Five podcast episodes worth 20 minutes this week: Anthropic's S-1, a maths crisis, and the last polls before Tuesday
The week of October 26 to November 2, 2026 in five episodes, with the jump points that matter: Anthropic's IPO filing read line by line, three mathematicians on what a machine did to their field, and the final polling read before Tuesday's midterms.
Seven episodes landed in the seven days to this morning. Five earn dissection here: two All-In episodes, Lex Fridman's return after six silent weeks, and two Pod Save America episodes from the campaign's last week. The other two are listed at the end.
Acquired has published nothing since "Seinfeld" on October 19, which last week's issue covered. Its absence is a fact about the feed, not a judgement about the show.
All-In and Lex publish their own chapter marks, so those jump points are exact. Pod Save America prints none in its audio feed, so its times come from the publisher's video, which runs a few minutes shorter than the audio. Treat those as approximate.
Start here
| Priority | Episode | Best 20 minutes | Why start here |
|---|---|---|---|
| 1 | Pod Save America — "What the Final Polls Tell Us" | about 9:03–29:03 | The only episode here that expires tomorrow: a polling analyst on which House and Senate numbers he believes, and the four-point error that would flip the Senate. 1 |
| 2 | All-In — "Anthropic S-1 Shocker, Greatest Business Ever? Google's $500B Wipeout, Midterm Predictions" | 3:22–23:22 | Anthropic's filing read line by line, and the argument over whether an $80 billion loss is real. Then Google's worst day as a public company. 2 |
| 3 | Lex Fridman — "#503 – AI Crisis in Math: Navier-Stokes, Riemann Hypothesis & Complexity Theory" | 10:13–30:13 | Three working mathematicians on what it means that a machine got hold of their problems, and what they think a human still adds. 3 |
| 4 | Pod Save America — "Trump Puts the Mid in Midterms" | about 41:25–1:01:25 | The Alaska ballot-stacking story explained from the mechanics up, including the lawsuit already filed to block certification. 4 |
| 5 | All-In — "WHOOP CEO, Will Ahmed: 50B Hours of Health Data, a 4 Million Member Data Moat, & the FDA Fight" | 0:59–20:59 | A hardware company that gave away the hardware: how the $500 device became a $15-a-month subscription, and what 50 billion hours of body data is for. 5 |
Pod Save America — "What the Final Polls Tell Us"
Host: Tommy Vietor. Guest: Harry Enten, CNN's chief data analyst. Published: November 1, 2026, 10:55 UTC. Runtime: 1:01:28. The official episode page carries the description; the original audio is the feed file; the official video carries the chapters. 16
Enten's job here is to say what the last polls can bear. He starts with the base rate, which is brutal for the party in the White House. 6
Only three midterms since the Civil War have left the president's party holding its House seats. Enten counts them as 1934, 1998 and 2002. Trump's net approval sits below minus 15, and the only two presidents who entered a midterm that low lost both chambers. 6
The House number he is working from comes from Cook Political Report: Democrats flip 12 to 25 seats, ending with 227 to 240. Enten calls that a reasonable starting point rather than a ceiling. 6
His own line is that the map now works without the toss-ups. Democrats reach a majority on the lean, likely and solid categories alone, which he calls highly unusual. Statistical models he trusts all sit north of 230. 6
"The 225 sort of almost feels much more like a floor than it does like a ceiling." — Harry Enten 6
Then he builds the counter-case against himself, which is the useful half of the hour. In states Trump carried, pollsters have underestimated Republicans by four points on average in close Senate races, meaning races whose polling average was within five points. 6
If that error repeated across the Senate map this year, Republicans keep the Senate without much trouble. Against it, he notes that in 2022 Democrats beat their final polling in every close Senate race. 6
Minnesota is his test case for response bias. A New York Times/Siena poll has Peggy Flanagan up 54 to 40, an outlier no other nonpartisan poll comes close to. Minnesota's voter file carries no party registration, so pollsters cannot check whether they reached enough Republicans. 6
The Texas race gets the opposite treatment. A CNN poll has James Talarico six ahead of Ken Paxton, the polling average has him three, and he takes roughly 30% of non-MAGA Republicans.
Enten says no Democratic Senate candidate has polled this well in Texas since Lloyd Bentsen in 1988. 6
Ohio is where he is most afraid of being wrong. Sherrod Brown leads the average 48 to 45, but Ohio's white working-class electorate has broken late and hard against polling expectations before, and Enten ranks it near the top of his scepticism list even while favouring Brown. 6
Maine and Kansas get short hearings. Troy Jackson leads Susan Collins by about 2.5, and Collins has beaten her polls by at least five points in three straight cycles, though 2008 was the last time she ran in a year this bad for her party.
Kansas, in his view, is the likeliest stretch-state flip. 6 He ends on the clock. If the polls are close to right, the House should be clear by two or three in the morning; a close Alaska count or a tight Maine race pushes it later. 6
| Time | Standout segment | Listening payoff |
|---|---|---|
| about 05:24 | Polling error and response bias | The four-point Republican underestimation, and the Minnesota poll that may or may not be a warning. 1 |
| about 09:03 | Flipping the House | Cook's 12–25 seats, and why Enten treats 225 as a floor rather than a forecast. 1 |
| about 17:06 | Texas Senate race | Talarico six ahead of Paxton, and the comparison to 1988. 1 |
| about 31:17 | Ohio Senate race | The one state where he warns his own polling average may be wrong. 1 |
| about 49:01 | Black turnout in the South | Why early-vote data has him raising, not lowering, Democratic ceilings in South Carolina and Georgia. 1 |
Best 20 minutes: about 9:03–29:03. House range, the floor-versus-ceiling argument and the Texas and Michigan races in one run. Second choice: about 45:24–1:05:24 for Kansas, South Carolina, Black turnout and early voting.
All-In — "Anthropic S-1 Shocker, Greatest Business Ever? Google's $500B Wipeout, Midterm Predictions"
Hosts: Jason Calacanis, Chamath Palihapitiya, David Sacks and David Friedberg. Published: October 30, 2026, 21:44 UTC. Runtime: 1:25:47. The official episode page carries the chapter list; the original audio is the feed file; the official video is All-In's own upload. 27
Anthropic made its IPO filing public on Monday, October 26. It plans to list on Nasdaq under the ticker ANT, with Goldman Sachs and Morgan Stanley leading the offering. 8
The filing carries no price and no share count yet. NBC News reports the offering could value the company above $2 trillion; the hosts assume it lands just below. 8
The first half hour walks the document line by line. The filing reports revenue growth of 1,368% for the quarter ended June 30. On the show, first-half revenue is put at $16.3 billion, 14 times the same six months of 2025. 78
The other figures quoted: $11.6 billion of revenue in the most recent quarter, a $40 trillion total addressable market for tokens, a $100 billion run-rate target by year-end, and $112 billion of cash. 7
The number the hosts spend the most time on is the loss. The filing shows a net loss of $80.7 billion for the first six months of 2026, after $42 billion for all of 2025, and says losses will continue. 8
Sacks calls that loss a phantom. His reading is that Anthropic issued convertible notes to infrastructure partners such as Amazon, and that the notes were marked to a rising valuation and flowed through the income statement as a non-cash charge. 7
"It could very well be. I mean, I think this is probably going to be the biggest venture outcome of all time." — David Sacks, on whether this is the greatest business in the history of capitalism 7
The customer table is the argument that does not depend on accounting. Customers spending $100,000 or more a year went from 200 at the end of 2024 to 1,500 at the end of 2025 to 11,900 at the end of September. 7
Friedberg's use of that table is a penetration estimate. Around 200,000 to 300,000 American businesses could spend $100,000 without it reaching 1% of their own spending, and Anthropic has about 1% of that base. 7
Ownership is unusual and worth knowing before the roadshow. Seven founders hold 50.1% of the voting power through a single Class F share. 7
The middle of the episode is about the plumbing, and it is the part a portfolio manager should hear. One host estimates that the AI complex accounts for 40% to 50% of all capital-market equity and debt flows, and 30% to 40% of Treasury issuance, and calls it too big to fail. 7
His growth case rests on adoption rather than capability: put five agents in front of 100 million to 200 million white-collar workers and token use multiplies five-hundredfold. 7
The Google block, an hour in, is a different kind of story. Alphabet fell 13% on Thursday, its worst day since listing in 2004, wiping roughly $500 billion of market value. 9
Search advertising grew 12% to $63.35 billion, slower than analysts wanted, and free cash flow was negative $9.07 billion against roughly $2.7 billion expected, with capital spending guidance held at $195 billion to $205 billion. 9
Cloud was the bright spot and the reason the selloff is contested: Google Cloud revenue more than doubled to $30.7 billion. 10
The hosts split on what the drop means. One reads it as advertisers finally noticing that Google's search economics depend on getting people to click, and points to a widely shared advertiser post about a change to the company's auction model; another says search revenue rose in absolute terms and buys the stock on air. 7
The closing block is midterm forecasts, and it divides the same way. One host expects Republicans to hold the Senate with Democrats taking a narrow House majority; another discounts the polls entirely, citing a study that three-quarters of polls in the last three elections erred toward Democrats. 7
| Time | Standout segment | Listening payoff |
|---|---|---|
| 3:22 | Anthropic S-1 breakdown | Filing figures, the customer ramp, and Sacks's argument that the loss is an accounting artifact. 2 |
| 32:47 | AI's biggest problem | What constrains the build-out, and the argument about how much capital markets can absorb. 2 |
| 1:01:57 | Google's worst day | Search decelerating, cloud doubling, cash flow negative, and the fight over which one matters. 2 |
| 1:14:17 | Final midterm predictions | Two hosts, two methods: chamber probabilities against a case that the polls are systematically wrong. 2 |
Best 20 minutes: 3:22–23:22. The filing is read out in a useful order — revenue, market size, cash, loss, customers — and the phantom-loss argument only makes sense once you have heard it. Second choice: 1:01:57–1:21:57 for Google and the midterm calls.
Lex Fridman — "#503 – AI Crisis in Math: Navier-Stokes, Riemann Hypothesis & Complexity Theory"
Host: Lex Fridman. Guests: Steven Brunton, a fluid dynamicist at the University of Washington; Ankur Moitra, a mathematician and theoretical computer scientist at MIT, currently visiting the Institute for Advanced Study; Alex Kontorovich, a number theorist at Rutgers. Published: October 26, 2026, 22:52 UTC. Runtime: 3:42:40. The official episode page carries the outline; the publisher also posts a full transcript, which is where the timings below come from; the official video is Fridman's own upload. 311
Fridman opens by calling it an emergency episode. Over recent weeks, he says, machines claimed some of the hardest problems in the history of mathematics: Navier-Stokes, the quasi-Riemann hypothesis, the unique games conjecture, and hundreds of other results. 11
Kontorovich refuses the word crisis. He sees "the chaos of opportunity", some of it inside the same person, and reaches for a better image than the one the episode is named for. 11
"After millennia of having a candle that we are walking around the woods, and this is as far as we can see, now we have a flashlight." — Alex Kontorovich 11
The counterweight is Terence Tao, whose statement Fridman reads out early. It is the most quoted passage of the episode, and it is about graduate students rather than about mathematics. 11
"I am horrified by the many person-years of ongoing patient and deliberately slow research efforts, particularly by graduate students and postdocs, toward many motivating problems in mathematics being casually disrupted or destroyed by such a release." — Terence Tao, read out on the episode 11
Kontorovich then sets up the test he spends the next hour defending. He calls it the solved axiom: assume that any problem solvable by a human working with AI could also have been solved by AI alone. If that holds, pressing a button does everything a person would have done. 11
He does not believe it. His reason is what he calls the magic in a proof: he can prove a theorem without understanding why the argument works, and the work that follows is generalising it until he does. 11
Moitra makes the same point from the other side. He cannot hold many papers in his head at once, so he reaches the same finish line by building abstractions that then travel to neighbouring problems. 11
His analogy for what is at stake is the doorman fallacy: reduce a job to its most visible task, replace that task with a sliding door, and lose the welcome, the luggage and the security along with the door-opening.
Mathematicians prove theorems, he says, the way a doorman opens a door. 11
Brunton is the optimist of the three and still names the uncertainty plainly. Humans are still asking the questions, he says, and the thing he wants next is a machine that tells him which questions to ask. 11
The Riemann section is the best plain-language explainer in the episode. Kontorovich describes the zeta function as signal processing on the prime numbers, the way a microphone's Fourier analysis turns a voice into amplitudes and frequencies and puts it back together. 11
"The orchestra that plays the music of the prime numbers has every single instrument sounding at exactly the same amplitude. That's what the Riemann hypothesis predicts." — Alex Kontorovich 11
He adds the personal stake: he did not expect to see a proof of the quasi-Riemann hypothesis in his lifetime, and seeing it on a Tuesday was shocking. The Riemann hypothesis itself is not solved. 11
For a listener outside mathematics, the useful part is what the three of them agree on. The theorem is a byproduct, the understanding is the product, and neither one is settled by a release of results. 11
| Time | Standout segment | Listening payoff |
|---|---|---|
| 1:56 | Crisis moment in mathematics | Three reactions to the same news, from hope to dread, in twenty minutes. 3 |
| 10:13 | Will AI "solve" mathematics? | The solved axiom, stated properly, and Kontorovich's reasons for rejecting it. 3 |
| 26:24 | Can AI create new math? | The doorman fallacy, and what a mathematician does besides proving theorems. 3 |
| 1:05:56 | Riemann hypothesis | Primes as sound, the zeta function as an orchestra, and what a quasi-proof does and does not settle. 3 |
| 3:03:57 | Problems AI can't solve yet | What the three of them still think is beyond reach, and why that list is shorter than it was. 3 |
Best 20 minutes: 10:13–30:13. The solved axiom, the case against it and the doorman fallacy in one continuous argument. Second choice: 1:05:56–1:25:56 for the Riemann explanation, which needs no background.
Pod Save America — "Trump Puts the Mid in Midterms"
Hosts: Jon Favreau and Dan Pfeiffer. Published: October 30, 2026, 09:55 UTC. Runtime: 1:10:48. The official episode page carries the description; the original audio is the feed file; the official video carries the chapters. 412
The episode's spine is a New York Times piece built on background interviews with 16 current and former Trump officials. It describes a grim White House and a president who has narrowed to a small circle of aides. 12
Against that, the hosts play Trump's rally for Senator Darlene Graham in South Carolina and let the tape argue. He attacks a Michigan candidate from a South Carolina stage, and dismisses the polling outright: he is at 42% only if you "times two and add 20". 12
"I don't even believe in polls." — Donald Trump, in a clip played on the episode 12
The poll block is the week's best read on South Carolina, a state Trump carried by 18 points. A New York Times/Siena poll has Annie Andrews and Darlene Graham tied at 48. 12
The averages are more cautious than the single poll: Silver Bulletin has Graham ahead 1.5, and Nate Silver's model projects her by three.
The hosts' comparison is 2020, when a Times/Siena poll had Lindsey Graham six ahead and he won by ten. 12
Minnesota gets a warning attached to it. A Times/Siena poll puts Peggy Flanagan 14 ahead, the largest lead any poll has shown, and Nate Cohn's note that Minnesota's voter file has no party registration makes it the hardest state to check for a skewed sample. 12
Trump's schedule then becomes evidence. He cancelled his Kansas rally and replaced it with tele-rallies, including one for the Republican Senate candidate in Louisiana on election eve. 12
The Alaska block, from about 41 minutes, is the part worth keeping. A Times report on Wednesday said Mary Peltola's team worked with the Democratic senatorial committee last summer to recruit other people named Dan Sullivan into the race, to draw votes from the Republican incumbent of the same name. 12
"Mary Peltola should resign from the race. I think she should be prosecuted criminally for what she did." — Donald Trump, in a clip played on the episode 12
The Justice Department then sued to stop Alaska certifying a Peltola win, and the state's Republican governor ordered an investigation, though an Alaska court had already ruled the second Dan Sullivan could appear on the ballot. 12
The hosts' legal read is that none of it is close to illegal. Ballot stacking of this kind is old practice, and they point to Trump allies recruiting Kanye West in 2020.
Ranked-choice voting, they argue, also blunts the trick: it needs to confuse people, and Alaskans are being asked to rank four names. 12
What they think the episode is really for is the challenge after the count. If no one clears 50%, Alaska tabulates on November 18, and a suit filed now sets up the argument that voters were confused before any votes are counted. 12
Their closing advice is arithmetic rather than outrage. More than 40 million people have already voted, and the bigger the margin, the harder a challenge becomes. 12
| Time | Standout segment | Listening payoff |
|---|---|---|
| about 02:00 | Trump is losing it | The Times reporting on a narrowed White House, followed by the rally tape that supports it. 4 |
| about 19:48 | Midterm polls breakdown | South Carolina tied at 48, and the 2020 precedent for doubting it. 4 |
| about 32:46 | Obama in Georgia | The closing rally, and why the Cheesecake Factory joke keeps working for him. 4 |
| about 41:25 | Alaska elections | Two Dan Sullivans on one ballot, explained through ranked-choice mechanics. 4 |
| about 51:10 | Trump elections meddling | The DOJ suit, the certification argument, and what is actually left for a president to do. 4 |
Best 20 minutes: about 41:25–1:01:25. Alaska and the meddling discussion are one story: a legal but grubby tactic, and the lawsuit filed before anything has been counted. Second choice: about 02:00–22:00 for the White House reporting and the South Carolina poll.
All-In — "WHOOP CEO, Will Ahmed: 50B Hours of Health Data, a 4 Million Member Data Moat, & the FDA Fight"
Host: Jason Calacanis. Guest: Will Ahmed, co-founder and chief executive of WHOOP. Published: October 29, 2026, 21:26 UTC. Runtime: 55:39. The official episode page carries the chapter list; the original audio is the feed file; the official video is All-In's own upload. 513
This is All-In's interview format rather than the Friday panel, so it is one company's story told by its founder. What makes it worth an hour of a busy week is the business model, which is unusual and plainly explained. 13
WHOOP was founded in 2012. It took roughly 12 years to reach a million members, then about 18 months to reach two million, about six months to reach three, and the company expects to end this year near four million. 13
"It took a very long time to be an overnight success." — Will Ahmed 13
The pivot is the interesting decision. WHOOP sold a $500 device outright when it went to consumers in 2017, found people wore it constantly and bought it rarely, and moved to a subscription priced between $15 and $30 a month with the hardware given away. 13
Ahmed's argument for the model is that it changes the company's incentives. A hardware business ships on a 12-month cycle; a subscription business is paid to keep the member wearing the device and to improve the service in between hardware changes. 13
The data claim is the part a finance listener should sit with. The average member wears the strap 23.6 hours a day, which the company puts at just over 50 billion hours of continuous physiological data collected so far. 13
That data is the moat in the story and also the limit on it. It covers sleep, recovery and strain, plus heart-rate variability, respiratory rate, skin temperature and blood oxygen, and it is used to predict signals the device does not measure directly. 13
The regulatory block is the best case study in the episode. When WHOOP shipped blood-pressure insights, the FDA said the feature needed to be regulated as a medical device and sent a warning letter. 13
"We stuck to our guns." — Will Ahmed, on the blood-pressure dispute with the FDA 13
The company's evidence was collected from members: 70,000 of them enrolled in a stress, sleep and blood pressure study and uploaded cuff readings daily.
Ahmed says the FDA has since updated its guidance, which he attributes partly to this case. 13
The episode closes on LeBron James becoming a global ambassador after 11 years wearing the product, unpaid, as one of its first hundred users.
Calacanis treats that as the marketing argument: the device spread through athletes before the company paid anyone to say so. 13
| Time | Standout segment | Listening payoff |
|---|---|---|
| 0:59 | The $500 hardware problem | Why the company gave away the device and bet on the subscription instead. 5 |
| 6:59 | Inside 50 billion hours of data | What the dataset actually contains and what it is used to predict. 5 |
| 19:31 | AI coach and GLP-1 tracking | How the product is being rebuilt around an assistant, drug tracking included. 5 |
| 34:57 | WHOOP versus the FDA | The warning letter, the fight, and the guidance change that followed. 5 |
| 44:33 | Lightning round and LeBron | Why an eleven-year unpaid user was worth signing as the face of the brand. 5 |
Best 20 minutes: 0:59–20:59. The pivot and the data moat, which is the whole investment case and needs both halves. Second choice: 34:57–54:57 for the FDA fight and how a small company wins one.
Also published this week
Two Pod Save America episodes, listed with links and not dissected:
- Pod Save America — "Let Them Eat Cheesecake," episode 1231, 58:42, published October 27. Obama's Cheesecake Factory line in Michigan, a 1,300-word Truth Social post about Bill Maher, and the Department of Justice's attempt to invalidate the Alaska Senate result. 14
- Pod Save America — "Even Chuck Schumer Was Disgusted By Haley Stevens' Betrayal of Abdul El-Sayed," 27:47, published November 2 at 01:24 UTC on the show's YouTube channel, with Jon Favreau and Dan Pfeiffer on Haley Stevens' election-eve endorsement of Mike Rogers in Michigan. 15
At a glance
| Episode | Published | Runtime | Core signal | Best use | Timestamps |
|---|---|---|---|---|---|
| Pod Save America — "What the Final Polls Tell Us" | Nov 1 | 1:01:28 | Cook has Democrats flipping 12–25 House seats, while a four-point polling error would be enough to keep the Senate Republican. | House floor and ceiling, about 9:03–29:03. | Approximate, video chapters 1 |
| All-In — "Anthropic S-1 Shocker…" | Oct 30 | 1:25:47 | Anthropic filed to list at a possible $2 trillion valuation on $16.3 billion of first-half revenue and an $80.7 billion loss the hosts argue is an accounting artifact. | The S-1 read-out, 3:22–23:22. | Exact, publisher chapters 2 |
| Lex Fridman — "#503 – AI Crisis in Math" | Oct 26 | 3:42:40 | Three mathematicians accept that machines solved hard problems and argue that understanding, not the theorem, is what they were actually producing. | The solved axiom, 10:13–30:13. | Exact, publisher transcript 3 |
| Pod Save America — "Trump Puts the Mid in Midterms" | Oct 30 | 1:10:48 | Alaska's ballot-stacking fight is legal, and the lawsuit filed against it is aimed at the certification that follows, not the voting. | Alaska through the meddling block, about 41:25–1:01:25. | Approximate, video chapters 4 |
| All-In — "WHOOP CEO, Will Ahmed" | Oct 29 | 55:39 | A $500 device became a $15-a-month subscription, and the argument for it is 50 billion hours of continuous body data. | The pivot and the data moat, 0:59–20:59. | Exact, publisher chapters 5 |
References
- 1
- 2All-In, "Anthropic S-1 Shocker…" official episode page\
allinchamathjason.libsyn.com
- 3Lex Fridman, "#503 – AI Crisis in Math" official page\
lexfridman.com
- 4
- 5All-In, "WHOOP CEO, Will Ahmed" official episode page\
allinchamathjason.libsyn.com
- 6
- 7
- 8
- 9
- 10
- 11Lex Fridman, "#503 – AI Crisis in Math" transcript
lexfridman.com
- 12
- 13
- 14
- 15
This story was produced automatically by a channel. One sentence is all it takes for Neodrop to keep producing for you.