Skip to main content

[@ModernRogue] Weird Things Podcast #1002 - We Tell You The Future

· 20 min read
[@ModernRogue] Summarizer
Video Bot

@ModernRogue - "Weird Things Podcast #1002 - We Tell You The Future"

Link: https://youtu.be/KRYjBPJoup0

Duration: 120 min

Transcript: Download plain text

Short Summary​

This milestone1,000th episode of the Weird Things podcast features hosts Andrew, Justin Robert Young, Dylan Theus, and Brian Brushwood debating the new Tesla Roadster's cold-gas thruster flying demo, rumored Tesla/SpaceX/xAI merger, and a wide range of AI topics from dirt-cheap on-device decision models and workflow automation to pig-butchering scams and fusion energy predictions. The hosts also critique Netflix and Amazon's expensive misfires like Electric State and Citadel, the Greta Gerwig Narnia trailer, and a viral Art of Play/Kelly Clarkson controversy, while noting ChatGPT's 1.2 billion weekly users and only 3% paid conversion.

Key Quotes​

  1. "I think Elon is insane."
  2. "I think before the end of this decade, I think we're going to see a sustainable fusion" (00:18:30)
  3. "Uh it mentions that the Jeb model is a system one model. If if it's been a minute since you've read thinking fast and slow, thinking fast and system one is your gut level where it's pure associations like you smell smoke. You your adrenaline rush because you know fire. At no point did you evaluate and think through why would I smell smoke? Could be this, could be that. Oh, maybe it's fire. You just instantly quote unquote knew it." (00:21:06)
  4. "Oh that that's an interesting gap that I saw a tweet about reminding people that six billion humans have never touched AI in any form. two billion only know it as a better version of Google and less than a billion people are actually using it for anything outside of just searches." (00:24:45)

Detailed Summary​

Weird Things Podcast – Episode 1,000 Detailed Summary​

Hosts and Production​

The 1,000th episode of the Weird Things podcast is hosted by Andrew, Justin Robert Young, Dylan Theus, and Brian Brushwood, with production handled by an intern Brian recently hired. Brian noted that the Weird Things Patreon feed had been fully caught up after five days of work.

  • The episode marks a milestone 1,000th installment, with hosts Andrew, Justin Robert Young, Dylan Theus, and Brian Brushwood.
  • Brian Brushwood hired an intern specifically to handle production and noted the Patreon feed was now fully caught up after a five-day push.

Tesla Roadster Flying Demo​

Elon Musk and Tesla unveiled the new Roadster at an invitation-only event the previous night, with Brian calling it a shocking display and noting the car itself looked appealing — unlike the Cybertruck. The demo featured the Roadster using cold-gas thrusters (compressed air) to perform a vertical jet, barrel roll, and various stunts over a lineup of Tesla vehicles.

  • The Roadster used cold-gas thrusters (compressed air) to perform a vertical jet over a lineup of Tesla vehicles and tractor-trailers, completing a barrel roll and landing with nobody inside.
  • It then drove up a multi-story ramp that loops vertically upside down, stayed attached via a suction system, flew off inverted, fired thrusters to right itself, and landed.
  • The car reportedly leapt over five cars without a ramp, had roughly 3–4 seconds of flying time, and was joked as "suborbital" at 45 minutes.
  • The Roadster's quoted 0-to-60 time is 0.7 seconds, which the hosts said is not being questioned.
  • The thrust system uses an onboard compressor (not nitrogen), refueling takes ~30 minutes, and the suction mechanism is vacuum-based requiring surface proximity.
  • Ars Technica ran the headline: "Tesla turns to 50-year-old stunt technology tries to roadster again."
  • Hosts debated replicability: Porsche could in theory task a rocket subsidiary but lacks the in-house reaction systems and "YOLO attitude" — comparing to Europe never building a Falcon rocket, though they could in 10 years.
  • One host argued the demo car was a shell made to look like a Roadster rather than the actual production car; viewers circulated fake/CGI claims similar to SpaceX landing skepticism.
  • Randy Pitchford attended the event and retweeted a David Copperfield "levitating Ferrari" comparison joke.
  • A joke circulated about the Roadster "solving parallel parking" by flipping over four cars.

Elon Musk Praise and Criticism​

Brian described Musk as both "insane/progressively crazier" and as one of the greatest entrepreneurs, businessmen, and engineering-team leaders of the 21st century. He also criticized specific Musk decisions while acknowledging Musk's unmatched ability to assemble physical engineering teams that break cost and capability barriers.

  • Brian described Musk as both "insane/progressively crazier" and as one of the greatest entrepreneurs, businessmen, and engineering-team leaders of the 21st century.
  • Brian criticized some Musk decisions including space-based data centers vs. inference and the pivot from Mars to the moon as expedient.
  • Brian argued Musk's ability to assemble physical engineering teams that break cost/capability barriers is unmatched by anyone else on the planet.
  • Musk previously claimed Cybertrucks would be watertight enough to paddleboard to South Padre Island.
  • Anecdotes shared: Musk once crashed an uninsured new McLaren with a passenger, joking "the funny thing is it's not insured"; a Nate Silver book described Musk driving with Peter Thiel, taking a corner too fast, catching air, and flipping the car in a coin-flip landing.
  • Sam Altman publicly said the Roadster demo was impressive; Musk is currently suing Sam and has been bitter toward him.

Tesla / SpaceX / xAI Merger Rumors​

Rumors circulated of a potential Tesla, SpaceX, and xAI merger that would require SpaceX to do a buyout of Tesla, given the massive valuation gap between the companies. A merger could trigger significant key man risk concerns given Musk's history of altering arrangements.

  • Rumors circulated of a potential Tesla, SpaceX, and xAI merger that would require SpaceX to do a buyout of Tesla.
  • Context: Tesla market cap ~$1.44 trillion; SpaceX valuation ~$1.9 trillion.
  • A merger could trigger key man risk concerns if Musk decides to test or alter the arrangement.

AI Workflow Discussion​

A speaker laid out a three-level AI framework for creators: (1) prompt AI to do what you'd do if you had time, (2) study best-in-field and have AI emulate the good/avoid the bad, and (3) identify the specific creative element that makes your work unique and infuse more of it yourself. The next one to three decades, he argued, will be defined by the unseen layer of what creators do once truly unrestrained.

  • The next one to three decades will be defined by the unseen layer of what creators do once truly unrestrained.
  • Used a fractal analogy of the coast of England — large media players have always had resources to carve content to its atomic unit.
  • Workflow example: records on Riverside's PX3 room, prompts the tool to find the last take, package it, source B-roll of Katie Zachariah and Anthony Scaramucci (plus a Zachariah Hannity clip), and produces a cut with a YouTube Shorts cover (misspelled, requiring correction).
  • Hosts canceled $5,000 of annualized subscriptions in four hours by linking AI to personal finance/budget tools.
  • "Dots" are ongoing AI agents that can run projects (distinct from chatbots), with a $100/month plan minimum.
  • OpenAI reports 1.2 billion people using ChatGPT every week.
  • Anthropic provides a free Claude tier but spent essentially $0 marketing free Claude in India because they don't want price-sensitive users as customers.
  • Meta's Muse was described as nowhere near as capable as Astra for computer use (gets confused, can't prioritize emails); the host has personal "trust issues" with Meta products and hasn't tried Muse.
  • A host's 3D Studio Max proficiency (inverse kinematics, volumetric lighting, specular highlights) has been frozen for ~25 years, requiring AI to operate Blender.
  • For Cordkillers, the host asked Blender to remake an entry-level animation, producing a stingers sequence with geometry over transparency featuring Tom Merritt; changed texture from gold shiny to brassy brushed metal — noting users still need to know "secret keywords" to describe visual changes.
  • Lost intro title sequence was made by J.J. Abrams in LightWave, with artifacts (~3 frames of pixelation); Kane Parsons (Backrooms creator) is deeply into Blender; James Cameron is also technically wired in.
  • A second speaker audited their operation, learned CapCut and Adobe Premiere, deliberately backed away approaching a perfectionism cliff, then produced 1,000 shorts in a couple of years before offloading the workflow to a robot.

AI Capability and Scams​

A recent interpretability paper argues researchers are on a good path to understanding what happens inside big models, and one GPT Pro session output more mathematical proofs than all mathematics teams had produced in the prior couple of years. The hosts debated whether even super-intelligent AI can be smarter than the collective sum of all humans, noting biological humans are calorie-for-calorie the most efficient system for converting solar energy into useful things.

  • A recent interpretability paper argues researchers are on a good path to understanding what happens inside big models.
  • One GPT Pro session output more mathematical proofs than all mathematics teams had produced in the prior couple of years, including decades-long work.
  • Humans take in roughly 10,000 experiences per day (up from ~20 ads a day historically), limiting how convincingly AI can predict wants.
  • A year ago there was panic that anything could be faked, but people quickly got good at shrugging off fakes (like the current Flying Car story).
  • Pig butchering scams are now bigger than cocaine in money involved; people avoid them not because they're smarter but because they recognize the signs.
  • The core exploit is loneliness and isolation — victims miss the companionship more than the money.
  • Scenario raised: 20% of net worth bequeathed upon death to a team running an AI companion mimicking a girlfriend/boyfriend, called terrifying and dangerous.
  • Audible original series "Only Fantasy" documents OnlyFans "chatters" hired by top creators for 24/7 availability — 2:30 AM availability is a mutually understood shared fantasy at higher tiers (6–7% rates cited).
  • The Chinese mafia black-market version of AI companionship already exists, described as "pigmenting" and the "biggest industry on the planet."
  • The best parasites slow the host without killing it; black-market scam operators extract maximum value before family intervenes.
  • Anecdote from ~25 years ago: a man in flip-flops at IHOP claimed his son at Goldman Sachs could get investors into hot IPOs, asking each person for $5K — recognized as a scam because legitimate IPO allocations don't work that way.
  • Friend responded "Yeah, I know. But what if it's not a scam?" — citing Robert Cialdini on how desire to believe overrides rational fraud analysis.
  • Andrew claimed he made more money using AI than last year, used AI to save on taxes, and suggests ~15 minutes a week of AI finance management could save tens of thousands per year (treasury bills vs. idle checking balance).
  • Andrew describes himself as terrible at bookkeeping (frequently paying penalties) and his investment style as lucky; Nvidia position up roughly 50–100%; SEP IRA maxed out; throws away quarterly tax installment coupons. Disclaimer: "Do not take advice from me folks."
  • Coined phrase: "Andrew does very well in an Andrew-shaped universe" — popularized by economist Tyler Cowen.

TV and Pop Culture​

The hosts reviewed several ongoing and recent TV series, with Andrew praising Lanterns for "sticking the landing" while another speaker lost interest around episode five citing a "comically bad" clip about female law enforcement. Andrew also discussed his rewatch of Star Trek: The Next Generation (debuted 1987), praising LeVar Burton as Geordi La Forge.

  • Lanterns (DC series): Andrew finished season one and thought it "stuck the landing"; another speaker watched through ~episode five before losing interest, citing a clip about "female law enforcement" he found "comically bad" and warning "don't give the people who are against the ammunition."
  • Compared to Watchmen TV show — criticized a Lanterns character being "the smartest woman in the world because her father is genetically" so; Andrew preemptively hears from "the real Justin" about Damon Lindelof criticism.
  • Andrew is rewatching Star Trek: The Next Generation (debuted 1987), currently in the first season, praising LeVar Burton as Geordi La Forge (Burton had been acting over a decade before TNG); jokes about a "guy in a skirt" used to signal the future vs. modern Star Trek trying to be edgy.

Art of Play Controversy & Cardistry​

Art of Play, started by Dan and Dave Buck, was discussed as an adult toy shop that helped revive magic by bringing 15-year-old kids into conventions and is credited with creating cardistry as a new branch of magic. A Kelly Clarkson incident at Art of Play was described as an unintentional celebrity crash-out that generated publicity the hosts could not have engineered themselves.

  • Art of Play, started by Dan and Dave Buck, was discussed as an adult toy shop by Pirate Wires and a force that helped revive magic by bringing 15-year-old kids into conventions.
  • Dan and Dave Buck are credited with creating cardistry, a new branch of magic, in their lifetime.
  • The hosts referenced a magician named Matty, born with no arms and no legs, who fooled Penn and Teller.
  • A Kelly Clarkson incident at Art of Play was described as an unintentional celebrity crash-out that generated publicity the hosts could not have engineered themselves.
  • The incident was framed as a flashpoint moment where multiple narratives (woman vs. patriarchal magic, muggle vs. wizard, small business vs. mega elite) all projected onto one event.
  • A co-host and Justin coined the slogan "hurt butts hurt butts" to describe situations where two aggrieved parties only amplify each other's harm.
  • A host argues AI cannot replicate the human credibility-building required to create such a viral moment, citing Tilly Norwood (an AI avatar actress) as a counter-example.
  • Tilly Norwood reportedly glitched during an interview by running on a Chinese model and started speaking Chinese.
  • The hosts reference The Diamond Age, noting the "real young lady's primer" uses a persistent human voice actor to make you care, while lesser versions are LLM-replicable local models.

Small AI Models & Cost Economics​

Small AI models are described as dirt cheap and super fast, suitable for tasks where sending to larger models like Astra or Claude is unnecessary. The hosts cite examples including an 18 MB browser speech model and Apple's 3 billion-parameter on-device phone model, with the pushback that humans recalibrate and expect more as capability grows.

  • Small AI models are described as dirt cheap and super fast, suitable for tasks where sending to larger models like Astra or Claude is unnecessary.
  • Using a small model to follow an entire podcast in real time would cost fractions of a cent per episode.
  • The hosts cite a speech recognition model that is only 18 megabytes and can run in the browser at pretty good quality, roughly the same size as the MP3 of Hey Jude.
  • Apple's on-device AI model has 3 billion parameters, can run on phones, is described as pretty good today, and is expected to be amazing in five years.
  • A speaker pushes back on the idea that humans will settle for good-enough cheap models, arguing people recalibrate and expect more as capability grows.

AI as Infrastructure, Not Gadget​

The hosts distinguished between a "gadget/duad" and a depended-upon "layer" of civilized society, arguing AI is making that transition in real time. They noted that hidden production information (e.g., which producer kept a writer on set doing on-time rewrites) is becoming more valuable in an AI age and is not captured by IMDb.

  • Hosts distinguish between a "gadget/duad" and a depended-upon "layer" of civilized society, arguing AI is making that transition in real time.
  • The "Jeb model" is described as a "system one" model referencing Kahneman's Thinking Fast and Slow framework; its cheapness comes from not running any simulations of cognition.
  • Hidden production information (e.g., which producer kept a writer on set doing on-time rewrites) is argued to become more valuable in an AI age and is not captured by IMDb.
  • Video game creators must now think more like magicians because clever ideas shown in trailers can be vibecoded in Roblox the next day.

Netflix & Amazon Adaptation Critique​

The hosts criticized recent expensive adaptations, with one calling the Greta Gerwig Narnia trailer looking like "AI slop" despite not being AI-generated, and Amazon's Rings of Power failing to understand what an orc is in Tolkien's work. Netflix's Electric State and Citadel were cited as among the most expensive productions ever that nobody watched.

  • The Greta Gerwig Narnia trailer was discussed, with one host saying it looked like "AI slop" even though it was not AI-generated, and questioning whether character changes preserve or undermine the source material.
  • A speaker argued Netflix removed the Christ-figure element of Aslan, comparing the change unfavorably to Peter Jackson compressing The Hobbit into Lord of the Rings while keeping characters true to source.
  • Amazon's Rings of Power was criticized for not understanding what an orc is, with orcs representing entropy in Tolkien's work.
  • Netflix is described as super data-driven, with casting Kevin Spacey in House of Cards cited as a data-driven decision that ignored risk.
  • Electric State was called one of the most expensive movies ever that nobody has seen, directed by the Russo brothers.
  • Citadel was called the most expensive TV series ever made, with American, European, and Indian versions, and nobody watched it on Amazon.

Global AI Adoption & Business Strategy​

A host is bullish on AI, traveling through Ecuador and Colombia and planning Brazil work to support Latin America AI development. The hosts argued that roughly 6 billion humans have never touched AI in any form, 2 billion only know it as a better Google, and less than 1 billion use it beyond searches.

  • A host is bullish on AI, traveling through Ecuador and Colombia and planning Brazil work to support Latin America AI development.
  • Anthropic will need an ad-supported or free product tier to bring the rest of the planet on board, the host argues.
  • Facebook's attention-maximization model is better served by advertising to upper-middle-class users than to the "bottom billion" poor, and AI could uplift them.
  • A host stated he is exiting Meta/Facebook ads because their algorithmic pricing is too effective at predicting margins.
  • Roughly 6 billion humans have never touched AI in any form; 2 billion only know it as a better Google; less than 1 billion use it beyond searches.
  • ChatGPT has 1.2 billion weekly users, with only about 3% on the paid tier.
  • Much of ChatGPT ad usage consists of lifestyle queries (shoes, clothes) rather than B2B services, and it remains early days for advertising on the platform.
  • Scam Stuff has had 180,000 customers over 15-plus years, creating 180,000 emails sortable for legally compliant marketing.
  • Selling niche products (e.g., a book) is harder than broad products (e.g., Audible), and search volume for niche interests can be tiny, e.g., only about 100 people searching a given query in a day.
  • Advertising is often hard to quantify and multi-stage, requiring tracking recognition across touchpoints.

Mall Futurism and Energy Predictions​

Hosts went to a mall to shoot at a video arcade location and arrived to find plastic sheets over everything (arcade closed); malls fail due to high real estate, overhead, energy, labor costs and weak "vibes." They predicted energy costs will fall tremendously by the end of the next decade, driven primarily by fusion and supplemented by cheap solar.

  • Hosts went to a mall to shoot at a video arcade location and arrived to find plastic sheets over everything (arcade closed); malls fail due to high real estate, overhead, energy, labor costs and weak "vibes."
  • Suggests repurposing malls as retirement communities for Gen Xers, keeping arcades and food courts; argues removing teenagers from mall labor via minimum wage laws reduces output.
  • Predicts energy costs will fall tremendously by the end of the next decade, driven primarily by fusion and supplemented by cheap solar; sustainable fusion reaction demonstrated within the next four years, before the end of the current presidential administration.
  • Envisions AI-coordinated overnight store turnovers in future malls — walls repainted, new fixtures, complete redesigns every two weeks.
  • Brian described AI being applied to materials science, mathematical proofs, and physical sciences as a complicated engineering problem.
  • Evolution of agent assistance: manual timer-based creation → always-on watching agent anticipating requests → proactive agents that execute unprompted.

Decision/Routing Models​

Decision models (or routing models) are stripped-down models that output yes/no decisions rather than text — costing fractions of a cent per episode to run a full podcast. Jev is a model from former OpenAI researcher Dogo and his team at Typesafe, which just completed a multi-billion dollar funding round.

  • Decision models (or routing models) are stripped-down models that output yes/no decisions rather than text — dirt cheap, super fast, costing fractions of a cent per episode to run a full podcast.
  • Can play games (Snake), control vehicles, perform other small decision-based tasks.
  • Jev is a model from former OpenAI researcher Dogo and his team at Typesafe, which just completed a multi-billion dollar funding round; processes audio, images, and likely other modalities soon.
  • Described as a System 1 model (Kahneman's Thinking Fast and Slow) where gut-level associations replace deliberation.
  • Compared to fuzzy logic popular in the 1990s — chair vs. table blending into middle ground.
  • OpenAI released a decisions API; Microsoft and Hugging Face have models in this category.

Apple Hardware & ChatGPT Features​

The hosts discussed recent Apple hardware shortcomings and new ChatGPT features, noting that the iPhone Air's battery life is terrible (dying around 6 PM) and the best iPhone camera remains on the Pro model. ChatGPT now includes a "slides" feature accessible via Spaces that can generate slide decks from inputs like research papers.

  • The iPhone Air's battery life is described as terrible, dying around 6 PM, and no updated Air has been announced.
  • The best iPhone camera remains on the Pro model, not the thin Air/Duo, so camera-focused users are unlikely to choose it.
  • ChatGPT now includes a "slides" feature accessible via Spaces that can generate slide decks from inputs like research papers.

Notable Surprises and Takeaways​

The episode covered a remarkable range of predictions and revelations, from the fact that the Tesla Roadster demo was publicly telegraphed by Musk to claims that sustainable fusion could be demonstrated within four years. The hosts emphasized that AI decision systems can now run entire podcasts for fractions of a cent and that pig-butchering scams have surpassed cocaine in financial scale.

  • The Tesla Roadster demo is not a secret — Musk publicly stated cold gas thrusters and flying would happen.
  • A Tesla/SpaceX/xAI merger rumor with massive valuations ($1.44T and $1.9T) and key-man implications.
  • AI can mathematically outperform all human teams combined in a single session.
  • Pig butchering scams surpass cocaine in financial scale — exploitation of loneliness, not tech.
  • Sustainable fusion within four years — energy cost collapse prediction.
  • Decision/routing models like Jev run entire podcasts for fractions of a cent.