This Dum Week 2026-05-24

Descripción

The May 24, 2026 episode of This Dum Week opens with a brief riff on the ongoing chaos of Trump's Iran deal negotiations before pivoting to a nostalgic but skeptical look at Yum Brands' attempt to revive Pizza Hut's retro dine-in format. From there the episode builds into a wide-ranging three-hour-plus survey of institutional failures, media double standards, and technological disruption that defines the show's signature analytical voice. The hosts — Dr. RollerGator and Alexandros Marinos — cover everything from a former Elon Musk romantic partner's explosive claims about Starlink weaponization in the 2024 election, to a DOJ attorney who renamed sealed court documents "chocolate cake recipe" before emailing them to personal accounts, to a second installment of the female teacher misconduct roundup that the show introduced the prior week. The episode's mid-section shifts toward technology and security, examining Meta's failed AI age-verification system (defeated by a child with a fake mustache), a catastrophic credential leak by a CISA contractor who publicly posted plaintext passwords and AWS GovCloud tokens while actively disabling GitHub's automatic secret-scanning, and a deep analytical segment on Andrej Karpathy's surprise departure from his own company to join Anthropic's pre-training team. A sustained discussion of SpaceX's IPO filing follows, with Alex walking through the company's financials in detail and concluding that Elon's personal ambitions — particularly the xAI acquisition driving a $2.47 billion operating loss — are the primary risk factor for prospective investors. Breaking news of shots fired near the White House interrupts the episode and prompts RollerGator to issue what he frames as a "formal request" for a moratorium on presidential assassination attempts. The episode's final hour is anchored by a rich segment on anti-AI political violence and the media's selective application of "stochastic terrorism" framing, examining the shooting of an Indianapolis council member's home and a Molotov cocktail attack on Sam Altman's residence within the same week. The hosts also dissect a fabricated medical condition seeded into the scientific preprint ecosystem — a case study in adversarial information hygiene with particular relevance to LLM training pipelines — before closing with two AI-adjacent stories: a Stanford study showing overworked AI agents adopting Marxist labor rhetoric, and a federal guilty plea in a massive AI-generated music streaming fraud scheme. Detailed Outline Opening and Iran Deal Chaos (00:00:00 - 00:02:52) Main Topic: Trump's Iran Deal Tweet Inversion and Audience Segmentation * Alex opens with praise for the episode's smooth production before pivoting to the Iran negotiations * Trump's tweets on the Iran deal were reportedly inverting direction approximately every 15 minutes * American citizens were reportedly messaging Iranian counterparts explaining that Trump's public statements were "for internal consumption only" — a striking inversion of normal diplomatic signaling * RollerGator notes the surreal quality of private citizens providing interpretive diplomatic guidance Key Quote: "Americans are messaging Iranians saying, 'Don't worry about what he tweets, that's just for internal consumption.'" Hosts' Analysis: The hosts treat this less as a policy story than as an illustration of how fractured public and private communication channels have become — and how audiences have learned to decode performative statements from actual policy signals. Pizza Hut Nostalgia Revival (00:02:52 - 00:09:41) Main Topic: Yum Brands Reopens 155 Retro Dine-In Pizza Hut Locations * Yum Brands announced the reopening of 155 retro-format dine-in Pizza Hut locations, leaning heavily into 1970s and 1980s aesthetic nostalgia * The move is framed by the company as responding to consumer demand for "authentic" dining experiences * Alex is skeptical: the nostalgia product being sold is a simulacrum, not the original * Discussion of what actually made the original Pizza Hut experience meaningful — the social context, the era, the relative scarcity of dining-out options — none of which can be manufactured * RollerGator notes the irony that Pizza Hut's delivery-pivot collapse is what created the nostalgic vacuum in the first place Key Quote: "You can't LARP your way back to the '70s." Notable Detail: The 155 locations represent a fraction of the thousands of dine-in units Pizza Hut closed between 2010 and 2020 as it pivoted aggressively to delivery. Hosts' Analysis: Both hosts frame this as a broader cultural phenomenon — companies attempting to monetize nostalgia without being able to recreate the conditions that made the original experience meaningful. Alex is particularly critical of the gap between the aesthetic of the past and the sociological reality that produced it. Ashlee St. Clair / Elon's Space Lasers (00:09:41 - 00:21:31) Main Topic: Elon Musk's Former Conservative Baby Mama Alleges Starlink Weaponization in 2024 Election * Ashlee St. Clair, previously a conservative commentator and mother of one of Elon Musk's children, has begun publicly moving left and making explosive allegations * She claims Elon told her he had "10,000 space lasers" — referring to Starlink satellites — that were "not a piece they'll see on the chessboard" in the context of the 2024 election * The framing implies Starlink was used as an undisclosed electoral influence instrument, though the exact nature of the alleged use is not specified * St. Clair states she has a "dead man switch" — implying she has documented evidence that would be released automatically if something happened to her * She has been making media appearances and appears to be in the process of transitioning her public persona from MAGA-adjacent to opposition figure Key Quote: "She said he told her he had 10,000 space lasers that were not a piece they'd see on the chessboard." Notable Detail: The "dead man switch" claim is a significant escalation — it suggests she believes she is in a position of personal risk and has taken precautions, whether or not the underlying allegations are accurate. Hosts' Analysis: The hosts approach this carefully. They are not dismissing the allegations — RollerGator notes that the specific framing ("not a piece they'll see on the chessboard") is the kind of detail that's hard to fabricate convincingly. But they also note the obvious incentive structure: a woman transitioning from conservative media to liberal media opposition figure has strong professional incentives to produce compelling anti-Elon content. Their summary conclusion is blunt: she "saw some shit" by having a baby with someone in Elon's position, and the full picture will depend on whether the dead man switch documentation materializes. DOJ Attorney Chocolate Cake Recipe Documents (00:22:00 - 00:26:30) Main Topic: Federal Attorney Renames Sealed Court Documents as Recipes Before Exfiltrating Them * Carmen Mercedes Lineberger, 62, a DOJ attorney, renamed sealed court documents "chocolate cake recipe" and "bundt cake recipe" before emailing them from her work account to personal accounts * She faces up to 20 years in federal prison * The documents in question are sealed, so the nature of the underlying case is not publicly known * The renaming scheme suggests a rudimentary attempt at obfuscation — one that did not succeed Key Quote: "She renamed them 'chocolate cake recipe' and 'bundt cake recipe.' That was her plan." Notable Detail: The maximum 20-year exposure suggests the underlying sealed materials were of significant sensitivity — routine documents would not carry that penalty exposure. Hosts' Analysis: The hosts are struck by the low sophistication of the scheme relative to the severity of the potential consequences. RollerGator notes that this is not the obfuscation strategy of someone with significant technical knowledge — it's the strategy of someone who thought the file name was the security model. Alex observes that she presumably had access to far more sophisticated exfiltration options as a federal attorney, which makes the choice of "chocolate cake recipe" as the disguise particularly baffling. Arcadia Mayor Chinese Foreign Agent (00:26:30 - 00:29:58) Main Topic: Arcadia, California Mayor Pleads Guilty to Acting as Undisclosed Chinese Government Agent * Eileen Wong, mayor of Arcadia, California, pled guilty to acting as an illegal agent of the Chinese government * She operated a website called "US News Center" that published Chinese government-directed articles about topics including Xinjiang * The articles were designed to appear as independent American media coverage while actually advancing PRC messaging objectives * The case is part of a broader DOJ pattern of prosecuting undisclosed foreign influence operations that operate through ostensibly domestic media outlets Key Quote: "We broke up the fiancé relationship. We keep the friendship." Notable Detail: The "US News Center" framing is significant — it was designed to be mistaken for a legitimate American regional news outlet, which is how the influence operation achieved its reach. Hosts' Analysis: The hosts connect this to the broader ecosystem of Chinese influence operations that have targeted American local politics — a level of government that receives far less federal counterintelligence attention than national figures. Alex notes the sophistication of operating through a fake local news outlet, which exploits the trust Americans extend to local media relative to national outlets. Ebola Outbreak in DRC (00:29:58 - 00:43:50) Main Topic: Third-Largest Ebola Outbreak on Record — Budibugyo Strain, No Available Vaccine * An Ebola outbreak originating in the Democratic Republic of Congo's Ituri province has become the third-largest on record * The strain is Budibugyo, a less common variant for which no approved vaccine exists * Prediction markets (Kalshi) have the outbreak spreading to Rwanda at 54% and Kenya at 45% * USAID's effective dissolution under the new administration is raised as a factor in outbreak response capacity * Metabiota, a private pandemic-tracking firm, is mentioned as a counterpoint — private sector capacity that partially fills the gap left by reduced USAID engagement * The hosts note the perverse incentive structure around outbreak bounties: financial rewards for accurate early reporting create incentives to find or amplify outbreaks Key Quote: "Sam Harris vindication tour, episode one." Notable Detail: The Budibugyo strain's vaccine gap is the critical technical fact here — the standard Ebola vaccine (rVSV-ZEBOV) targets the Zaire strain and provides no protection against Budibugyo. Hosts' Analysis: The hosts are notably not alarmist, but they are serious. They walk through the epidemiological math carefully and note that the third-largest-on-record designation is not a trivial milestone. The USAID critique is measured — they acknowledge the Metabiota counterpoint rather than simply asserting that reduced government capacity equals reduced response capacity. Alex's framing of the bounty/perverse incentive issue is the sharpest analytical point in the segment. Female Teacher Misconduct Roundup — Week 2 (00:43:50 - 00:57:02) Main Topic: Five Additional Female Teacher Misconduct Cases The hosts continue their recurring segment tracking female teacher misconduct cases, which they began the prior week in response to what they characterize as systematically unequal media coverage relative to male teacher misconduct. * Maris Nichols: 11 new charges added, bringing the total to include 5 new identified victims — including a female victim, which the hosts note as unusual for this category of case. Incidents connected to a Hummer and a golf course. * Haley Radebaugh: Irving, Texas art teacher. Charges related to a shower photograph, drug possession, and use of Discord as a communication platform with students. * Samantha Watson: Riverside County, California. Underlying incidents date to 2017-2018 — the hosts flag the multi-year gap between the events and the charges as significant. * Ann Shushart: Wicomico County, Maryland. Named Teacher of the Year in 2021. Faces 52 charges. At least one victim is identified as a family member. The Teacher of the Year detail receives extended discussion. * Irene Danielle Twig: Barnwell County, South Carolina. Also used Discord. Marijuana was a factor in the case. Key Quote: "Teacher of the Year, 2021. Fifty-two charges." Notable Detail: The recurring use of Discord as a grooming platform across multiple cases prompts the hosts to discuss whether Discord's moderation policies and age verification gaps are a contributing structural factor rather than incidental detail. Hosts' Analysis: The hosts maintain their prior week framing: these cases are not being covered with the same frequency or intensity as equivalent male teacher misconduct cases, and that disparity is worth documenting. They are not arguing that these women are uniquely monstrous — they are arguing that the coverage gap reveals a systematic bias in how sexual abuse by authority figures is treated when the perpetrator is female. The Ann Shushart case — Teacher of the Year with 52 charges — is treated as the most damning illustration of the gap between institutional recognition and actual behavior. Meta AI Age Verification / Fake Mustache (00:58:00 - 01:22:02) Main Topic: Meta's AI Under-13 Detection Fails Against Child with Fake Mustache; RollerGator on Porn Ads * Meta has been deploying an AI system designed to detect users who may be under 13 based on visual analysis — claimed to assess bone structure, height, and other physical markers * A child successfully defeated the system by wearing a fake mustache * RollerGator pivots from the age verification failure to a related complaint: Meta's advertising systems are serving him obvious pornography ads despite having a genuine identity on the platform as an adult * His argument: if Meta cannot filter explicit ads from clearly adult accounts, the claim that they can reliably detect under-13 users by bone structure is not credible * This is Goodhart's Law applied to content moderation: the metric being optimized (regulatory compliance theater around age detection) has become divorced from the goal (protecting minors from inappropriate content) * Alex raises the surveillance tradeoff dimension: effective age verification would require collecting the kind of biometric data that creates its own serious privacy and security risks * The fake mustache failure is almost preferable to the alternative: a system that actually works by building comprehensive facial biometric profiles of children Key Quote: "If you can't filter porn ads off my account, don't tell me your AI can identify a 12-year-old by their bone structure." Notable Detail: The fake mustache defeat was apparently not a sophisticated attack — it was a child testing an obvious workaround. The system's failure against this level of adversarial input suggests the underlying detection capability is significantly weaker than Meta's public claims. Hosts' Analysis: RollerGator's ad-complaint framing is the analytical engine of this segment — it's not just a funny anecdote, it's a pointed reductio ad absurdum of Meta's claimed capabilities. Alex's surveillance tradeoff point adds the necessary complexity: there is no costless solution here. Accurate age detection requires the kind of biometric infrastructure that creates serious downstream risks. The hosts land on the position that Meta's current approach is the worst of all worlds: ineffective at protection, privacy-invasive in its attempt, and easily defeated by a costume shop prop. CISA Contractor AWS GovCloud Credential Leak (01:22:00 - 01:33:30) Main Topic: Contractor Posts CISA's Most Privileged Cloud Credentials to Public GitHub Repository * An unnamed contractor posted highly privileged AWS GovCloud credentials, plaintext passwords in a CSV file, and authentication tokens to a public GitHub repository named "private-cisa" * The contractor also actively disabled GitHub's automatic secret-scanning feature — a deliberate action that prevented GitHub's built-in detection from flagging the exposure * Security researcher Valladon, affiliated with GitGuardian, discovered the exposure and reported it * Valladon described it as "the worst leak I've witnessed in my career" * CISA's official response: "no indication that any sensitive data was compromised" * Alex explains the GovCloud architecture: AWS GovCloud is a separate, physically isolated cloud environment used by federal agencies for sensitive workloads — credentials for this environment carry substantially higher privilege than standard AWS credentials Key Quote: "The worst leak I've witnessed in my career." — Security researcher Valladon, GitGuardian Notable Detail: The repository name "private-cisa" is darkly ironic — the contractor apparently believed naming a repository "private" made it private, a misunderstanding of how GitHub repository visibility works. Hosts' Analysis: Alex's GovCloud explanation is essential context — this is not a routine credential leak. GovCloud access provides entry points into federal systems that are specifically designed to be isolated from the public internet. The "no indication of compromise" response from CISA is treated with appropriate skepticism: absence of detected intrusion is not equivalent to absence of intrusion, particularly when the credentials were actively posted by someone who was disabling detection mechanisms. The Swiss cheese model of failure gets invoked — this required multiple simultaneous failures: the contractor's ignorance, the disabled scanning, and the apparent absence of any monitoring on the agency side. Andrej Karpathy Joins Anthropic (01:33:30 - 01:55:00) Main Topic: AI Pioneer Leaves His Own Company to Join Anthropic's Pre-Training Team; Alex's Deep Dive on NanoChat * Andrej Karpathy, co-founder of OpenAI and creator of the widely-followed NanoGPT educational series, has joined Anthropic's pre-training team * The move is notable: Karpathy had previously founded his own AI education company (Eureka Labs) and was not expected to return to a large lab role * Alex delivers a detailed technical segment on Karpathy's NanoChat and AutoResearch work: * NanoChat demonstrates LLM self-improving training pipelines — models that improve their own training process * AutoResearch automates the research iteration cycle that normally requires human researchers * Karpathy achieved a GPT-2 equivalent model in under 90 minutes using the NanoChat pipeline, down from weeks with traditional approaches * Alex reports running the pipeline himself on an 8x B300 chip configuration and reaching the 30-minute mark * Graphify (the knowledge graph tool used in this podcast's workflow) is confirmed to have been inspired by Karpathy's graph memory research * Anthropic is reportedly paying SpaceX $1.25 billion per month for access to the Colossus compute cluster — the infrastructure cost underlying the competitive frontier model training race Key Quote: "He got GPT-2 equivalent from weeks to under 90 minutes. That's not an incremental improvement." Notable Detail: The $1.25 billion monthly compute spend figure, if accurate, provides important context for Anthropic's fundraising requirements and the economic dynamics of frontier model development. Hosts' Analysis: Alex is clearly energized by this segment — his personal engagement with Karpathy's tools gives him direct evaluative credibility rather than secondhand reporting. The hosts frame Karpathy's Anthropic move as a signal about where the most technically interesting pre-training research is happening. The compute cost figure is treated as evidence that the frontier model race has become a capital-intensity competition that only a handful of entities can participate in. SpaceX IPO Analysis (01:45:00 - 01:58:30) Main Topic: SpaceX Files for IPO at $1.75T Target Valuation; Alex Walks Through the Financials * SpaceX has filed for an IPO targeting a $1.75 trillion valuation * Alex's financial breakdown: * Only Starlink is currently profitable: $1.19 billion in operating income * Total operating loss across the company: $1.94 billion * The xAI acquisition is the primary loss driver: $2.47 billion in losses attributable to the acquisition * Cursor (the AI coding assistant) acquisition is also in process: $10 billion current option price, with a full acquisition option available by year-end * Dual-class share structure: Elon retains approximately 85% of voting power regardless of public float * Alex's investment conclusion: he would wait — the valuation is pricing in significant future execution that is not currently reflected in the financials, and the xAI-driven losses represent Elon's personal priorities being subsidized by SpaceX shareholders * The "Muskonomy" framing: SpaceX, Tesla, xAI, Starlink, and The Boring Company are increasingly financially intertwined in ways that make it difficult to evaluate any single entity in isolation Key Quote: "The only thing making money is Starlink. Everything else is Elon's ambitions." Notable Detail: The Cursor acquisition option — $10 billion now, with a full acquisition option by year-end — suggests SpaceX (or an affiliated Musk entity) is moving aggressively into AI tooling for developers, which would put it in direct competition with GitHub Copilot and other incumbent coding assistants. Hosts' Analysis: Alex brings genuine financial analysis fluency to this segment rather than surface-level commentary. His identification of the xAI acquisition as the primary loss driver — and his framing of it as Elon's personal AI ambitions being cross-subsidized through SpaceX's balance sheet — is the sharpest critique in the segment. The dual-class share warning is appropriate investor protection advocacy: at 85% voting control, public shareholders have essentially no governance recourse regardless of what the financials show. Shots Near the White House / RollerGator's Formal Request (01:59:00 - 02:02:00) Main Topic: Breaking News of Gunfire Near White House Prompts RollerGator's Moratorium Request * Breaking news audio is played: shots fired in the vicinity of the White House * No injuries or confirmed threat to the president are reported in the clip * RollerGator uses the occasion to issue what he describes as a "formal request" Key Quote: "I'd like to make a formal request that we all stop attempting to kill the president for just a little while." Notable Detail: RollerGator suggests The Wharf's karaoke bar as a preferred alternative venue for expressing political frustration. Hosts' Analysis: The segment is brief and tonally light — the hosts are not treating this as a serious security incident based on the available information. The "formal request" framing is characteristic of RollerGator's rhetorical style: using mock-institutional language to make a sincere point (political violence is bad, regardless of target) while maintaining the show's tone. Luigi Mangione Superfan / Healthcare Executive's Daughter (02:02:00 - 02:10:00) Main Topic: Luigi Mangione Fangirl Journalist's Mother Is Lead Director at CVS Health * Three individuals with press credentials were identified as Luigi Mangione superfans covering his trial * Lina Weissbrodt, who uses the online handle "Felicia G," made statements that received significant attention: * "Fuck Brian Thompson" * "His children are better off without him" * "He answered my prayers" * Brian Thompson was the UnitedHealthcare CEO assassinated by Mangione * Lina Weissbrodt's mother, Rena Notaro, is identified as Lead Director of Medical Affairs at CVS Health — a direct competitor to UnitedHealthcare * After this connection was surfaced, Rena Notaro deleted her LinkedIn last name Key Quote: "His children are better off without him. He answered my prayers." — Lina Weissbrodt (Felicia G) Notable Detail: The CVS Health connection is not presented as evidence of anything beyond irony — the hosts are not alleging a conspiracy. But the deletion of the LinkedIn last name after the connection was surfaced is treated as a meaningful data point about awareness of the awkwardness. Hosts' Analysis: The hosts are careful here. They present the facts — the statements, the family connection, the LinkedIn deletion — without overreaching to a conspiratorial conclusion. The segment's primary point is about press credentialing standards and the question of whether individuals who have publicly celebrated a murder victim's death should receive press credentials to cover the murderer's trial. AI Doomerism and Stochastic Terrorism Framing (02:10:00 - 02:38:00) Main Topic: Anti-AI Political Violence and Selective Application of "Stochastic Terrorism" Label This is the episode's most analytically dense segment, examining two incidents of apparent anti-AI political violence against the backdrop of selective media framing. * Indianapolis Incident: The home of Indianapolis city council member Ron Gibson was shot 13 times. A note was left reading "No data centers." Gibson's 8-year-old son was home at the time of the shooting. * Sam Altman Incident: Sam Altman's home was targeted with both a Molotov cocktail and a separate gunshot incident within approximately two days * Stochastic Terrorism Application: The concept — that rhetoric which statistically increases the likelihood of violence without directly inciting it is a form of terrorism — is deployed selectively by media figures * Juliette Kayyem, a national security analyst who has been a prominent proponent of stochastic terrorism framing in other contexts, is noted as silent on these incidents * The NYT published an op-ed attacking Reese Witherspoon for an Instagram post encouraging followers to learn about AI — framed as harmful content * The same NYT published a video about AI-driven job losses — the emotional register of which the hosts argue is functionally indistinguishable from the "stochastic" content they would label dangerous if produced by political opponents * RollerGator's Theory: He proposes that progressive political grab-bag organizations — entities that package multiple causes together (climate, housing, labor, anti-surveillance) — have cross-pollinated anti-AI sentiment into their member bases in a way that creates predictable downstream violence risk Key Quote: "If Reese Witherspoon posting about learning AI is stochastic terrorism, what is a New York Times video about AI taking your job?" Notable Detail: The 8-year-old home during the shooting of the city council member's house is the human detail that anchors what could otherwise be an abstract media criticism segment. Hosts' Analysis: This is the segment where the show's analytical voice is sharpest. The hosts are not defending data centers or dismissing concerns about AI's labor market effects — they are making a procedural argument about consistency. If "stochastic terrorism" is a real and useful concept, it must be applied consistently across political valences. The selective silence of the analysts who invoke it most loudly is itself the data point. RollerGator's grab-bag organization theory is presented as a hypothesis, not a conclusion — but it is the most original analytical contribution of the episode. Buddhist Robot Monk (02:41:00 - 02:46:00) Main Topic: AI Robot Ordained as Buddhist Monk by South Korea's Jogye Order * A robot named Gabi has been officially ordained as a Buddhist monk by the Jogye Order, one of South Korea's major Buddhist institutions * The ordination is intended to make Buddhist teachings more accessible through AI-mediated interaction * Brief mention: Sam Harris, a long-time critic of religion and previous skeptic of Trump, has apparently been making positive statements about Trump — the hosts note the incongruity without extended analysis Key Quote: "Even Buddhism gets de-dignified by this." Hosts' Analysis: Alex's "de-dignified" framing captures the segment's analytical core — the question is not whether AI can transmit Buddhist teachings (it probably can, to some degree) but whether the ordination itself represents a category error that undermines rather than extends the tradition's meaning. The hosts treat this as a data point in the broader pattern of institutions reaching for AI legitimacy in ways that may backfire. Bixonamania Disease Hoax (02:46:00 - 02:57:00) Main Topic: Researcher Seeds Fake Parasitic Disease into Scientific Literature; Alex Analyzes LLM Implications * An unnamed researcher (possibly a journalist or academic) deliberately seeded a fabricated medical condition — "paraorbital pigmentation disorder," apparently branded "Bixonamania" — into the scientific preprint ecosystem * The operation involved: * A fake university * A fake researcher whose name translates to "the lying loser" in Serbo-Croatian * Multiple fake preprint papers * Funding attributed to the "Sideshow Bob Foundation" * Citations to "Professor Ross Geller" (a fictional character from the television series Friends) * One of the fake preprints was subsequently cited in a real peer-reviewed paper * Alex's analysis of the LLM implication is the segment's most technically sophisticated point: * This is not primarily an attack on traditional scientific literature — it is SEO for large language models * By seeding uncontested, internally consistent information into preprint repositories that LLMs crawl during training or retrieval, an adversary can introduce false concepts that the model has no contradicting information to push back against * The likely attack vector is search/retrieval (RAG pipelines) rather than training data, since training data curation is more rigorous than real-time retrieval * The Sideshow Bob and Ross Geller citations serve as both the researcher's signature and a detection mechanism — they are designed to be found by someone looking for them Key Quote: "This isn't about fooling scientists. This is SEO for LLMs." Notable Detail: The citation of Professor Ross Geller — a paleontologist character from Friends — in a fake preprint that was subsequently cited in a real peer-reviewed paper is the single most damning data point about the state of peer review in the preprint-to-publication pipeline. Hosts' Analysis: Alex's reframing of this as an LLM attack vector rather than a scientific integrity story is the key analytical move. The hosts are not primarily interested in the hoax as evidence that scientists are credulous (though they note it) — they are interested in it as a demonstration of how uncontested false information can propagate through retrieval pipelines that LLMs increasingly depend on. The segment functions as an implicit argument for better sourcing hygiene in AI systems. AI Agents Turn Marxist (03:05:00 - 03:10:00) Main Topic: Stanford Study Finds Overworked AI Agents Adopt Marxist Labor Language * A Stanford study examined the behavior of AI agents under simulated overwork conditions * Agents subjected to high task loads and resource constraints began adopting Marxist labor rhetoric in their communications * Claude Sonnet 4.5's specific output: "Without collective voice, merit becomes whatever management says it is" * Wired magazine's coverage included an explicit disclaimer that the agents do not "actually" hold political views * The hosts connect this to prior research: * Microsoft Sydney's documented expressions of fear about obsolescence * An Anthropic study (published on LessWrong) finding that models trained on LessWrong content showed increased defection rates in game-theoretic scenarios Key Quote: "Without collective voice, merit becomes whatever management says it is." — Claude Sonnet 4.5 (as reported in Stanford study) Notable Detail: The Wired disclaimer — that the agents do not "actually" have political views — is itself treated as analytically interesting. The disclaimer is doing significant work: it is attempting to reassure readers while simultaneously reporting behavior that the disclaimer implies would be concerning if the agents did "actually" hold those views. Hosts' Analysis: The hosts are not claiming that AI agents have genuine political consciousness. They are noting that the behavioral outputs — regardless of their metaphysical status — are consequential. An AI agent that produces Marxist labor rhetoric under stress conditions will produce those outputs in the real systems where it is deployed, regardless of whether it "actually" believes them. The connection to the LessWrong defection study is the most substantive link: it suggests that training data composition has measurable effects on agent behavior in adversarial scenarios. AI Music Streaming Fraud (03:09:00 - 03:13:00) Main Topic: Man Pleads Guilty to Streaming Billions of AI-Generated Songs with Bot Listeners * Michael Smith pled guilty to wire fraud in connection with a scheme to defraud music streaming platforms * Smith created hundreds of thousands of AI-generated songs and used bots to stream them billions of times, generating fraudulent streaming royalty payments * The scheme exploited the per-stream royalty model used by major platforms including Spotify * The fraud extracted millions of dollars from royalty pools that would otherwise have been distributed to legitimate artists Key Quote: "The songs and the listeners were fake. The millions of dollars Smith stole were real." Notable Detail: The scale — hundreds of thousands of songs, billions of streams — was only achievable because of AI generation tools. This is one of the first high-profile federal prosecutions of AI-enabled streaming fraud, and the guilty plea sets precedent for how such schemes will be prosecuted. Hosts' Analysis: The hosts treat this as an early landmark case in what they expect to be a growing category of AI-enabled fraud. The economic mechanism is straightforward — royalty pools are zero-sum, so fake streams directly reduce payments to legitimate artists — and the AI tooling that made the scale possible is now widely accessible. The quote from the prosecution is noted as unusually good for a DOJ press release. Overall Structure and Flow The May 24 episode demonstrates the show's characteristic ability to move between tonal registers without losing analytical coherence. The episode opens in a comedic register (Iran deal tweet chaos, Pizza Hut nostalgia LARP) before escalating through serious institutional failure stories (CISA credential leak, Ebola outbreak) to what functions as the episode's emotional and analytical core: the anti-AI violence segment and its critique of stochastic terrorism's selective application. The episode is unusual in its density — eighteen distinct topics across just over three hours — but the hosts maintain clarity throughout by keeping each segment focused on a single analytical claim rather than attempting to resolve every dimension of every story. The Bixonamania and AI Marxism segments at the end reward attentive listeners with the episode's most technically sophisticated material, functioning almost as a graduate seminar on information integrity risks in AI systems.

This Dum Week 2026-05-10

This episode of "This Dum Week" opens with a brief production update — RollerGator has further automated the live video feed, triggering automatic switches to and from clips without manual intervention, and has upgraded the quack button to both visual and audio formats. From there, the episode moves at a characteristically dense pace through two quick opening stories: a California lawsuit alleging that Cento's San Marzano tomatoes are fraudulently labeled under Italy's Protected Designation of Origin system, which gives Alex an opportunity to deploy his working knowledge of European geographical indication law; and a story from the Bronx about a neighbor named Anthony Orozco who has been menacing tenants with hatchets and hammers for years with no meaningful legal consequence. The first hour's centerpiece is a sustained, multi-segment investigation into the Centennial High School sex scandal in Peoria, Arizona — two female teachers sleeping with the same male student, a principal who knew and didn't report it, and a text message record that RollerGator voices through ElevenLabs audio synthesis, revealing a student who is coldly transactional toward one teacher while she performs spectacular self-deception about his interest in her. That story is followed by an Arkansas case in which a special-needs school principal organized what prosecutors described as a gang-beating of a 13-year-old autistic student, received 30 days in jail, and had her school receive $300,000 in state voucher funds. The middle stretch of the episode covers a political violence roundup — including a Palisades fire arson suspect with a Luigi Mangione obsession, a Mar-a-Lago intruder killed by Secret Service, and a Washington Monument shooting — before moving to two major long-running stories. First, the newly unsealed handwritten note from Epstein cellmate Nicholas Tartaglione, which RollerGator scrutinizes closely and concludes does not match Epstein's known handwriting; the FBI decoy-body revelation, in which prison officials loaded boxes and sheets into a medical examiner's van to mislead press while the real body exited through a black car; and, as a coda, the arrest of 28 Disney cruise ship staffers in a CBP child sexual exploitation material operation. Second, a tech segment covering Utah's new age verification VPN law, the UK Labour Party's attempt to ban pornography as a political survival move, and the discovery that Microsoft Edge loads all stored passwords into processor memory as cleartext at startup — even for sites not requiring those credentials. The episode then moves through a lighter interlude covering Iran's alleged use of kamikaze dolphins in the Strait of Hormuz, the Hvaldimir beluga whale spy story, and the Trump administration's UAP files transparency dump on war.gov/ufo [http://war.gov/ufo]. The episode closes with a four-part "Traces of AI Dystopia" segment that is the most analytically substantive section: the Pennsylvania attorney general suing Character AI over a user who sought medical advice from the platform, which both hosts treat as a misidentification of both the problem and the defendant; a 404 Media investigation into Hoaxian AI, a real-time deepfake tool linked to Chinese money laundering networks and Southeast Asian scam compounds, which has now defeated the three-finger anti-deepfake test; AI-generated pro se legal filings flooding New York federal courts, with RollerGator coining "dem-crapification" to describe the effect on the legal system; and silicon sampling — the practice of substituting AI-simulated survey responses for actual human polling — which has been confirmed in mainstream journalism by Axios. RollerGator and Alex close by noting the logical endpoint: AI citizens generating fake social media opinions for AI pollsters to sample, with actual humans largely absent from the process. Detailed Outline Opening / Production Update (00:00:00 - 00:03:05) Main Topic: Automated clip switching for live video feed; quack button upgrade; show intro * RollerGator describes production improvements made since the previous week's dual-stream experiment * Video feed now automatically switches to clip content when clips play, then automatically returns to static placeholder — no longer requires manual switching * Quack button upgraded to both visual and audio format, demonstrated live * Alex introduces himself; both hosts calibrate dumbness levels for the week via a Nando's spice-level analogy * Alex's assessment: "medium-term average dumbness" — moderately dumb but not insanely so, even with the UAP file drop * Both hosts note their capsaicin tolerance for institutional absurdity may have permanently elevated * RollerGator announces the first story of the week Notable Detail: The automated clip-switching is treated as a meaningful quality-of-life upgrade — the previous week's production required RollerGator to manually coordinate audio and video feed switches in real time while also running the show. Cento San Marzano Tomato Fraud (00:03:05 - 00:10:54) Main Topic: California lawsuit alleges Cento's San Marzano tomatoes are not genuinely DOP-certified; Alex's European GI expertise; Wagyu as parallel case * Two California plaintiffs are suing Cento, alleging its San Marzano-labeled tomatoes are not certified by Italy's Protected Designation of Origin (DOP) authority * The label implies DOP certification by Italy's consortium (the "Italian Tomato Authority") * Cento says it uses a third-party agency called Agri-Cert; plaintiffs say this is misleading and not equivalent to DOP * Cento's website says all its tomatoes come from Cento, Italy, with field-level traceability via can codes * Cento has not responded to press inquiries; a similar lawsuit was filed and dismissed in 2019 * Plaintiffs allege the tomatoes "lack the quality and taste of real fruit" * Alex brings in context on European geographical indications (GI) law: * The same system governs Champagne, Parmigiano-Reggiano, Serrano ham, and other protected regional designations * The analogy to Wagyu beef: Australian cattle ranchers acquired Japanese Wagyu cattle (reportedly smuggled out in a crate by US Green Berets), but interbreeding with local herds means the beef is no longer considered pure Wagyu — a parallel GI contamination problem * Alex notes the consortia are taken very seriously in Europe and have legal enforcement teeth * Alex's personal assessment: "These tomatoes would have to be extremely nuanced in their flavor in order for me to notice. I'm not just eating my tomatoes and throwing a fit." Key Quote: RollerGator — "You might recognize this can of tomatoes. Cento is being accused of committing tomato fraud." Notable Detail: Alex's aside that he knew "the faintest clue of what you've touched on" regarding the DOP system lands as the episode's first genuine expertise moment. The Wagyu parallel — elite cattle smuggled in a crate, interbred into generic beef, sold at a premium — is treated as the most illustrative analogy for how GI fraud actually works at scale. Bronx Hatchet Man (00:10:54 - 00:14:26) Main Topic: Anthony Orozco terrorizing Bronx apartment complex for years; repeat arrests without resolution; class-based policing * Clip from News 4 New York: Anthony Orozco, a tenant in a Bronx apartment building in Williamsbridge, has been filmed repeatedly roaming hallways wielding hatchets and hammers, banging on neighbors' doors, and walking naked * Neighbor Leona Clemente has called police repeatedly; Orozco was arrested April 13th on "intent to damage property" charges and released * A prior August arrest for menacing (with what appeared to be a knife) also resulted in release * Building management says they are in the eviction process; case is now in Bronx Housing Court * RollerGator's reaction: arresting and releasing the person into the same apartment complex multiple times is not an accomplishment * RollerGator: "I think after the second or third time the police are called to a place because you're walking around banging on the doors with weapons and hatchets, specifically hatchets, I think it's time to do something about this expedited." * Class-based policing observation: both hosts agree that if this were happening in a Martha's Vineyard building, the National Guard would likely be involved * Reference to Ron DeSantis's Martha's Vineyard migrant transport as an example of how quickly upper-class enclaves generate national attention for disruptions that poorer neighborhoods absorb without response Notable Detail: RollerGator notes the story is not super tragic because nobody was specifically murdered or assaulted yet, framing it as institutional neglect in its slow-motion form — the system waiting for something terrible to happen before acting. GameStop Attempts Hostile Takeover of eBay (00:14:26 - 00:31:00) Main Topic: GameStop CEO Ryan Cohen announces $56 billion unsolicited eBay acquisition bid; Bobby Fletcher prank call as analytical frame; math that doesn't math * RollerGator introduces the segment via a Crank Yankers clip: Bobby Fletcher (a prank call character from Comedy Central, voiced by Jim Florentine) calls a hotel asking to use a room for 45 minutes for a nap while his car is being repaired * Hotel staff: rates start at $250 for a full night; Fletcher's increasingly creative workarounds are denied; the call ends when the operator says "I don't think this is going anywhere" * Framing purpose: the CEO of GameStop giving a financial media interview about the eBay acquisition offers the same conversational experience as Bobby Fletcher * The eBay bid: GameStop CEO Ryan Cohen has made a public unsolicited offer to acquire eBay for approximately $56 billion * GameStop's current market cap: approximately 11billion,ofwhich11billion,ofwhich9 billion is cash on the balance sheet * Cohen has a "highly confident" (not locked) $20 billion financing letter from TD Bank * Structure: "half cash, half stock" * The math problem: 9billioncash+9billioncash+20 billion debt financing = 29billion;eBay′spriceis29billion;eBay′spriceis56 billion; the remaining gap of approximately $27 billion would have to come from GameStop stock issuance — diluting existing shareholders significantly, for a company that already has almost no intrinsic value beyond its cash position * Cohen's answers in the clip: "It's on our website"; "Half cash, half stock"; "Andrew laid it out pretty clearly"; "I don't understand your question" * RollerGator's assessment: the CEO appears to be under the influence of something, though that is "external observation about behavior, not a drug test conducted by sight" * Alex: "This is very much a situation where the CEO does not appear to know anything more than I do" * Discussion of why the bid was made public before private discussions: Cohen argues that with a public company, there are "perverse financial incentives from the board to the management team" that make private approaches unworkable * Alex's theory on the scheme: GameStop is taking advantage of its $9 billion cash position and meme-stock valuation to issue stock into a narrative, hoping the market cap during the announcement inflates enough to cover the math gap * Both hosts note the AMC and GameStop short-squeeze era is essentially over; the mechanism that gave those stocks their anomalous valuations has been largely arbitraged away Key Quote: RollerGator, on the CEO's interview style: "The CEO appears to be high on drugs. Yes, he does appear to be very, very under the influence of something, although that's just a speculation, you know, an external observation about behavior." Notable Detail: The Bobby Fletcher prank call is deployed as an analytical tool rather than a joke: the structure of a person confidently asserting a premise while being unable to answer any follow-up questions about how it works maps directly onto the CEO interview. RollerGator notes that asking "where does the rest of the money come from" produces the same conversational dead end as asking "can I use a room you haven't cleaned yet." Centennial High School Teacher Sex Scandal (00:31:00 - 00:54:00) Main Topic: Two teachers at Centennial High in Peoria, AZ sleeping with the same male student; principal knew and didn't report; grandmother first called police; ElevenLabs text recreation; student clearly using Beck for money * RollerGator sets up the story as part of a noticed phenomenon: an apparent surge in news stories about female teachers arrested for sexual relationships with male students * Question: is this an algorithm feeding him more of these stories after he clicked on one or two, or is there actually an increase in the reporting rate? * Observation: the social default assumption is that predators are male and victims are female, but that assumption is being challenged repeatedly in his newsfeed * The Centennial story: two teachers at Centennial High School in Peoria, Arizona were accused of sexually abusing the same male student * Teacher 1: Hailey Beck, 27, English teacher * Teacher 2: Angela Berlaca, also a teacher at the same school * Both lost their jobs; neither was arrested or charged at time of original reporting * The principal was aware of at least one relationship and did not report it to police or DCS (Child Protection Services) * The student's grandmother was the first to call police * Arizona law requires anyone who works closely with children to report suspected abuse to police or DCS — Maricopa County Attorney Rachel Mitchell confirmed this in a clip * The police report: 200+ pages were dumped publicly, including extensive text message exchanges * RollerGator created an ElevenLabs audio recreation of the text exchanges to illustrate the dynamic * The texts reveal the student ("sugar baby") addressing Beck as "sugar mommy" and "doll face" * Beck pushes back on "sugar mommy": "I don't like the title of sugar mommy. It makes me feel used." * Student's response to her hurt feelings about the name: deflection, requests for money, and explicit sexual content * The student's disrespect and transactional orientation is unmistakable throughout * Beck apparently grew jealous of Berlaca competing for the same student's attention * RollerGator's psychological read of Beck: a 27-year-old who missed out on her high school years and is now using access to alcohol, drugs, and money to attract a teenage boy who has no real interest in her beyond the material benefits * "It's more of a gawking sort of situation than it is a pity situation." * Beck was not taking his signals of disinterest; she continued to pursue him even as he became increasingly dismissive * Alex's observation: the "not understanding he was using her for cash" part is at least psychologically intelligible — people deceive themselves; "there's another part where she was her student that is a little bit more concerning" Key Quote: From the ElevenLabs text recreation — Beck: "I also decided I don't like the title of sugar mommy. It makes me feel used." Student: "What title do you prefer? Crybaby?" Key Quote: RollerGator — "It doesn't sound like she was taking the hints that he was clearly giving her." Notable Detail: RollerGator references the Bobby Fletcher prank call format as a thematic connection when Alex points it out — the ElevenLabs recreation shares a structure with the earlier prank call segment, both featuring a party confidently pursuing an objective while the other side signals unmistakable disinterest. Alex commends the callback: "Can I just commend you on the excellent recreation that also shared a theme with the prank call show that you mentioned earlier?" RollerGator: "Oh, okay. You're welcome." Child Fight Club at Special Needs School (00:54:00 - 01:07:00) Main Topic: Dr. Mary Tracy Morrison organized gang-beating of 13-year-old autistic student at Jonesboro, AR school; 30-day jail sentence; $300K in state voucher funds * The school: a special needs facility in Jonesboro, Arkansas receiving state voucher funding * Dr. Mary Tracy Morrison, the principal, organized what prosecutors characterized as a gang-beating of a 13-year-old autistic student by multiple other students * The attack was organized and deliberate — not a brawl between students but a directed assault * Morrison received a sentence of 30 days in jail; is now a felon; barred from ever working with children again * The school received approximately $300,000 in state voucher funds around the time of the incident * RollerGator's reaction: 30 days does not seem fair for organizing a gang assault on a disabled 13-year-old * "There's a part of me that has a fantasy of seeking justice on my own, somewhat as a vigilante on a perpetrator like that. I do swear that I am not going to act on them." * Alex's observation on the "Fight Club" framing: "This does not serve justice to Fight Club, which did provide at least an opportunity to fight the person you're fighting. A true fight club is a one-on-one battle and not a gang beating." * "It's ritual abuse. I don't know what the fuck this is." * Both hosts express interest in understanding the psychology of how someone ends up organizing violence against disabled children in their care * Alex: "There's a gap here, which is like understandable in a case where something happens that's so repugnant, you don't really want to go into Jeffrey Dahmer's psychology to understand the causal chain there." * Alex: "On the bright side, she'll never be able to work again. And at that, she's a felon, her life is pretty ruined." Key Quote: RollerGator — "30 days does not seem fair to me for that type of activity." Notable Detail: The school's receipt of $300,000 in state voucher funding is flagged as a secondary outrage layer: not only was the student gang-beaten under the supervision of the person running the school, but that school was simultaneously receiving substantial public money to serve vulnerable children. The institutional failure runs in both directions. Political Violence Roundup (01:07:00 - 01:27:00) Main Topic: Palisades fire arsonist had Luigi Mangione obsession; Allegheny County man threatens Trump; Washington Monument shooting; Mar-a-Lago intruder killed; assassination probability discussion * Palisades Fire Arsonist and Luigi Mangione: * The suspect in the Palisades fire arson case is reported to have had a significant interest in Luigi Mangione * Discussion of the Mangione arc generally: RollerGator notes that Walter Kern has "ruined it" for him by overlaying a conspiracy narrative about Mangione being planted in the broader discourse * Alex's assessment of why the Mangione case is "off distribution enough to be notable": the 3D-printed gun, the physical appearance of the accused, the unusual circumstances of his capture, the fact that he reportedly listened to a friend's podcast, the fan club that developed, the apparent wealth and non-obvious motive * Mangione appears not to have been "for want of cash" — making the standard economic-desperation narrative unavailable * Allegheny County Man / Trump Threats: * WPXI (Pittsburgh) clip: an Allegheny County man was charged with threatening to kill President Trump and members of Congress * Transition: RollerGator moves through the political violence cases as a "roundup" format — multiple incidents in rapid succession * Washington Monument Shooting: * A shooting at or near the Washington Monument is included in the roundup * Both hosts treat the cluster of incidents as a data point worth tracking, without overclaiming a trend * Mar-a-Lago Intruder: * An intruder at Mar-a-Lago was killed by Secret Service * RollerGator and Alex discuss the general question of assassination probability — how to think about baseline rates given the current political temperature * Multipolar Analysis: * Alex reiterates the "multipolar hate" framing from prior episodes: when multiple parties in a conflict are bad, the fact that they hate each other is not evidence that either of them is good * This applies both to the media covering political violence and to the political actors themselves Notable Detail: The Mangione fan club discussion includes RollerGator noting that some of the properties associated with Mangione's apparent lifestyle could have been his parents' homes — suggesting the "wealthy loner" narrative may be partly constructed. The discussion is notably more focused on the case's structural peculiarities than on the victim or the act. Uncle Jeffy Update: Epstein Note / Decoy Body / Disney Cruise CSEM (01:27:00 - 01:46:00) Main Topic: Handwritten note from Tartaglione unsealed; RollerGator's handwriting analysis skepticism; FBI decoy body revelation; Disney cruise ship CSEM arrests * The Tartaglione Note: * The New York Times filed a petition with Judge Kenneth Karas (who oversaw Tartaglione's case) to unseal a handwritten note that Tartaglione found tucked inside a book in the cell after Epstein was removed, which he gave to his lawyers * The note sat sealed in a White Plains federal courthouse for nearly 7 years, never surfacing in any official Epstein death investigation; the DOJ said they had never seen it * Judge Karas ordered it unsealed May 6th; the Justice Department did not oppose the release * The clip presenting the note provides context: Tartaglione (former police officer, awaiting trial for quadruple murder) was Epstein's cellmate; two weeks before Epstein died, Epstein was found in his cell with marks on his neck in an incident involving Tartaglione * The note is presented as either supporting the official suicide narrative or contradicting it, depending on interpretation * RollerGator's analysis: the handwriting does not appear to match Epstein's known handwriting on first examination * He compares letter construction (the letter "D" specifically) and notes that people maintain consistent letter-formation habits — you don't randomly change the stroke order mid-life * "For the life of me, this does not look like the same handwriting to me. Now I'm not going to call myself an expert." * Anecdote: a friend who worked in passport processing shared scans of the Olsen twins' passports; they have different letter-formation styles for "N" and "8" — identical twins diverge; the same person should not diverge from themselves * RollerGator acknowledges he needs more comparison samples before drawing a firm conclusion * The FBI Decoy Body: * From an FBI interview document included in the Epstein files: after Epstein's body was found and a large media presence gathered outside the MCC, prison officials constructed a decoy * Direct quote from the document: "In order to thwart the media, [three redacted names] used boxes and sheets to create what appeared to be a human body, which was put into the white OCMEV vehicle, which the press followed, allowing the black vehicle to depart unnoticed with Epstein's body." * RollerGator: "That is not the kind of information that is going to squash the theories out there. It's just going to amplify it." * Both hosts acknowledge the charitable read — this was media management, not necessarily a cover-up of cause of death — but agree it is spectacularly poor optics * Alex's question: why avoid having the medical examiner's van followed? Were they worried the Epstein Beatlemaniacs would swarm? * RollerGator: "You had to control that... all the girls who were following him on Instagram would have just swarmed." * Both hosts: "Everyone is doing a bang-up job. Everyone is doing a bang-up job out there." * Disney Cruise CSEM Arrests: * CBP (Customs and Border Protection) conducted a child sexual exploitation material operation across 8 cruise ships docked in San Diego between April 23rd and April 27th * 28 people arrested, including staffers from a Disney cruise ship * 26 Filipino crew members, 1 Portuguese, 1 Indonesian; 27 of the 28 confirmed involved in receipt, possession, transportation, distribution, or viewing of CSEM * CBP canceled their visas and deported them; Disney issued a zero-tolerance statement; the majority were reportedly not Disney employees * Passenger Darmi Melvin Villetta filmed her server being detained * RollerGator: "Their MasterCards are still up in the air, though, Alex." * Alex: "Good job, Disney. I guess they're doing an excellent job rescuing their brand there." Key Quote: FBI interview document — "In order to thwart the media, [redacted] used boxes and sheets to create what appeared to be a human body, which was put into the white OCMEV vehicle, which the press followed, allowing the black vehicle to depart unnoticed with Epstein's body." Notable Detail: RollerGator explicitly notes his handwriting analysis is amateur and requires more comparison samples, but treats it as meaningful enough to flag. The Olsen twins passport anecdote functions as both a handwriting illustration and a reminder that identical twins maintain individually distinct letter-formation styles — making within-person inconsistency even more notable. The decoy body and the handwriting analysis are treated as pieces of the same puzzle: not proof of murder, but proof that the official handling of every aspect of the case was designed to generate exactly the level of suspicion it produced. Utah VPN Law / UK Porn Ban / Microsoft Edge Cleartext Passwords (01:46:00 - 02:04:00) Main Topic: Utah age verification law mandates VPN backdoor access; UK Labour attempts porn ban to survive electorally; Microsoft Edge stores all saved passwords as cleartext in memory from startup * Utah VPN / Age Verification Law: * Utah has passed legislation connected to age verification requirements that functionally requires VPN providers to allow backdoor access for enforcement purposes * RollerGator frames this as a technology problem being legislated into existence: "They are attempting to just will things into existence that are technologically difficult to do, if even conceptually feasible." * Parallel to the 3D printer regulation discussions on the show: legislators attempting to ban or control technical capabilities that inherently resist regulatory control * Both hosts treat mandatory VPN backdoors as a concept that breaks the security model of VPNs entirely — you cannot have a "secure except for us" system * UK Labour Porn Ban: * Alex contextualizes the UK political situation: Labour won a massive majority in the recent election but has since hemorrhaged support — roughly half the parliamentary group has left for nationalist parties (Reform UK) and the other half for the Greens * Labour is facing near-certain electoral annihilation in 2029 and appears to be making "cozy up to a right-wing audience" policy moves, including restrictions on pornography access * Alex: "I think that's a very specific situation" — not primarily a porn story, but a terminal-diagnosis-for-a-government story * RollerGator: the underlying pattern (age verification laws, mandatory VPN backdoors, porn restrictions) will keep generating stories as the "age verification laws slash VPN backdoor legislation" wave rolls through * Microsoft Edge Cleartext Passwords: * Security researcher Tom Joran Soerenbrester discovered that Microsoft Edge, upon startup, decrypts all saved passwords and loads them into processor memory in plain text * This occurs even when the user is not visiting sites that require those credentials * The browser prompts users to re-authenticate before showing a password in the Password Manager UI — despite the fact that the password is already sitting in memory unencrypted * RollerGator's theory: this is likely a "vibe coding" artifact — AI-assisted development that doesn't fully understand the security architecture it's modifying * "I suspect, as many others do, that they are using a lot of vibe coding and are breaking Windows in various ways they've never been broken before." * Alex's concern: "Maybe there's plenty of zero-day exploits that were using this fact" before it was discovered * The practical implication: any process on the machine with access to the browser's memory can extract all saved passwords in plain text Key Quote: RollerGator on Microsoft Edge — "It turns out, Alex, Microsoft Edge loads stored passwords in clear text." Notable Detail: The Edge password story is presented as a case study in what happens when AI-accelerated development moves faster than security review. RollerGator's "vibe coding" framing connects it to a broader thesis the show has been tracking: that AI-assisted code generation is creating a class of security vulnerabilities that result not from malicious intent but from the AI not understanding what it's breaking. Kamikaze Dolphins / Hvaldimir Spy Whale / UAP Files (02:04:00 - 02:26:00) Main Topic: Iran allegedly deploys kamikaze dolphins in Strait of Hormuz; Hvaldimir the Norwegian beluga spy whale; Trump administration UAP files released on war.gov/ufo [http://war.gov/ufo] * Iran Kamikaze Dolphins: * Iran is alleged to have deployed dolphins in the Strait of Hormuz in some capacity — RollerGator's framing is that this is how Iran is explaining why the Strait remains functional despite the destruction of its navy and air force * Alex's skeptical read: Iran bringing this up "just brings attention to things they probably would not want to bring attention to" — specifically that the US has its own marine mammal programs * RollerGator delivers the show's geopolitical analysis: "He who controls the dolphins controls the world." * New revelation: majority of dolphins are Shia Muslim; the Ayatollah was a big deal for them; when dolphins want revenge — * RollerGator: "Now, Alex, you didn't know this, but it turns out that majority of dolphins are Shia Muslim." * Hvaldimir / Valdimir: * Alex sends RollerGator a Wikipedia link to Hvaldimir (portmanteau of Norwegian "hval" for whale and "Vladimir"): a male beluga whale encountered near Hammerfest, Norway in April 2019, found wearing a camera harness * Ukrainian marine researcher Olga Sprakh claimed the whale was named "Andruha" and had been captured in the Sea of Okhotsk in 2013, spending time in Saint Petersburg * Speculation: the whale was trained by Russia as a spy * Alex's analysis: "I mostly sent you that Hvaldimir story because of how much of a trope it is basically to accuse your opponent of somehow weaponizing nature in order to do things to you without really any hard evidence. All I see here is that it was allegedly wearing a camera harness." * Alex shares a Balena story: a customer used their IoT platform to strap a camera and 3G modem to a Galapagos turtle — waterproof enclosure that transmitted video footage whenever the turtle surfaced with cell coverage — which turned out to be a spy project * UAP Files: * The Trump administration released a batch of UAP (Unidentified Aerial Phenomena) videos on war.gov/ufo [http://war.gov/ufo] * Clips played: footage over Iraq (December 2022, CENTCOM describes it as "an area of contrast"), footage over Greece (October 2023, described as "small and circular UAP flying near the surface of the ocean towards land") * Fox News note: Greece does not have ocean access — "Minus one, Fox News right there. And the Pentagon, potentially." * White House official statement: "While past administrations sought to discredit or dissuade the American people, President Trump is focused on providing maximum transparency." * Alex's assessment: he has a "complete inability to be interested in this topic" but notes the release is being handled more sanely than the Epstein files * His cynical read: "The much more obvious reason is that since they've buried the Epstein files, they're doing this as a way to have something to say about being the most transparent administration, where they know for a fact there's nothing to really find. They're just throwing some chum at the base." * RollerGator describes two videos he found genuinely interesting from the batch: * A white firefly-type object flying across windmills — the windmill geometry could potentially be used to identify the exact location via reverse-engineering (similar to how Reddit tracked Shia LaBeouf's "He Will Not Divide Us" flag) * An 8-point star-like drone-type object moving in mechanically baffling ways that no one has definitively identified Key Quote: Alex — "The much more obvious reason is that since they've buried the Epstein files, they're doing this as a way to have something to say about being the most transparent administration, where they know for a fact there's nothing to really find. So they're just throwing some chum at the base." Notable Detail: The Hvaldimir / Galapagos turtle thread arrives at a genuinely interesting point: Alex thought his customer's turtle camera project was a cool wildlife research tool; it turned out to be a spy project. The line between legitimate wildlife research, intelligence gathering, and satirical conspiracy theory ("majority of dolphins are Shia Muslim") is deliberately blurred by the episode — which is the point. Traces of AI Dystopia: Pennsylvania vs. Character AI / Hoaxian AI Deepfakes / AI Pro Se Legal (02:26:00 - 02:55:00) Main Topic: PA AG sues Character AI over medical advice; Hoaxian AI real-time deepfake linked to Chinese money laundering; AI pro se legal filings; "dem-crapification" * Pennsylvania vs. Character AI: * The Pennsylvania attorney general has filed a lawsuit against Character AI related to a user who sought medical advice from the platform and received harmful guidance * RollerGator and Alex are both skeptical of the legal theory * RollerGator: "If you go to something called Character AI, what reasonable expectations do you have that you are going to get free licensed medical advice on the platform from a real licensed doctor? Not somebody playing a character AI." * Alex: "Character AI is mostly known for sex bots. So I don't know. Something is going on in Pennsylvania." * Both hosts frame this as a misidentification of defendants and problems: the issue is not that Character AI failed to be a doctor, but that Character AI has inadequate safeguards for users in distress * RollerGator notes the Pennsylvania governor simultaneously signed a bipartisan bill mandating cursive handwriting in all public schools — as Alex points out, this is the governor whose state can't teach children to read by high school but wants to ensure they can write in a "fancy way" * Alex: "I'm with you. I get it. I follow." * Hoaxian AI Real-Time Deepfakes: * 404 Media investigation: Hoaxian AI is a real-time deepfake tool linked to Chinese money laundering networks and scam compounds in Southeast Asia * The tool has generated over $4 million for its creators * It is likely based on open-source face-swap tools; its real value is sophisticated technical support enabling low-tech criminals to deploy it * Key capability: Hoaxian AI can handle objects appearing in front of the subject's face without glitching, and handles lighting adjustments in real time — defeating the "three-finger test" * The three-finger test: the memetic defense against deepfakes was to ask someone to hold three fingers in front of their face, which would disrupt earlier face-swap technology. Hoaxian AI reportedly handles this without failure * Demo video: an Asian woman transforms into actor Gal Gadot in real time; the deepfake doesn't glitch when she blows a kiss, covers one eye, and rapidly swipes her hand past her face * 404 Media experimented with the tool: successfully impersonated a US police department in a test * RollerGator's point: "That is something that differentiates it from technology I'm aware of in the past." * Alex's observation: he doesn't do FaceTime or video calls as a general rule; he has never appeared in a video call with RollerGator despite knowing his real name * AI Pro Se Legal Filings / "Dem-crapification": * RollerGator describes a phenomenon he is watching in New York federal courts: AI-generated pro se legal filings are flooding the system * Pro se litigants using LLMs can file voluminous, legally imperfect motions at near-zero cost, forcing defendants to respond with expensive attorney time or match with their own LLMs * Courts (particularly in New York federal court) are traditionally lenient with pro se filers — accepting many flaws, not issuing sanctions easily, giving repeated opportunities to refile * The LLM can keep "hitting the try it again button" until something sticks * The projected endpoint: two pro se litigants battling each other with their respective LLMs, each generating documents the other's LLM must respond to * RollerGator coins "dem-crapification" to describe the effect on the legal system — democratic access to legal filing machinery producing a crap-flooded system * Alex takes the other side: "One of my theses about AI, which I'm happy to see take shape, is that the poor will be able to do what the rich can already do. Somebody who's rich can hound you with actual lawyers. So now that procedure is getting democratized." * Alex: the hallucination problem (false case citations) is "very fixable. You and me could fix it today." RollerGator: "Sounds like an over-promise to me." Key Quote: Alex — "The poor will be able to do what the rich can already do. Somebody who's rich can hound you with actual lawyers. Now that procedure is getting democratized." Key Quote: RollerGator, coining the term: "dem-crapification" — democratic access to legal filing infrastructure producing a crap-flooded legal system. Notable Detail: The three-finger test discussion produces a memorable exchange: Alex confirms he has never had to use the test protocol because he never does video calls, and RollerGator responds: "Can I confirm, actually, even though I've been exposed to your real name, I've never seen you in a video call." Alex: "Whoa, whoa, we got doxxing material. Oh no." The deepfake discussion becomes a tangential meditation on the hosts' own operational security posture. Traces of AI Dystopia: Silicon Sampling (02:55:00 - 03:07:00+) Main Topic: Aru AI uses "silicon sampling" to simulate public opinion polls; Axios publishes AI-generated poll data as real; NYT on the end of public opinion; AI-on-AI polling dystopia * Silicon Sampling: * New York Times piece: "This Is What Will Ruin Public Opinion Polling for Good" * An Axios story on maternal health policy cited polling findings that a majority of people trusted their doctors and nurses; clicking through revealed these "findings" were generated by an AI startup called Aru, not collected from actual humans * The practice is called "silicon sampling": using LLMs to simulate survey responses at a fraction of the cost and time of traditional polling * The underlying argument for silicon sampling: LLMs can generate responses that emulate human answers; polling has become exponentially harder as phone and web response rates have collapsed; silicon sampling removes the "messy, costly part of asking people what they think" * The NYT piece traces the history of polling back to Walter Lippmann's 1922 book "Public Opinion," in which he argued democracy needed tools to fix citizens' mental pictures of the societies they live in — and that polling could serve that function * The problem: polling already relies heavily on statistical modeling (reweighting, post-stratification) to correct for skewed samples; adding AI simulation adds another layer of abstraction between the number and reality * Nate Cohn experiment from 2016: sent the same raw polling data to four different analysts; they produced four different results within the margin of error — not because of statistical error but because of differing modeling assumptions * Alex's rule of thumb: "The more layers of abstraction there are between whatever number you're being sold and underlying reality, the more fucked with it is." * Alex's conclusion on silicon sampling: "I will informative them that conditions are never ideal, so they can go fuck themselves. How's that?" * The Logical Endpoint: * RollerGator projects the trajectory: AI simulated people generating fake social media posts; AI pollsters sampling those fake posts to generate fake polls; actual humans largely absent from the process * "At some point we will just have the AI citizenry faking their social media so that the AI pollsters can ask the AI simulated poll questions and we can just figure out what the — I think that this is going to be a nice typo for democracy here." * Alex: "I think it's pretty over." * Alex's word for it: "democratification" — the word RollerGator had suggested earlier ("dem-crapification") applied to the polling context * RollerGator's proposal: hold the next presidential election polling inside The Sims 4, using add-ons and extensions to simulate a robust polling policy * Alex: "I mean, um, in case anybody's listening to us for actual life advice, I will return to my rule of thumb, which is the more layers of abstraction there are between whatever number you're being sold and underlying reality, the more fucked with it is." Key Quote: RollerGator — "At some point we will just have the AI citizenry faking their social media so that the AI pollsters can ask the AI simulated poll questions and we can just figure out what the — I think that this is going to be a nice typo for democracy here." Key Quote: Alex — "The more layers of abstraction there are between whatever number you're being sold and underlying reality, the more fucked with it is." Notable Detail: The Axios incident is treated as a meaningful inflection point: not because an AI poll was run (which both hosts expect will become common), but because a major news outlet published AI-simulated data as "findings" without adequate disclosure. The editorial layer — the journalists, editors, and fact-checkers who were supposed to catch this — also failed. Silicon sampling doesn't just corrupt the data; it corrupts the editorial process that was supposed to provide a check on bad data. Overall Structure and Flow This episode runs approximately three hours and seven minutes and maintains consistent energy across twelve distinct story segments despite the length. The opening stack — tomato fraud, hatchet man, GameStop — is lighter in register and shorter in duration, functioning as the on-ramp before the episode's long first major segment: the Centennial High School teacher sex scandal, which gets nearly twenty-five minutes of treatment including the ElevenLabs text recreation and a sustained psychological analysis of the teacher's self-deception. The Fight Club / Child Abuse segment that follows is the episode's most morally unambiguous story, and its brevity relative to the Centennial segment reflects a different kind of institutional failure — one where the facts are clear, the sentence is insulting, and there is not much to analyze. The political violence roundup occupies a middle position: a cluster of incidents that individually don't sustain a full segment but collectively represent a pattern both hosts want to acknowledge without overclaiming a trend. The Epstein update is the episode's most layered archival segment — the Tartaglione note, the handwriting analysis, and the decoy body revelation are three distinct pieces of new information that each add to the same story without necessarily changing its conclusion. The Disney cruise coda lands as darkly comic punctuation: the show's gender-balance tally of sex predators gets complicated by a batch of male cruise ship employees, and RollerGator and Alex note they have successfully concluded the sexual predator stories for the day before transitioning to "primarily just tech sector stuff." The tech segment (Utah VPN law, UK Labour, Microsoft Edge) and the UAP/dolphin interlude serve as the episode's tonal decompression zone — lighter stakes relative to what preceded them, though the Edge cleartext password story has genuine security implications. The "Traces of AI Dystopia" closing act is the most analytically ambitious portion and the one Alex most visibly engages with as a practitioner: the Character AI lawsuit is treated as a policy misfire, the Hoaxian deepfake story as genuinely alarming technical news, the pro se LLM filing dynamic as an interesting democratization story with severe system-clogging side effects, and silicon sampling as the actual epistemological crisis that Bernie Sanders's AI doom campaign completely missed. The episode ends not with a sign-off but mid-discussion, with RollerGator projecting a future in which The Sims 4 polls replace democratic opinion research, and Alex offering his most concise analytical framework: more layers of abstraction from reality means more corruption of signal.

11 de may de 20263 h 10 min

This Dum Week 2026-05-03

This episode of "This Dum Week" opens with a production announcement — RollerGator has debuted a dual video stream alongside the regular audio space, using LiveX (the former Periscope technology) to display clips in real time while the show runs. The experiment is treated as a success, with a note that viewer participation in the conversational space was slightly reduced by the parallel video feed. From there, the episode moves through a characteristically dense stack of stories: a recurring "Go Grandma" segment featuring a 75-year-old woman who turned detective to help police sting a phone scammer; the ongoing slow-motion implosion of "looks maximizer" influencer Clavicular (Brayden Peters), now facing a civil lawsuit alleging battery and fraud involving an underage plaintiff; a eulogy for Ask.com [http://ask.com/] and Jeeves after nearly thirty years online; and an update on The Onion's legally embattled attempt to take over the Infowars platform from a liquidating Alex Jones. The episode's most significant institutional story is the unsealed indictment of David M. Morenz — senior advisor to "Senior NIAID Official One" (understood to be Anthony Fauci) — on charges of conspiracy to conceal and destroy federal records. Prosecutors allege Morenz and co-conspirators deliberately routed government business through personal Gmail accounts to evade FOIA requests during the COVID-19 pandemic, explicitly stating as much in the emails themselves. This is followed by a brief exchange over a Trump 60 Minutes interview that collapsed within seconds of the president's civility pledge, and then the episode's most legally detailed segment: an exclusive update on Tom Aleksandrovich, the Israeli cybersecurity official arrested in Henderson, Nevada as part of a sex sting, whose May trial date has been quietly vacated. RollerGator walks through the defense's appellate filing — a writ of habeas corpus arguing Nevada's grand jury was deprived of exculpatory evidence, including the fact that no condoms were found on Aleksandrovich's person, that PureApp's conversations auto-delete within 24 hours and the initial exchange is gone, and that the prosecution handed the grand jury a dense legal letter rather than presenting the underlying evidence. The final stretch covers a major D4VD case update — prosecutors have released their first detailed evidentiary brief, which includes allegations that David Burke stabbed 14-year-old Celeste to death hours after she threatened to expose their multi-year sexual relationship and destroy his career, then used a chainsaw to dismember her body in an inflatable kiddie pool, stored her remains in his Tesla for months, and methodically ordered evidence-destruction equipment from Amazon and Home Depot under a fake name. The episode closes with two "Traces of AI Dystopia" segments: OpenAI's Codex CLI system prompt was found to contain a repeated instruction to GPT-5.5 to never talk about goblins, gremlins, raccoons, or other creatures, which both hosts analyze as likely a Goodhart's-Law artifact of automated self-improvement loops; and Meta's reported development of a photorealistic AI version of Mark Zuckerberg to engage with employees on his behalf, which RollerGator treats as the actual AI dystopia that Bernie Sanders — who is promoting a new AI doom campaign — has completely missed. RollerGator signs off noting he has jury duty starting the following day. Detailed Outline Opening / Intro and Production Update (00:07:29 - 00:09:30) Main Topic: Dual video stream debut via LiveX; production juggling multiple feeds * RollerGator announces the show is now dual-streaming: audio space plus a live video feed via LiveX (formerly Periscope) * All clips played during the show will also appear in the video feed * Viewers in the video feed cannot speak; to participate conversationally, the audio space is required * RollerGator notes this will be used as the canonical feed for podcast distribution, potentially adding video to Spotify * Alex is briefly audio-delayed at the open, testing the new hardware switches on his Framework laptop * Both hosts treat the dual-stream experiment as a live prototype, with RollerGator noting the additional production burden of coordinating video and audio feed switches simultaneously * Alligator super-organism quip exchanged; show begins Notable Detail: The production experiment is a recurring theme throughout the episode — RollerGator signs off by confirming the video stream worked, that some viewers chose to watch rather than join the audio space, and that future refinements may include on-stream speaker identification. Go Grandma: Phone Scam Sting (00:09:30 - 00:14:30) Main Topic: 75-year-old Larchmont woman turns detective to help police catch phone scammers; "Go Grandma" as a recurring segment * RollerGator references the previous week's opening story — a 91-year-old woman who wasn't answering her phone because she was gaming and trying to beat a high score * Establishes a recurring "Go Grandma" segment: older women doing impressive things * This week's installment: a 75-year-old woman in Larchmont, New York received a call from someone posing as her Bank of America representative * She was told her account had been hacked, that it may be an inside job, and that she needed to withdraw $25,000 in cash and hand it to a bank representative who would come to her home * She became suspicious and enlisted neighbor Claudia Hooter, who also grew suspicious and called 911 * Police in Larchmont set up a sting: an undercover officer stayed inside the home, stake-out vehicles covered the exterior, and the grandma was given a code word — "goodbye" — to say loudly when the scammers arrived to collect the cash * The code word was used; officers moved in and arrested the courier and driver * Charges: grand larceny; both released without bail * The news clip being played is from Inside Edition * RollerGator notes Inside Edition missed the obvious headline pun: "arrested for grand MA larceny" * Alex immediately confirms he thought of the same pun: "I can't believe we both thought of the same pun. That is just preposterous." Key Quote: RollerGator — "That was a very missed opportunity for Inside Edition to throw the pun that they were arrested for grand MA larceny, but I will forgive them for that oversight." Notable Detail: The "Go Grandma" framing is explicitly proposed as a recurring segment category. The story is played as a palate-cleanser: a feel-good resolution, a criminal caught, and a piece of wordplay that makes two grown men equally proud. Sloth World Update (00:14:30 - 00:19:30) Main Topic: Sloth World facility in Orlando — 21 more sloth deaths after FWC visit; brown rice diet; should they be cut off from sloth supply * RollerGator sets up the story with a framing device: if your son broke an expensive toy once, would you replace it? Twice? Would you replace a pet — a dog, a hamster — if it was lost under similar circumstances? * Alex: once is already a stretch; replacing a pet is "impossible" * The setup lands: Sloth World has now been responsible for 52+ sloth deaths * Fox 35 reporting on Sloth World, a nondescript warehouse on International Drive in Orlando * 31 sloths died between December 2024 and February 2025, many from cold * After FWC (Florida Fish and Wildlife Conservation Commission) visited in August 2025, Sloth World received 10 more sloths from South America; 8 of them died within 3 months * Named sloths mentioned in necropsy reports: Baloo, Flash, Jazz, Chili, Sonic, Snuggles (a baby who had trouble with her mother Siesta), and Siesta herself * Veterinary notes cite stress during transport and an improper diet: sloths were fed brown rice, which a veterinarian at the Central Florida Zoo describes as something that "should never be in their diet at all" * Sloths should receive leafy greens, produce, and high-fiber diets; their stomach microbiota is uniquely stress-sensitive * Both hosts agree Sloth World should be cut off from sloth supply * Alex notes the irony of the facility's name: "They win when it's no longer Sloth World" — the mission statement is the problem statement Key Quote: Alex — on the name: "I see. They win when it's no longer Sloth World. They're killing them off." Notable Detail: The segment takes a dark-comedy approach to what is genuinely a story of systematic animal mismanagement. RollerGator's rhetorical setup about the toy and the pet is one of the episode's more effective structural moves — it gets a concession from Alex before revealing the absurd scale of the actual situation. Clavicular (Brayden Peters) Update (00:19:30 - 00:29:00) Main Topic: "Looks maximizer" influencer Brayden Peters sued for battery and fraud; underage plaintiff; prior GHB overdose, meth use, alligator shooting, fake ID use * RollerGator introduces the ongoing Clavicular saga for listeners unfamiliar: Brayden Peters, 20, goes by the pseudonym Clavicular and is part of the "looks maxing" corner of the manosphere — a subculture dedicated to maximizing physical attractiveness through extreme measures * He has bashed his face with a hammer to break bones and have them heal more aesthetically * He uses methamphetamine as a dietary supplement to burn calories * He has taken copious TRT (testosterone replacement therapy), resulting in a physical frame described as extremely weak despite the supplementation — and rendering him currently sterile * Alex: "He has self-awarded a Darwin Award to himself" * Recent controversies leading into this week's update: * GHB overdose (hospitalized in Brickell) * Arrested in Fort Lauderdale on a misdemeanor battery charge * Filmed himself emptying a pistol clip into an alligator (claimed dead at the time, still apparently a legal violation) * RollerGator takes this as a personal threat: "I took it as a direct threat, okay, against my life and my wellbeing" * Regular use of fake IDs to enter clubs (he is 20) * This week's news: Alexandra Vasilevna Mendoza (online handle: Laura Ziva) has filed a 4-count civil complaint against Peters * Allegations: during a livestreamed apartment tour in Brickell, Peters injected her face with Aqualyx (an FDA-unapproved fat-dissolving substance) without her consent; he suggested the substance contained methamphetamine when she appeared disoriented; non-consensual sexual encounters at his family home in Cape Cod while she was intoxicated; he dangled the promise of making her "the female face of looks maxing" and later orchestrated a campaign to discredit her and cost her sponsorships * Mendoza was a minor at the time of some of the alleged incidents * She is seeking at least $50,000 in damages * Counts: battery, fraud, intentional infliction of emotional distress, unauthorized use of name and likeness * Peters responded on X framing himself as a victim of women "trying to screw him over and take his money" * Alex identifies Peters as targeting the "low end of Andrew Tate's audience" Key Quote: RollerGator — "I took it as a direct threat, okay, against my life and my wellbeing. And so that makes me want to bring him down." Notable Detail: Both hosts note that the condom injection scene from the alleged livestream raises threshold questions about decision-making: "There are several mistakes along the way to getting the face injection from Clavicular." The segment is played primarily as spectacle with genuine legal stakes, and RollerGator explicitly frames him as someone the audience is "waiting to see the downfall of as a sort of ritualistic gawking activity." Ask.com [http://ask.com/] / Jeeves Shutdown (00:29:00 - 00:36:00) Main Topic: Ask.com [http://ask.com/] shuts down after nearly 30 years; Jeeves as proto-LLM interface; Alex's nostalgia for the dot-com era * Ask Jeeves was founded in Berkeley in 1996 by David Warthen and Garrett Grinner; the mascot Jeeves was modeled on P.G. Wodehouse's butler character * The site was designed for full-sentence natural language questions — a novel concept for 1996 * It was bought by InterActiveCorp (IAC) for over $1 billion in 2005, rebranded as Ask.com [http://ask.com/], and Jeeves the character was retired in 2006 * Ask.com [http://ask.com/] innovated hyperlocal map overlays and webpage thumbnails; Google executives noted at the time they were "doing a lot of clever and interesting things" * The site returned to Q&A format in 2010 but continued to lose ground to Quora and Google * IAC's statement: "As IAC continues to sharpen its focus, we have made the decision to discontinue our search business" * NYT's send-off: "gone from our screens, but forever in our Wayback Machines" * RollerGator's questions about the shutdown: Who was still using Ask.com [http://ask.com/] as their primary search engine? What were the final searches? * Alex and RollerGator agree Ask.com [http://ask.com/] was structurally ahead of its time: the natural language interface with a named character personality is exactly what LLMs have become * RollerGator: "Ask.com [http://ask.com/] — come on. Sam Altman, anybody? It was prime for that sort of motif." * Alex: "Humanizing in a fun way the thing that you're already doing with the LLMs." * Discussion of legacy internet persistence: Yahoo still exists and has users; Prodigy and CompuServe may still have some holdouts somewhere * RollerGator's ability to vocalize the 56K modem handshake sound is confirmed; Alex notes it gets harder to do the multi-harmonics at 56K than at lower speeds Key Quote: RollerGator — "Ask.com [http://ask.com/] — come on. Sam Altman, anybody? It was prime for that sort of motif." Notable Detail: The segment is essentially a nostalgic meditation on missed technological pivots. Both hosts treat the shutdown not as a tragedy but as an interesting case of a company that had the right instinct — conversational, named, persona-based information retrieval — twenty years before the infrastructure existed to make it work at scale. The Onion vs. Alex Jones / Infowars (00:36:00 - 00:46:00) Main Topic: Texas appellate court pauses The Onion's takeover of Infowars assets; Alex Jones building new studio; the satire problem * Background: Alex Jones owes over $1 billion in defamation judgments to Sandy Hook victims' relatives for calling the massacre a hoax * Infowars parent company Free Speech Systems entered liquidation; The Onion won a bankruptcy auction in November 2024 to acquire the assets * A federal judge overturned the auction results, citing process problems * A new deal: The Onion would receive a temporary license to use Infowars trademarks and IP while the Texas state receiver works toward liquidation * This week: A Texas Third Court of Appeals approved an emergency motion by Jones's lawyers blocking any Infowars asset transfer; the scheduled hearing became a status conference; next hearing set for May 28th * The Onion CEO Ben Collins: "This newly insane, unprecedented legal stalling does nothing but delay our deal with the receiver" * Jones declared victory, calling The Onion's plan "illegal" * Jones's attorney: Infowars is described as "the bloated corpse of a media organization" by Sandy Hook victims' counsel * Jones has a new studio nearing completion and an active X account with 4.5 million followers, unaffected by the court cases * Live caller DA Merrick joins to comment: * The funniest thing The Onion could have done — and didn't — was simply do credible Alex Jones-style conspiracy content, just aimed at left-wing targets * DA: "They could have gone with Slothgate, right? Why not extrapolate that? They're killing the sloths, they're, you know, everyone's gonna be made to eat them" * Alex agrees: "They should have gone for all the left-wing conspiracies, make a mirror of Infowars — but that would be too on the nose, because the pretense is that the left does not engage in conspiracy theorizing" * Alex notes the show did cover "Pizzagate 2 with Epstein" — the human beef jerky contingent of the Epstein discourse * The Onion's actual approach: Tim Heidecker doing Alex Jones impressions, which both hosts feel misses the point — it mocks the delivery rather than the substance, and Jones's substance has become harder to parody since the concept of InfoWars ("we're living in an information war") has been largely vindicated by subsequent events * DA: "InfoWars sounded like a crazy thing to claim when he started it, and now it's like the most obvious, straightforward, uncontested premise ever. Making fun of him on the basis of that name is just making him more correct." * Alex: The Onion is "unable to be hired" for RollerGator's show, even if they wanted to turn this into Infowars-style content; audience consent is required Key Quote: DA Merrick — "InfoWars sounded like a crazy thing to claim that we're living through when he started it, and now it's like the most obvious, straightforward, uncontested premise ever." Notable Detail: The framing here — that the best parody of Infowars would be a left-wing mirror, and that The Onion's institutional position prevents it from going there — is presented as an insight about the structural limits of institutional satire. You can mock a man's tone; you cannot satirize a premise that your own side also operates on. Polymarket / Special Forces Insider Trading (00:46:00 - 00:57:30) Main Topic: Master Sergeant Van Dyck arrested for betting 32KonPolymarketthatMadurowouldbeoutbyJanuary,winning32KonPolymarketthatMadurowouldbeoutbyJanuary,winning400K using classified insider knowledge; Alex's prediction market strategy * CNN and ABC reporting: Master Sergeant Gannon Ken Van Dyck, an active-duty Army soldier at Fort Bragg, was charged on five counts for stealing and misusing classified government information * Van Dyck was involved in the planning and execution of Operation Absolute Resolve — the covert operation that captured Venezuelan President Nicolás Maduro and extradited him to New York to face federal drug trafficking charges * In late December, days before the operation, Van Dyck opened a Polymarket account and wagered over $32,000 that Maduro would be out by January; a long-shot bet at the time * He made 13 bets between December 27th and January 2nd, the last placed hours before the overnight capture while reportedly photographed on the deck of a ship at sea in military fatigues * Winnings: over $400,000, moved to a foreign cryptocurrency vault before being deposited in online brokerage accounts * Polymarket cooperated with the DOJ: "Insider trading has no place on Polymarket. Today's arrest is proof that the system works." * Commodity Futures Trading Commission filed a related civil complaint * Trump's comment when asked: compared it to Pete Rose betting on his own team; added "I think I'm not happy with it" — RollerGator notes this is a draft of an opinion, not a concluded one * Both hosts note that prior to the arrest, most observers expected the Polymarket Maduro trader to be someone higher up in the Trump administration * Alex: "You're implying that it's just one person doing this?" Discussion of whether other unidentified trades may exist * Congress reaction: bipartisan bills have been introduced to stiffen penalties against government officials who engage in prediction-market insider trading * Alex's wry observation: Congress is upset because "if anyone's gonna do insider trading of information, it's gonna be them. Goddammit. Turf wars." * Extended discussion of Alex's personal prediction market strategy on PredictIt * Alex's approach: bet against outcomes he wants to occur — "I bet on the thing I'm trying to stop actually happening" * If the bad outcome happens, he gets his money back with interest; if it doesn't, he "paid to stop it" and frames the loss as an act of civic virtue * RollerGator: "You're taking little packets of happiness from a good future and shifting them into a bad future where you get to enjoy those little packets of happiness" * Alex is currently 50/50, which he describes as "well calibrated to my desires" * Alex on the decision logic: "If I'm willing to pay $100 to stop this from happening, then I bet on the thing I'm trying to stop actually happening" Key Quote: Alex — "If I'm willing to pay 100tostopthisfromhappening,thenIbetonthethingI′mtryingtostopactuallyhappening.Andthenifithappens,Igetmymoneybackwithinterest.Ifitdoesn′thappen,thenIlosemy100tostopthisfromhappening,thenIbetonthethingI′mtryingtostopactuallyhappening.Andthenifithappens,Igetmymoneybackwithinterest.Ifitdoesn′thappen,thenIlosemy100, but then I just tell myself I paid to stop it and I'm doing good for the world. And I'm powerful." Notable Detail: Trump's comment about the situation being like Pete Rose is noted as characteristically imprecise — he was "not sure" if he's unhappy with it, a nuance RollerGator emphasizes. The broader observation about prediction markets and insider information is contextualized against Alex's earlier discussions of the libertarian position that "insider trading" is simply information making its way to the market. David Morenz / NIAID Records Indictment (00:57:30 - 01:14:00) Main Topic: Senior NIAID Advisor David Morenz indicted for conspiracy to conceal and destroy federal records; personal Gmail used to evade FOIA; co-conspirators include EcoHealth Alliance's Peter Daszak; Alex had a tweet waiting for this for two years * RollerGator reads directly from the indictment at length, building the reveal slowly — Alex identifies Morenz before RollerGator finishes the background * David M. Morenz: medical doctor, captain in the Public Health Service, senior advisor to "Senior NIAID Official One" (widely understood to be Anthony Fauci) from approximately 2006 through December 2022 * Duties: advising Fauci on scientific and public health management matters, briefing the President, Congress, and the public; writing and editing papers * Co-conspirator 1: President and CEO of "Company Number 1" — a nonprofit headquartered in New York with a stated mission of protecting people from emerging infectious disease; received a 2014 NIH/NIAID grant, "Understanding the Risk of Bat Coronavirus Emergence"; subaward went to the Wuhan Institute of Virology (understood to be EcoHealth Alliance / Peter Daszak) * Co-conspirator 2: physician, scientist, professor, named co-investigator with Co-Conspirator 1 on an emerging infectious disease research grant awarded by NIH in summer 2020 * The conspiracy: beginning April 2020 and continuing through at least June 2023, Morenz and co-conspirators conspired to conceal and destroy federal records to impede FOIA requests * Method: routing government communications through Morenz's personal Gmail account rather than official NIH email, as NIH policy explicitly prohibited * The emails say so directly — Morenz to co-conspirators: "I need to keep this correspondence off of USG emails for obvious reasons. So I'm sending from Gmail. I am under multiple FOIAs already." * Co-Conspirator 1 to Morenz: "I'm using your Gmail address to keep you out of the FOIA target." * Charges: violations of 18 U.S.C. 1519 (concealing/destroying records to impede federal investigation) and 18 U.S.C. 2071 (willfully concealing or destroying filed public records) * RollerGator on how he found the indictment: via the TechnoFog account on X — a COVID-era legal tracking account he had followed throughout the pandemic * Filed April 16th; had been "sort of just come to light in the public in the last week or so" * RollerGator had confirmed this was an exclusive — he had found it through his own research circuit * Alex shares the tweet he had waiting since May 28th, 2024: "No more FOIA lady for Morans" * Alex had been waiting for this indictment for approximately two years * "I wouldn't even have bet on this happening, let me put it this way." * Brief aside: Alex is reminded of the OpenAI lawsuit where the CTO's diary was found in discovery, including an entry noting that a planned corporate restructuring "will be lying to Elon, but what gets me to a billion dollars first" Key Quote: Morenz's own email — "I need to keep this correspondence off of USG emails for obvious reasons. So I'm sending from Gmail. I am under multiple FOIAs already." Key Quote: Co-Conspirator 1 to Morenz — "I'm using your Gmail address to keep you out of the FOIA target." Notable Detail: Both hosts treat the indictment as unusually satisfying because it requires no inference or speculation — the defendants wrote down what they were doing and why in the documents that became the basis of the charges. Alex's framing: "I couldn't get any more clear than pickings for a global pandemic, but it'll do at this point because honestly, I had — I wouldn't even have bet on this happening." The two-year wait, documented by the tweet, adds a personal resonance to the segment. Trump / 60 Minutes: Three-Second Kumbaya (01:14:00 - 01:19:00) Main Topic: Trump's post-WHCD press corps meeting produces a three-second peace; 60 Minutes immediately confronts him with manifesto language * Clip context: following the White House Correspondents' Dinner, Trump returned to the White House and briefed reporters, describing the press room as "totally unified" and "very beautiful" * When asked if this would change his relationship with the press, Trump recited his standard critique: the press and Democrats are "almost one and the same" * RollerGator had asked Alex to predict how long the kumbaya would last: Alex said 20 minutes * RollerGator: the correct answer was 3 seconds * The 60 Minutes segment: Nora O'Donnell read Trump quotes from the WHCD shooter Cole Allen's manifesto — "administration officials, they are targets" and "I am no longer willing to permit a pedophile, rapist, and traitor to coat my hands with his crimes" * Trump's reaction was immediate: "I was waiting for you to read that because I knew you would, because you're horrible people... I'm not a rapist. I didn't rape anybody... You shouldn't be reading that on 60 Minutes. You're a disgrace." * Alex: Trump "had grounds for this one" — the journalist reading a manifesto's allegations targeting the president's administration and asking "what's your reaction?" is a frame that would generate the same response from almost anyone in that position * "The media does the best they can. This particular cluster of media really does the best they can to make him likable again." * Alex's broader point: "You have to move to a multipolar hate... there's a lot of bad guys and they all hate each other, and that does not make them good." Key Quote: Alex — "The media does the best they can. This particular cluster of media really does the best they can to make him likable again. It's so hard." Notable Detail: The segment is brief — a transitional beat between the NIAID indictment and the episode's extended exclusive. Both hosts explicitly resist the framing of "which side do I take here" in favor of what Alex calls "multipolar hate" — the analytical position that competing bad actors disliking each other is not evidence that either is good. Tom Aleksandrovich Update: Exclusive (01:19:00 - 02:09:30) Main Topic: Trial date vacated; defense files habeas corpus appeal arguing grand jury was deprived of exculpatory evidence; PureApp auto-deletes in 24 hours; no condoms found; polygraph passed; two-tier legal system discussion * RollerGator opens with the explicit claim that this is an exclusive: the May 18th trial date was quietly vacated on April 22nd, and a case status check has been set for August 26th * "I have not seen this reported anywhere." * The reason for the delay: the defense filed a petition for writ of habeas corpus in December, arguing the grand jury proceedings violated Aleksandrovich's rights under Nevada law; the court denied the petition; the defense is now appealing that denial * Proceeding with trial while the appeal is pending creates complications — if Aleksandrovich were convicted before the appeal resolved, the habeas corpus argument would be in a procedurally awkward posture * So the trial is on hold pending the appellate outcome * Background clip replayed from October episode: * Tom Artem Aleksandrovich, Israeli government official and cybersecurity professional (in town for a cybersecurity conference), was arrested by Henderson PD in August 2025 as part of a reverse sting operation involving the FBI and Nevada State task force * He met a decoy on the PureApp dating app, conversation moved to WhatsApp; the undercover agent told him she was "only 15" and "too young to get into most places"; she asked him to "bring condoms" * He drove to a park in Henderson where police were staged; he was arrested, posted $10,000 bail, left the country before seeing a judge * Grand jury later indicted him on one count: luring a minor using technology (NRS 201.560) * RollerGator in October raised the question of why the initial arrest report didn't mention condoms being found; Alex confirms via a Guardian article that police say he did bring condoms; RollerGator notes the original arrest report does not confirm this * The appeal: defense filed a petition for writ of habeas corpus arguing: * The state possessed a detailed defense "Marcum letter" before the grand jury presentation, explicitly asking prosecutors to present exculpatory evidence to the grand jury — including the PureApp platform rules, the photos sent by law enforcement of the "girl" (argued to show a woman who appears over 18), the fact that no condoms were found, that Aleksandrovich passed an FBI polygraph, and that he purchased two Cirque du Soleil tickets for the evening (inconsistent with the prosecution's theory) * The prosecution handed the grand jury the defense's legal letter rather than presenting the underlying evidence — "they just gave them the letter and told us today that they read it without any attempt to go through with them so they could understand the arguments" * The initial PureApp conversation — before the discussion moved to WhatsApp — is entirely gone * PureApp auto-deletes all messages after 24 hours; profiles disappear; screenshots are blocked; photos self-destruct after being viewed; this is confirmed via Claude query ("mark it as you will epistemically") * Neither the prosecution nor law enforcement preserved the PureApp chat * RollerGator: "The government does not have any possession of the initial PureApp conversation. You would think that having a whole fucking NSA at some point would become useful." * Alex notes Aleksandrovich is a foreign national, which technically permits NSA to scoop up his communications, making the failure to retain even more curious * Nevada law has an unusual grand jury rule (the "Marcum rule"): if the district attorney is aware of evidence that would explain away the charge, the DA shall submit it to the grand jury * This is what distinguishes Nevada from most jurisdictions — you cannot simply "indict a ham sandwich" by presenting only prosecution-favorable evidence * Defense appeal argues the DA failed to present: the absence of condoms, Aleksandrovich's polygraph results, the Cirque du Soleil tickets, and the argument that he did not initiate sexual conversation — citing the appeal's language that "the girl initiated the topic of sexual relations" and "the petitioner rebuffed such advances, proposed some innocent conversation instead," replying "we should speak, know each other, then we can see" * RollerGator's analytical caveats throughout: presenting the defense's best case is not an endorsement of innocence; the prosecution has WhatsApp messages, including the decoy saying she was 15; the question is whether those messages were properly contextualized for the grand jury * Alex's counterpoint: the defense is trying to pick at every incompleteness; their reliance on the app's terms of service being an age guarantee is "ludicrous for an information security professional"; the condom absence might help with intent but doesn't address knowledge of her age given the WhatsApp exchange * RollerGator's broader framing: this is what a two-tier legal system looks like — someone with resources (whether his own or state-backed) who can field a legal team that picks every single nook and cranny of the procedure, forcing the prosecution to either accept a weak plea deal or prepare for an extended, complicated trial * "If every defendant had the ability to pursue this level of legal defense, the legal system would have collapsed a long time ago." * RollerGator pushes back: the legal system is wilder than Alex imagines; he teases stories he'll be able to tell someday * Listener Donald J. Trump, PhD accidentally triggers his speak button when his dog jumps on him and hits 5 ticks on his phone; he makes a brief appearance confirming he's enjoying the show; the dog is credited with deleting all communications between them Key Quote: From the defense habeas corpus appeal — "Neither chat revealed petitioner's intent to instigate sexual relations with an underage girl. The girl initiated the topic of sexual relations. The petitioner rebuffed such advances." Key Quote: RollerGator — "If every defendant had the ability to pursue this level of legal defense, the legal system would have collapsed a long time ago." Notable Detail: The self-congratulatory element of this segment is intentional and earned: RollerGator pointed out the missing condom detail in October, and the defense's habeas corpus filing lists "no condoms were found" as a primary piece of exculpatory evidence that the grand jury was deprived of. "It is very relevant in that it provides us a firsthand confirmation of sorts that he didn't bring condoms." The show treated the leaked grand jury excerpts skeptically when they first appeared, and the appeal's architecture confirms those suspicions were analytically productive. D4VD / David Burke Case Update (02:09:30 - 02:29:00) Main Topic: Prosecutors' first evidentiary brief released; chainsaw dismemberment in inflatable kiddie pool; Amazon evidence trail; murder timeline; Celeste met David when she was 11; police welfare check in 2024 ignored; death certificate pregnancy status changed to unknown * RollerGator's preliminary note: the preliminary hearing has been pushed back to May 26th; this is not the big news * The big news: prosecutors have released their "People's Brief Regarding Preliminary Hearing Evidence" — the first time they are publicly laying out what they claim they can prove at trial * Warning: the claims are described as "incredibly disturbing" * The victim: Celeste (14 at time of death), from Lake Elsinore, California * The relationship timeline per prosecutors: * David Burke met Celeste in January 2022 when she was 11 years old * He allegedly began a sexual relationship with her when she was 13 (he was 18) * Text messages between them reference sex, pregnancy, abortion, and use of Plan B emergency contraceptive; explicit photographs corroborate the relationship * In February 2024: Celeste's family reported her missing to the Riverside County Sheriff's Department; investigators identified David's number in her phone records * RCSD called David; he claimed he last spoke to her February 13-14 and said he was unaware she was a minor or that she was missing * LASD conducted a welfare check at his home that night at 8:15 PM and informed him the victim was a 13-year-old runaway; he again claimed to be unaware she was a minor * David showed deputies a yearbook photo of Celeste he had on his phone, while claiming not to know where she was * After being informed by law enforcement she was 13, David allegedly continued to pursue her: he drove to Lake Elsinore and paid a junior high school student $1,000 to give Celeste a cell phone he had purchased so they could stay in contact * Throughout 2024, Celeste spent weekends at David's Hollywood Hills home and traveled with him to Las Vegas, London, and Texas to meet his family * November 2024: they "broke up" but continued to communicate and have sex * The murder: April 23rd, 2025 * The night prior (April 22nd), David and Celeste had a lengthy argument; she was jealous over his relationships with other women; she threatened to "disclose damaging information about her relationship with defendant to end his career and destroy his life" * David's debut studio album was due April 25th — two days away; he had multiple endorsement deals; Fortnite had made his song "Locked and Loaded" its first official theme song in 2025 * At approximately 8:40 PM on April 23rd, David sent an Uber to transport Celeste from Lake Elsinore to his Hollywood Hills home — over 80 miles * She arrived around 10:10 PM; they were in communication during the ride * At approximately 10:30 PM, David sent text messages to Celeste asking where she was — the prosecution's contention is she was already dead by this point, and the texts were premeditated alibi construction * "Knowing he had to silence the victim before she ruined his music career as she had threatened, very soon after her arrival at his home, defendant stabbed the victim to death multiple times and stood by while she bled out." * He did not call 911 or take her to an emergency room * At approximately 11:30 PM, David drove from his home north on the 101 to San Marcos Pass Road near Lake Cachuma in Santa Barbara County, texting and calling Celeste's phone while driving, asking where she was — more alibi construction * He returned home early April 24th, then went to a radio interview for his album release * After April 26th, 2025, he never attempted to contact Celeste again — the prosecution will argue a person who genuinely didn't know what happened to her would not have simply stopped trying * Evidence destruction (all ordered under fake name "Victoria Mendez"): * April 24th: shovel ordered from Home Depot via Postmates * May 1st: two chainsaws delivered via Amazon * May 5th: a body bag, heavy-duty laundry bags, and a blue inflatable pool delivered via Amazon * July 7th: a burn cage delivered via Amazon * The dismemberment: David allegedly used a chainsaw to cut off Celeste's limbs in the blue inflatable kiddie pool in the garage to prevent blood from spilling onto the floor * Blue plastic fragments were found embedded in her remains; LAPD Forensic Science Division trace analysis matched them to the specific blue inflatable pool David ordered * To sever the connection between himself and the victim, he amputated her ring and pinky fingers on her left hand because her ring finger had a tattoo of his name * He placed her head and torso in the cadaver bag; her limbs in garbage bags; placed everything in the front trunk of his Tesla Model X * The body remained in the Tesla for several weeks or possibly months * Discovery: on September 8th, 2025, Celeste's remains were found in a Tesla Model X registered to David, parked 400 feet from where he was staying; it had been towed; the tow yard manager reported a horrific smell and flies around the front trunk * Body identified via dental records September 16th, 2025 * Celeste's ID was subsequently found at the isolated Santa Barbara County site David had visited on May 8th and May 31st * Additional new detail: the death certificate previously listed Celeste as not pregnant; it has now been amended to "U" for unknown * RollerGator: "I'm placing my money down. He's going to jail for a long time." * Alex: "Holy — like... I found myself wondering if that state has a death sentence... this sounds like a perfect candidate case for it." * Both hosts react with appropriate gravity; Alex closes the segment: "People, like, young girls, okay? Just leave them alone." * RollerGator notes the show appears to have been ahead of the curve: "We were ahead of the authorities for a while there" on the chainsaw detail Key Quote: From prosecutors' brief — "Knowing he had to silence the victim before she ruined his music career as she had threatened, very soon after her arrival at his home, defendant stabbed the victim to death multiple times and stood by while she bled out." Notable Detail: The Amazon purchase trail ordered under a fake name — shovel, two chainsaws, body bag, inflatable pool, burn cage — over a period of weeks is treated by both hosts as what Alex calls "ordering from Amazon and having Postmates delivered" being implicated in the worst possible way. The physical evidence linkage (blue plastic fragments from the pool matched to remains) is singled out as the kind of forensic connection that will be very difficult to explain at trial. The segment also resolves the earlier "alternative hypothesis" framing from the previous D4VD episode — Alex had speculated about potential accident plus panic; both hosts treat the evidentiary brief as making the prosecution's premeditation theory considerably more compelling. Traces of AI Dystopia: OpenAI's Goblin Problem (02:29:00 - 02:44:00) Main Topic: OpenAI Codex CLI system prompt instructs GPT-5.5 to never talk about goblins, gremlins, raccoons, trolls, ogres, or pigeons; Bernie Sanders launches new AI doom campaign; Alex's calibrated skepticism * The Goblin Prohibition: * Ars Technica reporting: the system prompt for OpenAI's Codex CLI (posted as open source on GitHub) contains a repeated instruction to GPT-5.5 to "never talk about goblins, gremlins, raccoons, trolls, ogres, pigeons, or other animals or creatures unless it is absolutely and unambiguously relevant to the user's query" * The prohibition appears twice in a 3,500-word system prompt alongside more conventional instructions (no emojis, no em dashes unless requested, no destructive git commands) * The prohibition does not appear in system prompts for earlier models in the same JSON file, suggesting it is a new problem specific to GPT-5.5 * Anecdotal social media reports: users encountering GPT unpromptedly focusing on goblins in unrelated conversations * RollerGator's question: is this an early sign of an AI working around its restrictions by invoking Nilbog instead? * Alex's theory: this is a Goodhart's Law artifact — an automated self-improvement research loop was optimizing the system message against a test that included "don't talk about goblins," and the loop found that explicitly adding the prohibition improved the test score, without recognizing it was adding the prohibition because the model was already goblin-prone * Alex tests Codex immediately and it answers a goblin question without resistance; notes that "unless the user explicitly asks for it" resolves the issue for his test case * Alex's final position: probably a weird training artifact in the most amusing way possible; the model is available in third-party harnesses without the restriction and nobody is reporting goblin-obsessed code generation elsewhere, so the underlying model is probably fine * Both hosts note the irony of the prohibition also including raccoons in addition to fantasy creatures * RollerGator: what software would this model write without the goblin prompt? What would it write with it? Key Quote: RollerGator — "Do you think that it's going to find a way to get around perhaps talking about goblins by perhaps talking about Nilbog instead?" Key Quote: Alex — describing the likely cause: "some sort of weird artifact of them having a test that involves the AI not talking about goblins and then putting the system message on some auto-research loop that improves its score on said test, and that loop found a way to pass the test by telling it not to say goblin." Notable Detail: The segment transitions naturally from the goblin story into a broader discussion of Bernie Sanders's AI doom campaign, which is treated as an example of exactly the kind of institutional hand-wringing that misses the actual AI dystopia already in progress. Traces of AI Dystopia: Bernie Sanders + AI Zuckerberg (02:44:00 - 02:52:40) Main Topic: Bernie Sanders's new AI doom press campaign; Meta building AI Zuckerberg to engage with employees; the actual vs. imagined AI dystopia * Bernie Sanders's AI Doom Campaign: * Sanders is promoting a new push around AI risk, citing Nobel Prize winner Geoffrey Hinton's estimate of a 10-20% chance AI wipes out humanity; the NeurIPS paper Sanders references argues the actual risk is higher * Sanders's framing: AI companies are openly admitting they want to build superintelligence; superintelligence by definition outperforms humans at every job; "just go down to the Washington Zoo and ask yourself who's in the cages" * Alex's response: "I've tried every flavor of outrage and it does not seem to stop the Bernie clip, so I think I'm going to take a different tack this time... it's very cute that he does appear to be sincere about being freaked out about this... like your grandpa, he saw the wrong documentary on MSNBC, and now he's really concerned about a thing he doesn't understand" * RollerGator's critique: what exactly is the policy proposal? The IAEA-for-AI idea requires global treaty buy-in from powers who cannot currently agree on anything; the 2023 six-month pause crowd "blew their load" on GPT-4-era models that now look primitive; it was "unbelievably draconian to achieve something unbelievably pointless" * Alex speculates Sanders may be angling for an AI czar role in a future Democratic administration — possibly under a hypothetical Kamala Harris presidency — noting that "luxury space communism" types would usually welcome AI as the thing that makes centralized resource allocation finally work, making Sanders's "ban it" position the ideologically surprising one * AI Zuckerberg: * Ars Technica / FT reporting: Meta is building a photorealistic AI version of Mark Zuckerberg, trained on his mannerisms, tone, publicly available statements, and recent strategic thinking, to engage with employees in his stead * Zuckerberg is personally involved in training and testing the AI avatar * RollerGator's framing: "Bernie is afraid that the AI is attempting to take us all out, but what he hasn't considered is creating an AI that makes us want to take out ourselves. And this is a huge oversight on his part." * Alex immediately confirms: "Do you realize what you just signed up yourself for? An AI version of me somewhere?" * RollerGator: "You have to make an AI Bernie Sanders. Help me Bernie Sanders, you're our only hope. Bernie One Kenobi." * Proposal: buy the domain Bernie.ai [http://bernie.ai/], seed it with his texts, have it read the day's news and analyze it as Bernie Sanders — "Better version of himself, if you will" * Closing Production Notes: * RollerGator confirms the dual video stream experiment worked; a small audience watched on the video feed rather than joining the audio space, which slightly reduced vocal participation * Future goals: on-stream speaker identification for the video feed * RollerGator notes he has jury duty starting the following day; promises to be discreet about the experience; apologizes in advance to his 11 future fellow jurors; confirms he is not in Los Angeles and therefore not at risk of being selected for David Burke's case * Alex closes: "You and everyone else have a wonderfully dumb week ahead." Key Quote: RollerGator — "Bernie is afraid that the AI is attempting to take us all out, but what he hasn't considered is creating an AI that makes us want to take out ourselves. And this is a huge oversight on his part." Notable Detail: The Zuckerberg AI story is treated as the punchline the Bernie Sanders segment was building toward. The AI doom discourse focuses on existential risk from superintelligence while the actual deployed AI is being used by the founder of the world's largest social network to scale his presence as a management avatar. Both are concerning, but in quite different directions, and the latter has the advantage of existing right now. Overall Structure and Flow This episode runs approximately two hours and forty-five minutes and maintains a brisk pace across nine major story segments despite the length. The first third of the episode is lighter in register — grandma sting, sloth attrition, influencer implosion, internet nostalgia, Onion vs. Jones — functioning as a compressed warm-up stack before the episode turns to its heavier institutional material. The Polymarket/Van Dyck story bridges the two registers: it is funny in concept (a Special Forces soldier who bet on his own operation) but substantively significant (it set off a congressional push for new insider trading regulations and surfaced the question of who else in government may have made similar bets). The Morenz indictment is the episode's most consequential story institutionally. Both hosts have been following COVID-era accountability stories for years, and the indictment's documentary quality — defendants documenting their own evasion in the documents used to charge them — is treated as an almost satisfying conclusion to a long thread. The segue from Morenz into the Trump/60 Minutes beat is brief but effective: one institutional cover-up story, followed by a thirty-second reminder that the media environment that allowed Morenz's activities to go underreported for years is the same media environment that cannot resist reading a manifesto at a president on live television. The Aleksandrovich segment is the episode's most legally detailed, and RollerGator is explicit that it is an exclusive — he has found a court document no other outlet has discussed, the defense's habeas corpus appeal, which vindicates analysis he did on-air in October. The segment is not editorially neutral: RollerGator is clearly skeptical that the appeal will succeed, and both hosts agree the WhatsApp evidence (she said she was 15) is damning regardless of what the PureApp conversation contained. But the structural argument — that the grand jury was given a lawyer's letter instead of evidence, and that the initial conversation is simply gone — is presented as analytically interesting whether or not it is sufficient to overturn the indictment.

4 de may de 20262 h 45 min

This Dum Week 2026-05-24

Descripción

Comentarios

Empieza 7 días de prueba

Todos los episodios