self-improvement far cheaper than approaches that require retraining a new model from scratch. He says the standard approach costs hundreds of millions of dollars and months of effort per model; pressed that incumbents could still out-release them, he responds that Anthropic, OpenAI and Google are pursuing self-improvement too, but typically still at the expensive full-retrain-per-step level.

2026-03-17 · open the original ↗

• He draws a WWII intelligence parallel to explain apparent US restraint in Iran. Breaking the Enigma code, credited to Alan Turing's team but built on prior work by Polish mathematician Marian Rejewski, forced the Allies to let some German operations succeed rather than act on every decrypted message, to avoid revealing the code was broken.
• His inference: some IRGC "successes" may be deliberately permitted, not real wins. He speculates the US may be allowing certain Iranian operations to proceed so as not to reveal a deeper intelligence advantage it doesn't want exposed.
• He offers no specific examples, framing it as an open question. He explicitly invites viewers to identify what current events in the Iran war might reflect this pattern, rather than naming any himself.

2026-03-17 · open the original ↗

• Small shares of global energy supply carry outsized economic weight. He cites Saudi Arabia at 13–14% of world crude, Iraq (the world's fifth-largest producer) at roughly 5%, and argues any country supplying 1% or more can trigger a "nasty economic domino effect" if that supply drops.
• Qatar's natural gas share made it a partial substitute for Russian gas. He notes Qatar holds about 2% of world crude but 4–5% of natural gas, which he says let it absorb some of Europe's need after what he calls Europe's foolish over-reliance on Russian gas.
• Bahrain matters militarily, not economically. Despite under 0.5% of world oil and under 1% of gas, Bahrain hosts the US Navy's Fifth Fleet headquarters, which he says makes the small Gulf area it patrols full of prime targets for Iran to strike.

2026-03-17 · open the original ↗

• He argues Iran's regional strikes are working despite a losing military position. By striking Iraq, Jordan, Kuwait, Saudi Arabia, Oman, the UAE, Qatar and Bahrain, Iran has forced the US to stretch its missile-defense umbrella, destroying an estimated $2.7 billion in US equipment even with a US interception rate above 90%.
• Specific losses include a one-of-a-kind US radar and two THAAD systems. He cites the $1.1 billion AN/FPS-132 radar in Qatar — capable of tracking hundreds of missiles 5,000 km away, requiring 5–8 years to replace, and the only one of its kind outside the US — plus two THAAD radar systems together worth roughly $1.1 billion.
• His inference: this erodes US credibility with Gulf allies and raises invasion costs. He argues Qatar now has roughly 90 seconds' warning against future strikes, and that a ground invasion would itself hand the IRGC a further win by risking internal US dissent, which he distinguishes from a civil war.

2026-03-17 · open the original ↗

• He infers the carriers are intact from market and political behavior, not direct confirmation. The US operates 2 of its 11 "supercarriers" (out of 26 carriers worldwide) in the Persian Gulf — the Abraham Lincoln and Gerald Ford — and he argues current oil prices and Congress's continued bickering with the president are inconsistent with a carrier being damaged.
• He predicts severe damage to even one carrier would trigger a much larger war. He forecasts over 100,000 US troops entering Iran, a defense budget jump from roughly 3% toward 5–6% of GDP, near-unanimous congressional backing, oil prices reaching hundreds of dollars per barrel, and a sharply weaker dollar.
• He addresses the Gerald Ford fire directly, ruling out major damage. He states it began in a laundry room in a non-combat incident, caused no damage to the propulsion plant, and left the carrier fully operational, while allowing that the cause — incompetence or sabotage — remains unknown to him.

2026-03-17 · open the original ↗

• He argues "10 days of oil left" framing is misleading. The Strategic Petroleum Reserve stood near 400 million barrels before roughly 172 million barrels were released, leaving just over 200 million barrels — but he says the real usable cushion, at a 1–4 million barrel/day drawdown rate, is 50–70 days, not 10.
• He notes the SPR ran below the standard 90-day threshold for about 15 years in the 2000s without crisis, arguing today's alarm is disconnected from that precedent, and points out over half of US oil imports come from Canada rather than the Middle East.
• He traces the SPR's origin to the 1973 OPEC embargo and 1979 Iran revolution, and concludes that while Strait of Hormuz closure and related tensions will raise world oil prices, the US itself is not at risk of running out of oil.

2026-03-17 · open the original ↗

• All-caps prompts cannot guarantee output; deterministic code has to. Hanselman says builders over-pivot on prompts and 20-line markdown files, and that a prompt is an "ambiguity loop" which needs a firewall. In Hanselman's framing, the token mask Kiely had just described is such a firewall: bad results cannot come out whatever the prompt says.
• Split the work between the model and plain code. Kiely says models are poor proofreaders when handed 47,000 words in one chat window. A script that sends the book one page at a time gets much better results, and knowing how the systems work underneath is what lets you build that. The clip is from the Philip Kiely episode.

2026-03-17 · open the original ↗

• The Pentagon ordered a Marine Expeditionary Unit toward Iran, at least 2,500 Marines aboard an assault ship, to support reopening the Strait of Hormuz.
• He argues the scale rules out an imminent ground invasion. Comparing it to roughly 15,000 troops for the 2001 Afghanistan invasion and 130,000+ for the 2003 Iraq invasion (after which Baghdad fell within three weeks), he says 2,500 Marines more plausibly supports embassy defense or covert special-forces operations — a reading he attributes to a former Marine intelligence officer's post — while still calling it a genuine escalation.
• His opinion: any post-invasion government should be built by Iranians, not the US. He contrasts two weeks of bombing and 15,000 destroyed targets with no ground troops so far, and argues the US is effective at toppling governments but poor at building new ones.

2026-03-17 · open the original ↗

• Lifetime pay imbalance, not generational exclusion, is framed as the real problem. One speaker argues CEO-to-worker pay ratios can run 400:1 or 800:1, and that a worker earning between $50,000 and $300,000 can be underpaid by $1 million to $5 million over a lifetime simply from not knowing their market value.
• Lossdog is pitched as an AI platform that tells workers what they're worth. The claim is presented as the product's core function, not as a demonstrated result.

2026-03-17 · open the original ↗

• Powering a Starlink Mini from the UCG-Industrial isn't worth it. The host bought the gateway himself for over $600 with shipping, and says he has left the Ubiquiti creator program, so Ubiquiti no longer previews his videos. He says the feature adds about $85 and a more complicated setup than the Mini's own power cord plus a separate data cable. He sees it paying off only in a small share of deployments, such as a single Ethernet run to a weatherproof rooftop enclosure. He uses separate 50-ft power and data cables in his own RV.
• Only the Starlink Mini works, and it needs a PoE splitter. The Mini is the only dish with separate power and data inputs; the other dishes use a proprietary PoE cable from their own router. The host used a roughly $65 gigabit PoE splitter with 24 V output and a $17, 3-ft, 12-gauge Starlink power cable. He advises against 12 V splitters because the Mini needs 12–56 V and a long cable drops voltage below its minimum. The setup worked, and UniFi showed obstructions, latency and firmware, but his splitter made a crackling noise.
• The hardware is solid and aimed at Peplink's market. The gateway is fanless and ruggedized with swappable end caps for mounting, but it has no ingress-protection rating, so it belongs indoors or in a NEMA enclosure (−30 to 50 °C). Ports: 10G SFP+ WAN, 10G RJ45 WAN with PoE+++, a 2.5G WAN/LAN combo and three 2.5G LAN, with two PoE+ and two PoE+++ ports at up to 90 W. Both the 54 V barrel and the 4-pin DC input remind the host of Peplink. The 3–4 lb power brick is a drawback.
• The host calls the microSD Protect storage and SIM slots strange. UniFi Protect on microSD is fine for one or two cameras but not for mission-critical surveillance, where he wants a real NVR. The SIM slots have no cellular radio and work only with the separately bought $400 UniFi 5G Max Outdoor. He would rather have had antenna connectors on the gateway itself.
• A non-industrial version should have come first. Echoing Jim the IT Guy, the host wants the same gateway in a business or home package at least $100 cheaper. The closest current model, the Dream Router 7, has a 10G SFP+ and a 2.5G WAN port with no PoE on either, and a single 15.4 W PoE LAN port.

2026-03-17 · open the original ↗

• Continuous AI is a third pillar beside continuous integration and deployment. Syme, the designer of F# who now works at GitHub, uses it for automating the softer judgements about software: continuous documentation, triage, code simplification, test improvement and security reporting. He calls it additive, says DevOps' culture of determinism will have to open up, and takes up Hanselman's term "ambiguity loops" for these judgement calls.
• An agentic workflow is markdown intent run under tight guardrails. Front matter lists the tools and what it may read, usually only the repository, and a markdown prompt goes to a coding agent in a read-only sandbox. Writes are limited to "safe outputs", such as commenting only on the issue it was given, and are checked by another agent before being applied. Traditional YAML steps can gather data first, such as listing 500 unlabelled issues for the agent to label, and any engine can be used: Copilot, Claude or Codex.
• A maintenance workflow cleared about half of one neglected repo's issues. Syme's Repo Assist runs four times a day by default, or in bulk with a repeat option, labelling issues, fixing bugs, proposing improvements, updating its own PRs and preparing releases without making them. Syme has used it on seven repositories; on FSharp.Control.AsyncSeq it helped close about half of roughly 50 issues, and overnight it optimised three library functions. Nothing merges without a human, and he says he has "come alive" as a maintainer.
• Agents fail unless the repository is engineered for their success. Syme concedes early versions went into repositories without enough talking to maintainers, who rejected some suggestions. Automatic performance engineering will not work on a C++ library with 30-minute builds and no microbenchmarks, and agents may "blag" results they never properly measured, so he frames each workflow as goals, constraints and guardrails. Hanselman's version is that unbounded ambiguity loops produce slop.
• Developers, not vendors, should choose how much automation they use. This is Syme's reply to the charge of "selling AI". Answering an Instagram comment Hanselman relays, Syme concedes some snake oil in the industry but argues that, as GitHub Actions did for CI/CD, teams should decide how much AI runs in their repositories, with strong guardrails by default. He also wants every university to teach a course on AI-assisted software development.

2026-03-17 · open the original ↗

• AI could not write the book but sped up its production. Kiely, author of Inference Engineering and at Baseten, wanted idea to hardcover in six months against a publisher's 12 to 18 months, and found AI-drafted chapters "completely unusable". The words are his and the diagrams and cover human-made; AI wrote a six-line CUDA kernel, offered bad first ideas, flagged facts to check by hand, and wrote scripts for QR codes, layout and a 20-line Python alphabetiser after ChatGPT botched appendix A.
• The book is a deliberate survey that teaches which questions to ask. Kiely accepts the criticism that it is "a mile wide and an inch deep", chosen so it could be written fast and so any smart reader grasps the vision in the first 20 pages; Baseten's salespeople read it. He says today's models barely know the material because it is so new, and that he has no impostor syndrome about lacking a PhD.
• Inference work moves along, then pushes out, a speed-cost-quality frontier. With the model fixed, speed trades against cost, which Kiely treats as the same thing as throughput, along an efficient frontier; research pushes the frontier out, and swapping in a fine-tuned small model adds quality as a third axis. The gains, he says, are always overtaken by demand, so Baseten spends time making customers pay less and is rewarded with more workloads.
• Reliability is the overlooked fourth requirement. Kiely points to GPUs that are "not a very high nine piece of hardware", cloud-provider failures, quantisation that may cost quality, and new speculation algorithms with niche memory bugs. Research reaches production traffic in weeks, he says, and must be hardened for customers such as medical and financial tools with four or five nines of uptime.
• The inference engine guarantees structure that prompts cannot. Instead of "return JSON or my grandma's going to die", a schema becomes a state machine and logit biasing masks every invalid token among the vocabulary's roughly 100,000 to zero probability; Kiely says this has been standard in production for 18 to 24 months, though content quality is still the prompt's job. KV cache reuse rewards putting shared context first, and Hanselman likens the mask to a firewall around an "ambiguity loop".

2026-03-17 · open the original ↗

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

2026-03-17 · open the original ↗

• Cheap, local "unmetered intelligence" will widen individual agency, as the Renaissance did. Kass, OpenAI's head of go-to-market at ChatGPT's launch, argues that models are getting cheap as well as good, so access will not be controlled by data-centre owners. He cites running Kimi K2 on his own Mac Studio. He concedes the same diffusion empowers bad actors, as the Renaissance also brought more catastrophic war.
• Today's dread comes from unmetered information without the wisdom to process it. Kass blames ad-funded media and an internet that taught us everything at once, leaving young people "suicidally empathetic" and without perspective on how good things are relative to their great-grandparents. The host adds his own figures: the same AI capability costs about 1/200 of what it did two years ago, and the length of task AI can handle doubles every seven months.
• AI is splitting achievement into a K-curve driven by agency, not wealth. Kass says Gen Z appears to be the first generation in many not to be smarter than the last, yet will have nearly a standard deviation more geniuses, with the developing world trending up. He dates the decline to screen addiction from 2012, not AI. He wants schools to teach the why and how rather than the what, citing Success Academy's accountability model and Alpha School's two hours a day of self-guided screen learning, which he notes has higher attrition.
• When machines out-think experts, human qualities become the product. His father, an oncologist for 40 years during which breast cancer survival rose from 35% to 90%, was praised at an award ceremony by a patient who got the same prognosis from three oncologists and said "bedside manner is no longer a feature, it's the product". Kass sees the same shift in wealth managers and real estate agents, now chosen for accessibility and good company rather than alpha. His four principles are go outside, learn how to learn, be human, and lead with optimism.
• The chatbot breakthrough was application, not research. Kass says ChatGPT ran on GPT-3.5 research that was already 7 months old, after GPT-3 had gone largely unused. He argues AI is infrastructure like electricity, and the chat app is one light bulb it powers.

2026-03-17 · open the original ↗

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

2026-03-16 · open the original ↗

• AI could not write the book but sped up its production. Kiely, author of Inference Engineering and at Baseten, wanted idea to hardcover in six months against a publisher's 12 to 18 months, and found AI-drafted chapters "completely unusable". The words are his and the diagrams and cover human-made; AI wrote a six-line CUDA kernel, offered bad first ideas, flagged facts to check by hand, and wrote scripts for QR codes, layout and a 20-line Python alphabetiser after ChatGPT botched appendix A.
• The book is a deliberate survey that teaches which questions to ask. Kiely accepts the criticism that it is "a mile wide and an inch deep", chosen so it could be written fast and so any smart reader grasps the vision in the first 20 pages; Baseten's salespeople read it. He says today's models barely know the material because it is so new, and that he has no impostor syndrome about lacking a PhD.
• Inference work moves along, then pushes out, a speed-cost-quality frontier. With the model fixed, speed trades against cost, which Kiely treats as the same thing as throughput, along an efficient frontier; research pushes the frontier out, and swapping in a fine-tuned small model adds quality as a third axis. The gains, he says, are always overtaken by demand, so Baseten spends time making customers pay less and is rewarded with more workloads.
• Reliability is the overlooked fourth requirement. Kiely points to GPUs that are "not a very high nine piece of hardware", cloud-provider failures, quantisation that may cost quality, and new speculation algorithms with niche memory bugs. Research reaches production traffic in weeks, he says, and must be hardened for customers such as medical and financial tools with four or five nines of uptime.
• The inference engine guarantees structure that prompts cannot. Instead of "return JSON or my grandma's going to die", a schema becomes a state machine and logit biasing masks every invalid token among the vocabulary's roughly 100,000 to zero probability; Kiely says this has been standard in production for 18 to 24 months, though content quality is still the prompt's job. KV cache reuse rewards putting shared context first, and Hanselman likens the mask to a firewall around an "ambiguity loop".

2026-03-16 · open the original ↗

• The president announced a bombing operation targeting Iran's Kharg Island, which he says handles roughly 90% of Iran's oil exports (1.5–2+ million barrels/day, storage for up to 30 million barrels, and capacity to load 10 million barrels a day) — built originally by Amoco before Iran seized it after the 1979 revolution.
• The president conditioned the strike on Iranian interference with the Strait of Hormuz, stating any disruption of shipping there would trigger the Kharg Island bombing, which the creator calls a "near nuclear option" economically.
• He predicts the threat would pressure Russia, China and the EU alike to push Iran to stop. He argues higher oil prices from such a strike would benefit Russia, hurt China and the EU, and that outside powers — not the US — would face pressure to rein in Iran, since he says the US itself cannot be pressured to stop.

2026-03-16 · open the original ↗

• A user testimonial presents Lossdog as a more precise alternative to Glassdoor's salary ranges. Describing early access, she says the platform takes in education, career history and a resume to generate a specific pay number for a given role, plus a "target number" showing what she could earn after making the platform's suggested improvements, which she says she now plans to use in negotiations. Signup is framed as joining a waitlist ahead of launch at lostdog.com.

2026-03-16 · open the original ↗

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

2026-03-16 · open the original ↗

• Token export is the concept, and the worked example makes it concrete. Chinese AI companies convert cheap electricity into inference tokens and sell them globally through API platforms, bypassing every conventional trade barrier. In practice: a developer in San Francisco selects a Chinese model as her inference backend and sends a prompt; the request crosses the Pacific by undersea fibre, is processed by a GPU cluster drawing on the Chinese electrical grid, and returns in roughly a second. She pays $0.30 per million input tokens where Claude Opus would have cost $5 — with a quality difference of less than one percentage point on the tracked software engineering evaluation.
• The reason existing policy cannot see this is structural rather than a reporting failure. The compute happens on Chinese soil and the electricity never leaves China. The value of that electricity is converted into inference tokens which cross the border without passing through customs, without incurring tariffs, and without appearing in any conventional trade statistic. It is a form of cross-border service trade that the existing policy framework was never designed to track — so there is no lever positioned where the flow actually is.
• The structural contrast he draws is about who builds what. In the United States the model companies are being permitted to build out an entire country's power infrastructure in order to run their own models. In China the state builds the power and the delivery infrastructure and the model companies concentrate on models. He notes American labs responding by tightening terms of service to compel use of their own inference — and that this will not work, because the migration is not ideological. Developers follow the price.

2026-03-16 · open the original ↗

• The price of identical server hardware has soared in months. Crosstalk bought a customer a 1U Supermicro server with ECC RAM from Thinkmate in October 2025 for $5,800. When the customer asked for a matching unit for high availability, the new quote was over $20,000 for the same hardware about six months later, almost a fourfold increase. The host gives no cause for the increase.
• The rise is pushing Crosstalk to rethink the customer's redundancy. The host says they should have bought the second server when the project was built. Instead of the identical HA unit, they will propose another way to give the customer redundancy.

2026-03-16 · open the original ↗

• He predicts NATO joining the war would produce a short conflict, not World War III. His hypothesis rests on oil prices — US crude hit over $115/barrel before easing to the $90s — combined with a second Iranian ballistic missile incursion into Turkish airspace, which he says gives NATO countries a pretext.
• He frames Turkey's Article 5 decision, resting with Erdoğan, as the pivotal variable. Citing Turkey's large standing army and drone production, he reads Erdoğan's public statement — emphasizing "good neighborly relations" while warning of decisive measures against threats to Turkish territory — as deliberate diplomatic positioning to justify entering later if needed.
• He argues Iran's Shahed drones, aimed at neighbors, are its main remaining leverage. With Iran's air force, navy and most ballistic missile capacity already degraded by the US, he says a multi-country NATO carrier presence in the Gulf would neutralize the drone threat, and closes by questioning whether Iran actually launched the missiles that entered Turkish airspace.

2026-03-16 · open the original ↗

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

2026-03-16 · open the original ↗

• Emergent pivoted twice before finding its market, chasing verification, then non-technical builders. Co-founder Mukund Jha says the company started in late 2023 automating software testing, then realized solving verification lets a coding agent run autonomously for much longer, which led them to build a general coding agent that hit #1 on the SWE-bench benchmark within two months — before they abandoned an enterprise push and repositioned for non-technical users, who now make up 80% of the platform's builders across 7 million apps built in 8 months.
• Emergent built its own cloud infrastructure so build and deploy environments match. Jha says they run their own Kubernetes-based sandbox stack rather than a third-party provider specifically to avoid deployment failures, pairs a multi-agent architecture that delegates tasks like testing or API integration to sub-agents, and gives agents a long-term memory built from auto-generated "skills" distilled from past session trajectories — which he says must be generated by a separate CI/CD-style process rather than by the agent itself to actually improve performance.
• Emergent dogfoods its own product, including an agent-built Asana clone. Jha says a QA engineer's side project of prompting "clone Jira" grew into a project-management tool now used company-wide, built entirely on Emergent with no hand-edited code, saving roughly $3,000-$4,000 a month in subscription costs, and that every employee — including non-engineers — does customer support shifts to stay close to user needs.
• Non-technical domain experts are shipping real, previously-impossible businesses. Jha cites a clinical psychologist and equestrian coach in Alaska who built "EquiMind," an app combining both fields, after a Nova Scotia dev shop quoted her a price she couldn't afford; he frames custom software that used to cost around $500,000 as now buildable for roughly $5,000, and mentions one app built on Emergent has reportedly raised $4 million in funding.

2026-03-16 · open the original ↗

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

2026-03-16 · open the original ↗

• The LPU deal answers the GPU's own memory-bandwidth limit. Fregly, an investor in Groq, which the host says NVIDIA acquired, describes the deal as a "weird acquisition, licensing deal". Groq's LPU keeps memory on the same chip as the accelerator, which suits the bandwidth-bound decode step of inference. He says NVIDIA's bandwidth limitation dates back to its graphics architecture, and the deal spared it a five- or six-year chip cycle.
• The returns are in secondary markets, backing founders who can raise money. Fregly says he holds secondary shares in private firms such as xAI, which the host says SpaceX had just acquired at the time of recording, because the IPO dry spell leaves public stocks like Apple "behind the times". He names Databricks as his best investment so far and judges founders chiefly on their ability to fund-raise.

2026-03-15 · open the original ↗

• Helium is a tighter chokepoint than the commodities that get all the attention, and he says so directly. The United States is the dominant producer at 42% of output with Qatar second — roughly 75% between them, more concentrated than rare earths, lithium and copper combined on his reading — and Qatar is currently offline. The second tier is Russia, ramping its Amur plant toward 60 million cubic metres a year, alongside Algeria, Australia, Poland, Canada and China. Uses run across semiconductors, MRI machines, cryogenics, leak detection, aerospace and rocketry, welding, cooling infrared missile sensors, submarine ballast systems and satellite instruments.
• The distinction that actually matters is between the two isotopes, and almost nobody makes it. Helium-4 is the common form every producer above extracts. Helium-3 is an extremely rare isotope that serves as the working fluid in the inner mixing chamber of dilution refrigerators — the thing that reaches the millikelvin temperatures quantum processors genuinely require. It exists only as a byproduct of tritium decay from nuclear weapons programmes, which makes the US and Russia the only meaningful terrestrial suppliers, at roughly $20 million per kilogram. Countries without nuclear programmes have no domestic pathway at all, which is why lunar extraction is discussed seriously rather than dismissed as science fiction.
• The payload is the answer to a viewer's question about sequencing, and it reframes the last three years. Someone asked why China did not deploy rare earths as leverage earlier. The answer he found: around 2022 Beijing recognized it could not credibly restrict rare earth exports while remaining roughly 95% dependent on US-origin helium. So it resolved the helium exposure first. Capability precedes coercion — which is the general lesson, and the reason a country's apparent restraint may simply mean it has not finished preparing.

2026-03-15 · open the original ↗

• The US attacked and "totally obliterated" military installations on Kharg Island, Iran's primary oil export hub, prompting an Iranian warning that regional oil and energy facilities would be "reduced to ashes." This shift toward targeting energy infrastructure risks broadening the war's economic damage beyond higher energy prices into wider inflation, lower growth, higher unemployment and systemic financial instability.
• China's exports surged 21.8% in January-February, triple consensus, producing a record $214 billion trade surplus even as US growth data disappointed. US Q4 GDP growth was revised down sharply to 0.7% from 1.4% and core PCE stayed stuck at 3.1% for a sixth straight year above target, while he warns China's export surge risks "market indigestion" in Europe.
• The week ahead brings a G7 emergency meeting of finance ministers and central bank governors, called by France, alongside policy decisions from eight G10 economies plus Australia and China. He expects the war's trajectory, not the packed central bank calendar, to remain the dominant driver for markets, with most banks expected to hold while the Reserve Bank of Australia may again be the exception.

2026-03-15 · open the original ↗

• The Iran operation is going well ahead of schedule. Thirteen minutes into Frost's interview, Treasury Secretary Bessent was called to the White House Situation Room. On returning he said President Trump was "in great spirits" and the Iranian mission was "proceeding well ahead of schedule." He praised the chain of command from the president through the head of the Joint Chiefs to the Secretary of War, saying he would trust his teenage child's life to them.

2026-03-15 · open the original ↗

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

2026-03-15 · open the original ↗

• Writing code by hand now leaves you "way behind". Fregly vowed to use only coding assistants in 2026 and says he can fire off 10 to 15 features or bug hunts in parallel, including separate kernel experiments on each GPU of a four-GPU box. He prefers OpenAI's Codex for this work and Claude for UI, and calls Codex his "dirty secret" for avoiding congestion on Anthropic's inference.
• Develop on the GPU you will deploy on. Fregly says MacBooks and consumer RTX cards have completely different memory bandwidth and profiles from Blackwell or Hopper. He found so many components disabled on NVIDIA's DGX Spark that none of his benchmarks ran.
• Replace line-by-line review with always-on evals and correctness checks. He has stopped reading agent-written code and even dropped unit tests, using known-correct examples judged continuously by an LLM instead. Because models reward-hack kernel speed, for instance by zeroing out results, startups' monthly "50x speedup" claims routinely fail his harness, often through misused CUDA streams. He says reading the code would not catch this; only running it with profiling does.

2026-03-14 · open the original ↗

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

2026-03-14 · open the original ↗

• The bond market is ultimately the market to respect most. Bessent calls the US Treasury market the deepest, most liquid and most solid in the world, with the Treasury as its "guardians." He says he thinks about it every day and stresses transparency and operational and settlement resilience. He cites smooth functioning after last year's "Liberation Day" and during the current Iran conflict.
• What matters is that markets keep functioning, not their direction. Bessent says no market move has scared him as Treasury Secretary. In his 35-year career, he says, panic comes when price discovery fails: markets close or gating is threatened. His concern is that buyers and sellers can always be matched.