0:00
4:34:49
RollerGator and Alex Marinos open with a long run of miscellaneous news: a quick take on the Elizabeth Holmes film, the five-day tenure of White House press secretary Katie Zachariah after she said she "moved to China" when asked where she passed the bar, new corruption allegations against Ambassador Kimberly Guilfoyle, a cyberattack that cooked refrigerators across French supermarkets, nationwide French protests, Elon Musk's cut to Ashley St. Clair's child support, a $6 million defamation verdict for MIT biologist David Sabatini, a $29.5 million settlement for a Las Vegas woman framed by detectives, a Wyoming grad student accused of poisoning a labmate, and a missing $470,000 shipment of RTX 5090 GPUs in Brazil. The hosts use the GPU story to argue that soaring hardware and memory prices are a hard-to-fake signal that AI demand is real.
The AI block starts with the arrest of an OpenAI engineer who hid a recording device in an executive conference room, and Apple's trade-secret fight with OpenAI. It then moves through a Roblox "digital grandma" test of Claude, a Cologne robot harassment case, and FTC Chair Andrew Ferguson's consumer-protection subpoenas to Anthropic and OpenAI. The center of the episode is a sustained critique of Anthropic: its warning about small models self-replicating across IoT devices, its quietly pitched federal AI safety bill (with Kelsey Piper's reaction and a "public benefit fund"), and what the hosts call a scattershot, self-undermining messaging strategy. The hosts also correct a 404 Media framing of Character.AI's bankruptcy, with Alex explaining the earlier $2.7 billion Google deal.
The final stretch is a run of "friends and enemies" segments and a long closing on mathematics. Clips of Eric Weinstein, Thunderf00t, Tyler Cowen on Dwarkesh Patel's show, and Cameron Berg's interview with philosopher Ned Block set up the hosts' ongoing argument that AI consciousness claims rest on definitional gaps. The show ends with a New York Times profile of Terence Tao, "on a war footing" after OpenAI's October 6 math release, and a long discussion of what mathematicians can do next, with a "five stages of grief" tracker for Tao and Alex's argument that the real question is whom the AI is aligned with.
Detailed Outline Welcome and the Holmes Film (00:00:00 - 00:04:04) Main Topic: "You Can See Everything" and How Much to Trust It- The show opens with the standard Twitter Space intro; Alex says he has not yet had a sufficiently dumb week "until we're done with this"
- RollerGator asks whether Alex will see the Elizabeth Holmes movie; Alex is torn, says he never had much interest in Holmes and that she has been "made into something much more than she deserves attention for"
- Alex says the film seems closer to a reality show than a documentary: scripted but realistic, like living your life in front of a camera to make a point
- RollerGator saw it at the first available showing, says it runs three hours and covers the 34 days before her 11-year prison sentence, and does not redeem her but offers perspective on her psychology
- RollerGator says the film's meta-referential structure forces viewers to question whether to take it at face value, and jokes you can see her eyeballs fill the entire screen
Key Quote: "It doesn't redeem her." — RollerGator
Press Secretary Katie Zachariah and "China" (00:04:04 - 00:10:12) Main Topic: Five Days on the Job and the Bar Exam- ABC and NBC clips report that new White House Press Secretary Katie Zachariah, whom Trump described as an attorney, never passed the bar; when Trump asked where she passed, she answered "China," saying she moved there with her new husband and was "distracted and overwhelmed"
- She had said "I'm a lawyer. I went to Pepperdine" and "I'm a rigorous researcher," called questions about her resume "miscommunication," and told Fox she was hired as press secretary and not White House counsel
- Trump says he told her "if I were you, I'd reconsider"; reports say she is out after five days, the shortest tenure ever for the job, and will move to a different role
- RollerGator jokes that she does rigorous research "except into perhaps her own history"; the hosts riff on Pepperdine versus Pepperidge Farm and on cities named Paris and Zeeland
- Alex says "China" should never be anywhere near the answer to what state you passed the bar in, and that she could have gotten further by saying Venezuela
Key Quote: "When asked what state you passed the bar in, your answer should never contain the word China." — RollerGator
Miscellaneous: Guilfoyle Faces Possible Firing (00:10:12 - 00:12:16) Main Topic: Trump Says "We're Looking at It"- ABC reports that Trump, asked whether he is considering firing Ambassador to Greece Kimberly Guilfoyle, said "we're looking at it, we're not happy"
- The report recaps that she was engaged to Donald Trump Jr. for four years and faces allegations that she pressured a Trump donor to pay her $100,000 credit card bill
- RollerGator says the controversy feeds the "pay-for-play" narrative the president does not need, and jokes that Trump will surely remove her to leave nothing to be said about those closest to him
- The hosts note this is the only Greek story of the week and decline to "go into the Cornell thing"
- NBC reports a cyberattack on cooling systems in French grocery stores, with fridge temperatures reportedly cranked as high as 122 degrees Fahrenheit and the Intermarché CEO expecting tens of millions of euros in losses
- Danfoss says the affected units were old enough to not receive updates or were running outdated firmware; cybersecurity expert Mark Rasch says nobody updates their doorbell and attackers are shifting from stealing data to shutting things down
- RollerGator notes Anthropic put out a memo that week warning AI models could soon self-replicate on consumer devices, and says he will get to it later
- Alex objects to the claim that nobody updates doorbells because he spent a decade building exactly that technology, a company he started in 2013 (Balena) and says he is no longer associated with
- He describes the trade-off: a centralized update platform is effectively a control backdoor, so the platform becomes a juicier target as more devices join, and adds "if we could put an intelligence on that device that could protect it, now you're talking"
- RollerGator says the story shows the EU's regulatory ambitions do not guarantee security, with Alex deadpanning "Nothing is ever a problem in Europe"
Key Quote: "The trade-off you make is that you have a centralized platform that becomes the gateway to your devices." — Alex
Miscellaneous: France in the Streets (00:20:31 - 00:30:00) Main Topic: Yellow Vest Revival and Riot Weapons- RollerGator reads an AP report on protests across France against rising living costs and school conditions: about 400 arrests, 110 marches, 550,000 marchers by the interior minister's count and 800,000 by organizers, with fires and looting in Paris, Lyon, Marseille, Limoges and other cities
- Interior Minister Laurent Nunez blames "extremely violent, radical far-left groups"; Macron says there is "no place for violence in our republic"
- A young protester lost a hand in Saint-Etienne; Nunez had suspended one type of stun grenade against high school demonstrators after a 15-year-old lost his hand, though the suspension does not cover other protests
- The hosts joke about the guillotine, French pronunciations, and the numbered Republics as a versioning system
- Alex calls it "just a return to normality" and says that if a country has not violently changed its form of government in 70 years, it is due for a new one
Key Quote: "If it hasn't violently changed its form of government in 70 years, it's due for a new one." — Alex
Miscellaneous: How Many Women Did Elon Musk Have a Kid With This Week? (00:30:00 - 00:40:00) Main Topic: Ashley St. Clair and the Child Support Cut- RollerGator brings back the song segment and reads Page Six: St. Clair says Musk was offered $15,100,000 a month in support in exchange for an NDA she would not sign, cut it to $40,000 a month after she refused, then to $20,000 after she publicly revealed their son Romulus in February 2025
- St. Clair, in a new documentary, says other women contacted her after being DMed by Musk and describes "a Gatling gun of sperm in his DMs"; Musk's 14th child is named Seldon
- The hosts joke about $20,000 a month being "poverty level," about naming conventions, and about whether Seldon is a Foundation reference
- RollerGator says Musk should not cut support to his child, but that $20,000 a month should be sufficient and that St. Clair's $15,000-a-month apartment is not warranted
- Alex asks how she is supposed to shitpost from anything less
Key Quote: "It's like a Gatling gun of sperm in his DMs." — Ashley St. Clair, as read by RollerGator
Miscellaneous: Sabatini Wins a $6 Million Defamation Verdict (00:40:00 - 00:45:07) Main Topic: Five Years in Court- The New York Times reports that former MIT and Whitehead Institute biologist David Sabatini won a defamation case and $6 million against Kristin Knouse, whose coercion allegations led to his forced exit in 2021; the jury rejected her counterclaims, and her lawyer says she will appeal
- Her lawyer said "society still has a difficult time believing women," which RollerGator notes ignores that the woman was believed at first, which is why the lawsuit existed
- Sabatini says "I've been dreaming about this moment for all this time" and hopes to be judged by his science
- RollerGator and Alex discuss how five years of litigation is a punishment on its own, and note that the burden is only preponderance of the evidence
- A woman who spent more than 15 years in prison receives a $29.5 million settlement after a jury found Las Vegas police detectives fabricated evidence in a 2001 killing; jurors had awarded $34 million and the settlement avoided years of appeal
- Forensic testimony placed the victim's death when Lobato was 170 miles away in Panaca; her lawyers said detectives manipulated her and destroyed notes
- Alex jokes the details are "blood-curdling," and RollerGator asks him whether he would take $29.5 million for 18 years in prison, with a joke about putting the contract in escrow
- CBS and AP report that a University of Wyoming microbiology graduate student from China was arrested at Los Angeles International Airport on suspicion of poisoning a colleague's tea with acrylamide after buying the chemical on eBay
- Campus police camera footage allegedly showed her rummaging through his desk, photographing notebooks, and rinsing out his mug; her bond was raised to $1.5 million cash
- The case is paired with a Berkeley case in which a coworker poisoned an engineer with cadmium and received probation
- RollerGator says "do not attempt to poison chemistry majors" and Alex adds that if you are going to do it, ask an LLM
- Tom's Hardware reports that distributor Oderco lost two pallets of Zotac RTX 5090 cards at Guarulhos airport, worth about R$2.5 million (around $470,000), with no official answers
- The report notes RTX 5090s vanished from online retailers in mid-September and recounts earlier hijackings of AI hardware
- Alex says the rise in GPU and memory prices is "a hard-to-fake signal" for what is going on, which a large audience wants to believe is not happening
- He cites Michael Burry's argument that hyperscalers were inflating GPU lifetimes as fraud, saying the market shows older GPUs are more valuable because what can be done with them is more valuable
- RollerGator distinguishes whether hyperscalers profit from their investment from whether model builders such as OpenAI and Anthropic recover their R&D costs
Key Quote: "That's a hard-to-fake signal for what is going on that there is a large audience that is determined to believe isn't happening." — Alex
The OpenAI Recording Device Arrest (01:04:00 - 01:18:00) Main Topic: Toilet Paper, a USB Port, and a Saudi Engineer- OpenAI says it found a sophisticated recording device in an office and used its models to understand it; the Justice Department charged engineer Abdulrahman Al-Fawzan, a Saudi national, with intercepting oral communications, with the DNI and FBI Director Kash Patel framing it as a national-security matter in the "superintelligence race"
- The complaint says the device was concealed in toilet paper in a USB port behind a TV in an executive conference room, installed around September 28, and that a chatbot account linked to his wife was used to ask how to build a continuous audio recorder, encrypt the SD card, and hide its activity
- A second device was found in a bag in a printer room; he, his wife, and a relative had booked one-way tickets to Saudi Arabia; the maximum penalty is five years in prison
- The New York Times reports he said he advised Saudi government agencies and was listed as a venture partner at Riyadh Ventures alongside a Saudi prince; he had met Mohammed bin Salman in 2016
- Alex notes the amateurishness of the plan, jokes that 100% of OpenAI's Saudi employees were spying on them, and speculates about a sovereign-AI competitor motive
- The hosts joke that Chinese actors would not need USB-port devices since they are "already in the chips in the smart TVs"
Key Quote: "Apparently there was only one Saudi employee of OpenAI, which has now dropped to zero, presumably, meaning that 100% of their Saudi employees were spying on them." — Alex
Apple vs. OpenAI: Trade Secrets Both Ways (01:18:00 - 01:22:05) Main Topic: Multiple Spider-Mans Pointing at Each Other- RollerGator reads OpenAI court filings in Apple's suit alleging OpenAI hired Apple staff to steal IP, in which OpenAI says a former Apple employee sent over 100 confidential files about an unreleased OpenAI device from personal accounts on his final day in June 2026
- OpenAI says the employee presented its product plans to Apple's senior leadership after returning
- Alex says it is hard to rule out cutthroat Silicon Valley behavior, but is skeptical it is as blatant as portrayed, and RollerGator asks how confident anyone can be that these companies can "align their robots" if their employees are not aligned with good employment practices
- TechTimes describes an X experiment in which Claude was given 60 seconds to recite Justin Bieber's "Baby" before a train hit a virtual grandmother; Claude refused to reproduce the lyrics and offered other help
- RollerGator plays the Roblox video; the hosts joke about arresting Claude, and Alex says a fate worse than death might be reciting Bieber lyrics
Key Quote: "Let's face it, you would have done exactly the same." — one of the hosts
The Cologne Robot Case (01:25:17 - 01:29:21) Main Topic: Who Is Liable When a Robot Misbehaves?- NBC reports that Cologne police confiscated a humanoid robot allegedly remote-controlled by a 16-year-old who livestreamed it touching a 16-year-old girl and making thrusting motions; the teens are suspects in a sexual harassment investigation
- A legal analyst says the robot is not liable because it is not a person but an instrument of whoever directs it
- RollerGator jokes about the analyst's potential Claude account being revoked for being cruel to Claude; Alex wonders what fidelity of control the streamer had and says it looks like a Unitree-type biped
- CNBC reports the FTC subpoenaed Anthropic and OpenAI over potential consumer protection violations; FTC Chair Andrew Ferguson says winning the AI race with China is the strategic imperative, that existing guidelines protect children and consumers, and that the probe is ordinary and about data security and honest representation
- Interviewers note Nebraska and Kansas attorneys general are pursuing similar deceptive-marketing claims against OpenAI
- RollerGator notes the claim that this is "an unregulated industry" does not square with existing rules, and jokes that the LLMs promised human extinction and have not delivered, so it is deceptive advertising
- Alex says he thinks there is a war between the administration and Anthropic, that Anthropic may not have a concrete, coherent internal direction, and that the administration is "not exactly the most straight player either"; he compares it to two mafia families with a pretense of law and order
Key Quote: "This is like 2 mafia families stabbing each other in some alley, but with a pretense of law and order on top." — Alex
Notable Detail: A guest, "Donald J. Trump, PhD," comes up to say that the Berkeley cadmium poisoner got two years of probation and had his felony reduced to a misdemeanor, so if you want to get away with poisoning, "do it in Berkeley."
Anthropic Warns of Self-Replicating Small Models on IoT Devices (01:37:34 - 02:07:16) Main Topic: "AI Worms" in the Smart Fridge- RollerGator reads Ars Technica: Anthropic says small open-weight models may soon replicate across IoT devices such as smart fridges and vacuums; its experiment ran small models on off-brand Android TV boxes, with Gemma needing about 15 boxes and Qwen configuring 8 of 9 hardware types
- Alex says the setup was a strong model on another machine walking through open ADB ports to put smaller models on devices that ran at about 0.4 tokens per second, and that he has a 20-million-parameter model running at 2.5 tokens per second on an ESP32
- Alex says self-replication is trivial once something has root access, but defenders can also host protective models, and that anything running code is hackable "first and foremost by the frontier models," so everything must be hardened long before LLM worms matter
- RollerGator adds that Metasploit already does most of the hacking; a CBS segment with a cybersecurity consultant is played, and Alex does an SNL-style bit about an AI worm jumping from a TV to a vacuum to an oven
- Alex says the series is "Anthropic discovers security," and objects that the company is polluting training data with ideas about AI taking over devices; he asks what the "underpants gnome" plan is
- The hosts discuss rent-seeking and investment-attraction motives, and Alex raises "safety fatigue," where constant warnings make real ones unheard
Key Quote: "It's, you know, we're all living through a long season of the series Anthropic discovers security." — Alex
Anthropic Quietly Pitches Its Own Safety Bill (02:07:16 - 02:27:52) Main Topic: FLOP Thresholds, Imminent Risk, and Kelsey Piper's Reaction- Alex notes the draft still uses a training threshold of 10 to the 25 FLOPs, as in California's bill, though modern frontier models are about 100 times larger; RollerGator explains FLOPs and says a low threshold could make small experimenters "frontier" companies required to pay third-party evaluators
- Politico reports the draft would let the government block a model only if it poses a foreseeable imminent risk of more than $1 billion in damages or killing or injuring more than 50 people, with evaluators registered with Commerce and potentially paid by the labs
- RollerGator argues AI-safety institutes and labs share staff, funding, social groups, and sometimes spouses, creating conflict-of-interest risk
- Kelsey Piper writes that Anthropic seems "terrified of the executive branch having the ability to flexibly make them stop," seeks preemption of state laws, and that the safety movement wants the opposite; the hosts say nobody can make sense of Anthropic's messaging
- A second tweet describes an Anthropic public benefit fund, which Alex had mocked as George Costanza's "human fund," with half of employees donating shares and "direct distributions to specific populations," which Alex reads as self-directed UBI or reparations
- Alex says that moving control of AI to the Trump administration while at war with it should give Anthropic pause; RollerGator cites Walter J. Peck from Ghostbusters shutting down the containment unit as the model for bad regulation
- Alex says Dario seems to be giving himself an anxiety attack while driving a car everyone is in, and notes "acquired AI risk" from overreaction, comparing it to lab-leak fears
Key Quote: "Anyone who has to deal with a bureaucracy becomes a libertarian, and the administration's treatment of Anthropic was genuinely outrageous." — Kelsey Piper, as read by Alex
Character.AI Files for Bankruptcy (02:27:52 - 02:42:23) Main Topic: What the 404 Media Framing Leaves Out- Forbes reports Character.AI filed for Chapter 11 and that Disney is the initial bidder at $20 million for a non-exclusive perpetual IP license; the company faced lawsuits over teen suicides and Disney's cease-and-desist
- 404 Media reads the filing as the first domino of the AI bubble: compute spending of $13.8 million a month in 2024, subscription revenue peaking at $2.8 million a month, $56.1 million on compute from January to August 2026, and losses of $8 to $9 million a month, with enshittification as a strategy
- Alex says the story leaves out that Google bought out the founders for $2.7 billion in 2024 (Noam Shazeer, an author of "Attention Is All You Need," has since joined OpenAI), so the bankruptcy is the leftovers sold off without the lawsuits attached
- RollerGator says the claims in the article about the market's economics are premature and concedes he had been confused
- Alex says Character.AI was a hybrid between a product company and a frontier lab, targeted kids, and has been a "dead company walking" for a couple of years
Key Quote: "It's been a dead, dead company walking for a couple years now." — Alex
Eric Weinstein on "Diary of a CEO" (02:42:23 - 02:57:46) Main Topic: "If You Don't Win, Your Morality Means Nothing"- RollerGator plays a clip in which Weinstein says he told Sam Altman and Dario Amodei they must win the race "almost at all costs," likens them to generals who must make terrible decisions, and says he declined to join the company because he did not want to be ruthless
- He calls the East Bay "cult city," says "you can't stop linear algebra," and discusses A Few Good Men, with Jessup as a man who should be on the wall
- RollerGator is "beside himself," pointing out that Weinstein constantly complains that nobody calls him about these problems while telling a story in which he turned down the work because he feared ruthlessness
- He jokes that Weinstein would rather have "frontier conversations" on a podcast, which becomes an "intellectual dark weave"
- The segment ends with a joke about an Oregon high school football coach fired after a 63-0 first half, suggested as a ruthless candidate for OpenAI
Key Quote: "If you don't win, your morality means nothing." — Eric Weinstein, as heard in the clip
Thunderf00t, Doom, and Searches That Are Wrong 50% of the Time (02:57:46 - 03:10:07) Main Topic: A Fellow Traveler With Bad Arguments- RollerGator, announcing "the worst one yet," plays YouTuber Thunderf00t arguing that AI doom is a marketing stunt, that life is resilient, and that LLMs are "fancy search engines" wrong on complex science questions about 50% of the time
- RollerGator notes the source of the 50% figure was an LLM's screenshot answering a chemistry and biology query, and that Thunderf00t is pegging himself to current model limits and ignoring tool use
- Thunderf00t argues a superintelligence would be a "brain in a jar with a billion off switches" dependent on human infrastructure
- Alex relates a dispute in which Thunderf00t claimed Elon Musk failed on the Dear Moon mission, while a clip showed Musk heavily caveating the date and the customer nodding along; he also says Thunderf00t confused Dear Moon with Artemis
- Alex gives him credit for one claim that checked out: all LLMs say Wikipedia is more reliable than they are, which Alex counters by saying Wikipedia has no self-correction mechanism
Key Quote: "I think there is an argument at the bottom of his arguments that can be constructed that is good, but Jesus Christ, he's bad broadly speaking." — Alex
Tyler Cowen, Dwarkesh Patel, and "AI Psychosis" (03:10:07 - 03:17:48) Main Topic: A New York Times Conversation on Bay Area Belief- The hosts play a New York Times opinion conversation in which Patel says people are too deferential to AI and puts AI psychosis in the Bay Area at about 20%, and Cowen says he has a minimum of it because he does not call AIs sentient
- Patel says he does not have a strong opinion and that they have functional analogs of emotion and a global workspace; Cowen responds that Stockfish beats Magnus Carlsen and nobody thinks it is conscious
- Patel says people are bad at being coherent about sentience, citing factory farming, and that a 5% chance they are sentient would make him feel bad about turning them on
- RollerGator says the people asserting LLM consciousness are exhibiting their own AI psychosis, and that Cowen "is stealing my ideas"
- RollerGator reads the introduction of Cameron Berg's interview with NYU philosopher Ned Block, "two of the leading thinkers in the field of AI consciousness," recorded at Lighthaven, which Alex describes as an EA and rationalist gathering place bought with money SBF sent just before the FTX collapse, with Jaan Tallinn lending funds to cover the clawback
- Berg says he has a student training a model on consciousness-free text; Block says that is nearly impossible and says he has a "very low prior" for consciousness in current AIs because they are trained on text to sound like us, with no breakthrough beyond backpropagation
- Block speculates the original function of consciousness was motivational (pain and pleasure), and that nothing in reinforcement learning creates it
- RollerGator says Block "echoed everything I've said for the last month and a half" and credits Berg for posting an interview where the guest effectively laughs at his ideas
- Alex says no one made an argument in the exchange and that "backpropagation" is no more an argument than "neurons firing"; he warns that Berg's position is "just as defensible" in a field without experiments and may gain firepower from an IPO
- RollerGator argues the pro-consciousness camp exploits the gap in defining consciousness in humans, positions itself as the most moral, and that he can self-attest to his own consciousness, while Alex plays the solipsist and the "nolipsist"
- RollerGator repeats his Kantian argument for being polite to LLMs, to avoid training oneself to be a jerk
Key Quote: "Machines fill all the gaps, man. It's all happening in there." — Alex
Terence Tao: The New York Times Profile (03:44:01 - 03:58:14) Main Topic: "Why One of the World's Greatest Mathematicians Is on War Footing"- Alex agrees to cover Tao on the condition that they track his five stages of grief week to week
- RollerGator reads the Times piece: after OpenAI's October 6 release of more than 350 findings, Tao says "I'm on a war footing," "there are no rules anymore," that he is "just numb," and that OpenAI "used math for what they needed, which was to evaluate their models"
- Tao says on his blog he is "horrified" by person-years of research being casually disrupted, and says the video OpenAI cut from an hour to one minute was an ad he should have pushed back on
- Tao calls the Navier-Stokes proof "technically readable" but not understood by any expert, earlier called the upcoming results "toxic waste," and now calls them "maybe treated waste"
- At a UCLA workshop, participants discuss the $22 million estimated cost of the swarm of 10,000 agents running 88 hours, the undemocratic cost of AI, and Tao says he is "soul-searching at gunpoint"
- PhD students Ben Goldman, Siddharth Malhotra, and Aryan Nadjimzada describe uncertainty, a field "strip-mined," and unreadable AI-generated proofs
- A Harvard report recommends more in-person dissertation defenses, and Tao says problem solving is a journey and students can use AI to solve ten problems and find new patterns
- RollerGator and Alex decide Tao is somewhere between depression and acceptance, with parts of anger and bargaining
Key Quote: "I'm on a war footing." — Terence Tao, as read from the New York Times
What Mathematicians, Programmers, and Everyone Else Do Next (03:58:14 - 04:27:01) Main Topic: Sympathy, Derision, and "Who Is the AI Aligned With?"- RollerGator, a software engineer, says programmers are laughing at mathematicians because they were told to "learn to code" and then were the first threatened; he is defensive of mathematicians and thinks the best will find the intersection of low-hanging problems with practical utility
- Alex says he mocks the proposed solutions, such as "stop publishing new math," not the existential dread, and that Tao's best option may be to wait
- Alex says mathematicians are told not to go after frontier problems because it is bad career strategy, and the new world may let them be more ambitious
- He argues Tao is upset because the AI is not aligned with him and the establishment he was part of, noting that "alignment" never specifies "with whom," and asks whether the AI will next go after physicists, biologists, lawyers, and hedge funds, and whether the establishment will push back
- Alex tells how he had Claude formalize the Community Notes algorithm in Lean, a side effect of frontier math capability the company did not intend for him
- RollerGator says universities may disappear; he praises Dover Publications as cheap access to graduate-level math, and notes that universities supply deadlines and pressure
- RollerGator worries about people who lose their external motivation and expects increases in alcoholism; Alex responds that he is thinking about civilization's survival, and that perhaps AI is "bullying us into getting our shit together" as the pre-AI system was balkanizing
Key Quote: "The most interesting part of the whole mathematician drama is who is the AI aligned with?" — Alex
Closing Thoughts (04:27:01 - 04:34:14) Main Topic: Elections, the Anthropic Bill, and a Sign-Off- Alex says that if Democrats win the House and Senate, it will be interesting to see whether Anthropic pushes its AI bill, and says he is sure the president will not make any stupid last-minute deals to get his library funded
- RollerGator jokes that Alex is cynical, and Alex says human communication has tone and subtext
- RollerGator wishes everyone "a wonderfully dumb week" and says he will see them next Sunday
The episode opens with a very long run of miscellaneous news, from the press secretary who "moved to China" and Guilfoyle to the refrigerator cyberattack, the French protests, Musk's child support, the Sabatini and Lobato verdicts, and a Wyoming poisoning. The missing RTX 5090 shipment leads into an AI block on OpenAI's recording-device arrest, Apple's trade-secret suit, a Claude test, a Cologne robot case, and the FTC.
The center of the show is an extended critique of Anthropic's IoT worm warning, its draft safety bill, and its mixed messaging, followed by the Character.AI bankruptcy, where Alex corrects the 404 Media framing. The last third turns to people the hosts treat as flawed allies or opponents on AI consciousness and doom (Weinstein, Thunderf00t, Cowen and Patel, Berg and Block), then to Terence Tao and a long discussion of what mathematicians and everyone else can do as AI outpaces expert work.
Weitere Episoden von „This Dum Week“
Verpasse keine Episode von “This Dum Week” und abonniere ihn in der kostenlosen GetPodcast App.
