Episode #574: Evals, Ontologies and the Unmappable World of Business

In this episode of the Crazy Wisdom Podcast, host Stewart Alsop sits down with Ryan Marsh of thestack.io to explore what it really takes to build production AI systems. They discuss how production AI has evolved from simple prompt-to-API demos into complex systems requiring evaluation suites, human-in-the-loop feedback mechanisms, and sophisticated approaches to handling context and data retrieval. The conversation covers the challenges of domain mapping, the fundamental difficulty of translating messy human business processes into structured systems, and the role of ontologies in AI development. Ryan and Stewart also examine the limitations of LLMs, the debate between specialization versus generalization in AI models, consciousness and cognition in system design, and the regulatory landscape facing AI companies. They touch on infrastructure constraints, the democratization of AI through open source models, and whether we'll eventually hit a ceiling where human intelligence can no longer distinguish between increasingly capable AI models. Key Insights1. Production AI systems today fundamentally differ from demos through their reliance on comprehensive evaluation suites that function like unit tests to measure and maintain quality, though they cannot be as deterministic. The key distinction is that production systems require clearly defined metrics for what good looks like, along with feedback mechanisms that allow the system to evolve over time. Without this foundational understanding of success metrics and continuous improvement processes, a system is not truly production-ready regardless of how many users it serves.2. The fundamental challenge in building production AI systems is not the technology itself but rather mapping business domains into structured formats that models can work with effectively. This problem of translating messy, subjective human processes and language into precise specifications has plagued software engineering for decades. Different people within organizations use the same words to mean completely different things, and humans naturally operate with assumed context and imprecision that must be explicitly defined for AI systems to function reliably.3. Large language models excel at generalization but struggle with specialization, which creates friction in production environments where specific outputs or styles are required. While they can code in any programming language, getting them to write code exactly the way a particular engineer wants remains extremely difficult. This explains why professional documentation and specialized coding tasks often require extensive prompting and fighting with the models, as they naturally gravitate toward their trained patterns rather than highly specific user preferences.4. Modern production AI systems primarily solve classification problems wrapped in natural language interfaces rather than requiring true open-ended cognition. The models work best when they can leverage reasoning over provided information to make verifiable decisions, but they still lack common sense despite their vast knowledge. Success comes from teaching models everything about your specific domain and what good and bad outcomes look like, rather than relying solely on their general intelligence.5. Context retrieval in production AI systems is fundamentally a data storage, search, and retrieval problem that has been solved many different ways throughout computing history. The appropriate solution depends entirely on the type of data being accessed, whether through vector databases, graph databases, relational databases, or even simple text search. The harnesses and frameworks for orchestrating AI agents have matured significantly, making the real challenge the quality and structure of the data being fed to these systems.6. Human-in-the-loop feedback mechanisms are essential for production AI because models will inevitably encounter situations they have not been trained to handle. When confidence is low or novel scenarios appear, systems should flag these for human review rather than proceeding blindly. The feedback provided during these interventions must be captured and scored so it can be incorporated into the permanent behavior of the system, creating a continuous improvement cycle similar to how model vendors perform reinforcement learning on their base models.7. The rush to regulation in the AI industry is driven primarily by the fact that these companies are currently completely exposed to existing consumer protection and liability laws with no legal precedents to protect them. If AI agents cause harm, companies could be sued into oblivion under current law. By establishing compliance frameworks through regulation, these companies can create carve-outs and exemptions that limit their liability when they follow prescribed rules, similar to how heavily regulated industries like airlines and banking have become nearly impossible to enter due to compliance requirements. Timestamps00:00 Welcome and introduction to Ryan Marsh discussing production AI systems and what companies need to understand about building them at scale05:00 The cognitive load challenge of working with invariants and probabilistic systems, discussing how LLMs function like PhD students without common sense10:00 Building eval suites to measure AI performance, handling the long tail of edge cases in complex contracts using human-in-the-loop feedback systems15:00 Context as a data retrieval problem and the fundamental challenge of mapping business domains when humans struggle with imprecision and ambiguous language20:00 How different departments use the same words with different meanings and why LLMs generalize well but don't specialize effectively for specific coding styles25:00 The ontology debate and intractable problem of mapping subjective human systems to rigid structured formats with diminishing returns on perfect mapping30:00 Why humans struggle distinguishing what is from what ought to be and the challenge of creating SOPs when companies lack updated documentation35:00 The liability exposure AI companies face and why they're begging for regulation to protect themselves from existing consumer protection laws40:00 Open source knowledge transfer between Chinese and American labs, chips and power as the real constraint not algorithms for frontier models45:00 Reaching intelligence ceiling where specialists can't distinguish state-of-art models and fundamental physical laws limiting LLM scaling through layered optimization strategies LinksWebsiteX

Stewart Alsop sits down with Antoine Marcel, CEO and cofounder of Flourish, to explore the future of home robotics and what he calls "solarpunk" automation. Antoine discusses why Flourish is building wheeled robots with arms instead of following the humanoid hype, arguing that the expensive legs and teleoperation approach isn't practical for real consumers today. The conversation ranges from the philosophy of robots that disappear when you're home (rather than becoming constant companions) to the bureaucratic nightmares of modern life that robotics might actually solve. Antoine explains how their $3,500 robots can be trained in thirty minutes using just your phone to handle personalized household tasks, and why saving people two hours a day—not just ten minutes—is the threshold for consumer adoption. They also dig into whether Paris will become a solarpunk or cyberpunk city, why Apple represents solarpunk design principles, and how the first 50 Flourish units are about to ship worldwide at flourish-robots.com. Key Insights1. The core innovation behind Flourish is building mobile-based robots with arms but no legs, moving on wheels rather than attempting humanoid locomotion. This design choice stems from practical considerations since even humanoid robots cannot reliably handle stairs, and legs represent approximately twenty thousand dollars of the cost in humanoid platforms. By eliminating legs, the company can focus on delivering functional value to consumers today rather than developing research products. The founder emphasizes that accessibility and usefulness are paramount, rejecting the demonstration-focused approach that dominates the robotics industry where companies raise billions on teleoperated demos that never ship to actual consumers.2. The founder's background in automation across logistics, software, and sales led him to recognize that the most painful tasks in life occur outside the computer. After automating digital work processes, he realized that returning home after a long workday to clean, cook, and tidy remained exhausting and unaddressed by existing technology. This insight, combined with advances in physical AI over the past three years, drove him to develop the first robot prototype that runs in his own home, constantly testing and iterating based on real-world use cases like navigating doorways and picking up scattered items from floors.3. The company is building toward a solarpunk vision rather than cyberpunk future, where robots disappear from conscious awareness and operate silently in the background. The founder envisions robots emerging from closets when people leave home, performing cleaning and tidying tasks, then retreating before residents return. This approach contrasts sharply with the prevalent humanoid robot vision featuring teleoperated machines with cameras potentially monitored by remote operators. The goal is reducing technology's presence in daily life to create more time for family, creativity, and meaningful activities rather than introducing intrusive robotic presences into living spaces.4. Consumer robotics faces fundamental challenges that explain why almost no companies successfully sell to end users despite massive investment in the sector. Industrial robotics offers clearer paths to revenue, while consumer products require extensive market education, safety certifications, and must genuinely change lives rather than provide marginal improvements. The founder argues that saving people ten or thirty minutes daily is insufficient and forgettable, but saving two hours daily creates transformative value that sustains product adoption. This high bar for meaningful impact explains why the consumer robotics market remains largely unaddressed despite technological capabilities.5. The business model centers on teaching robots specific tasks through a thirty-minute training process using a smartphone, then fine-tuning AI models for those particular workflows. Rather than pursuing general physical AI that can handle any task, Flourish enables users to create custom automations for their specific needs, like retrieving items from the refrigerator or collecting clothes for the washer. This approach acknowledges that general physical AI is not yet ready while still delivering practical value through task-specific learning. The company plans to launch fifty units at thirty-five hundred dollars, shipping worldwide, with rapid sellout expected due to existing demand.6. The robot extends vertically to reach from floor level to table height, enabling tasks like cleaning surfaces and accessing shelves while maintaining a compact mobile base that can navigate through doorways. Current limitations include lack of waterproofing, preventing dishwashing tasks, and challenges with delicate object manipulation requiring advanced tactile sensing. The founder uses first-generation prototypes in his own home, scheduling the robot to pick up scattered items like socks and trash throughout the day. This real-world testing revealed critical design issues like initial prototypes being too large to pass through standard doorways, problems only discoverable through actual home deployment rather than warehouse testing.7. The rise of AI-powered robotics fundamentally transforms hardware development by enabling software engineers without traditional robotics expertise to launch robotics companies after just weeks of learning. This democratization stems from downloadable models from well-funded research companies like Physical Intelligence, whose work the founder incorporates into Flourish robots. This represents a shift from the industrial age's capital-intensive factory model to an information age paradigm where substantial capability can be rapidly deployed through software. However, this transformation raises questions about competition intensity, market education for skeptical consumers, and whether the benefits will broadly distribute or concentrate among those with will and intelligence to build while others remain excluded and fearful. Timestamps00:00 Stewart welcomes Antoine Marcel, CEO of Flourish, who explains they're building accessible home robots with wheeled bases and arms rather than expensive humanoid forms for practical domestic use.05:00 Antoine discusses testing prototypes in his home for picking up clothes and trash, explaining how real-world testing revealed issues like doors being too narrow for early designs.10:00 The conversation shifts to how AI hype enables useful research, with Antoine noting he uses physical intelligence models in his robots that wouldn't exist without funding bubbles.15:00 Antoine emphasizes robots must save users two hours daily to matter, explaining why consumer robotics is harder than industrial applications due to certification requirements and market education needs.20:00 Discussion of solarpunk philosophy versus cyberpunk futures, with Antoine envisioning robots that hide away when not needed, emerging only to clean and organize while families enjoy life.25:00 They explore how different cities might adopt solarpunk or cyberpunk approaches, comparing bureaucracy as inherently cyberpunk while discussing Argentina's potential as a solarpunk capital.30:00 Antoine positions Flourish like early Apple, making complex robotics simple and accessible, rejecting teleoperated solutions in favor of onboard AI that learns from users.35:00 The business model discussion centers on unit economics, with Antoine arguing domestic robots should monetize time saved rather than attention captured like social media platforms.40:00 Antoine advocates against robot screens and rental models, favoring voice control and ownership so robots learn individual routines and preferences through personalization.45:00 They discuss robots disappearing from view when unneeded, contrasting home automation with dark factory models, before exploring robot aesthetics and art possibilities.50:00 Antoine announces Flourish's launch of fifty units at $3,500 that users can train in thirty minutes using phones, teaching specific tasks the AI then fine-tunes for daily execution. LinksFlourish Robots

In this episode of the Crazy Wisdom Podcast, host Stewart Alsop sits down with Neil Bortolus, founder of Esteem Biomechanics and Stewart's personal trainer for the past year and a half, to explore the intersection of functional movement, human biology, and emerging technology. Their conversation ranges from the principles of biomechanics and the evolutionary blueprint of human movement—standing, walking, running, and throwing—to the dangers of modern movement practices that prioritize aesthetics over function, the role of fascia and connective tissue in chronic pain, and why most gym exercises fail to respect our biological needs. They tackle bigger questions about where technology is taking us, from AI and LLMs replacing human cognitive work to transhumanism and whether we're headed toward a cyberpunk dystopia or a solarpunk regenerative future, all while emphasizing the importance of first principles thinking, the need for humans to remain in the loop, and creating environments that promote biological and mental regeneration rather than degeneration. Visit Neil's website atesteembiomechanics.comTimestamps00:00 Stewart introduces Neil Bortolus, founder of Esteem Biomechanics and his trainer for a year and a half, discussing their interesting conversations about movement and health05:00 Discussion shifts to negative and positive pressure in the body, how athletes naturally create proper core tension, and the differences between training high-performance athletes versus clients with cerebral palsy or stroke recovery10:00 Stewart shares his journey through yoga and discovering modern postural yoga's surprising origins in late 1800s British calisthenics mixed with Indian anticolonial movements, questioning why modern movement practices are fundamentally flawed15:00 Neil explains functional patterns' first principles approach, emphasizing standing walking running and throwing as evolutionary movement patterns that should guide all exercise design20:00 Conversation explores Brian Johnson's technology-focused longevity experiments versus optimizing natural biological function, questioning whether peak biology or technology interventions lead to better health outcomes25:00 Neil describes environmental factors like Southern California and Hawaii providing optimal conditions for human health through fresh food, sunlight, and natural grounding environments year-round30:00 Discussion of solarpunk versus cyberpunk futures, examining how technology could either support biological health or create dependency, with Stewart sharing his Invisalign experience as cautionary tale35:00 Neil and Stewart analyze education systems, influencer marketing, and Jacques Ellul's concept of technique autonomously taking over human goals and replacing both muscle and mental labor40:00 Neil questions Stewart's first principles for approaching technology, leading to discussion about using AI to accomplish tasks faster while maintaining human oversight and decision-making authority45:00 Stewart explains the black box problem with LLMs and neural networks, admitting he's built applications he can't fully understand, questioning whether understanding mechanisms matters if outputs work50:00 Debate about benevolent AI governance using El Salvador's transformation as analogy, questioning whether society should surrender control to AI systems for farming or other critical infrastructure55:00 Final discussion on determinism versus free will, alignment problems in AI development, and Neil's concept of regenerative versus degenerative environments for optimal human biological and mental healthKey Insights1. Movement training must be adapted to individual needs and capabilities rather than following a one-size-fits-all approach. Neil explains that working with clients ranging from those with cerebral palsy and stroke recovery to high-performance athletes requires fundamentally different techniques. While stroke and cerebral palsy clients experience every improvement as monumental, such as being able to walk a mile, elite athletes already possess optimal biomechanics and need only tiny refinements. The challenge lies in keeping athletes motivated through detailed adjustments while helping severely impaired clients build basic functions from the ground up.2. Modern fitness and yoga practices are built on flawed first principles that prioritize aesthetics over functional movement. The conversation reveals how modern postural yoga was actually invented in the late 1800s through a combination of British calisthenics and Indian anticolonial movements, not ancient wisdom as commonly believed. Similarly, conventional gym exercises and bodybuilding focus on how bodies look rather than how they function evolutionarily. The key insight from Functional Patterns methodology is that if a technique cannot help someone who is broken, it should not be applied to someone who is healthy. This principle exposes how much of modern fitness is based on working with people who are already genetically gifted rather than developing universally applicable methods.3. Human movement should be designed around four fundamental patterns that reflect our evolutionary blueprint: standing, walking, running, and throwing. These movements represent the biomechanical demands that shaped human evolution and remain essential for optimal health. Hunter-gatherer tribes that still exist today demonstrate this principle, as their members typically exhibit athletic builds, low body fat, minimal injury rates, and few mental health disorders despite having no access to modern gyms. The critical insight is that 80 percent of human movement occurs horizontally through space rather than vertically, yet most conventional exercises focus on vertical force vectors like squatting and deadlifting, which do not align with how humans actually move in nature.4. Technology should be used to optimize our environment for biological health rather than replace biological function entirely. The discussion contrasts two approaches to human enhancement: using technology to create better living conditions versus using it to bypass the need for healthy habits. Examples include using renewable energy to reduce electromagnetic frequency exposure, ensuring clean water access, and designing indoor lighting that mimics natural sunlight patterns. The key is maintaining environments that support regenerative rather than degenerative processes. This regenerative approach considers whether food, exercise, and environmental factors promote cellular regeneration and joint health, rather than simply preventing obvious harm.5. The rapidly changing technological environment is outpacing human biological adaptation in unprecedented ways. While human populations historically had thousands of years to develop epigenetic adaptations to their environments, such as Icelandic peoples developing larger bodies for heat retention or Sub-Saharan Africans developing efficient heat expulsion, modern society changes too quickly for such adaptation. The past two hundred years have seen dramatic shifts, and the pace is accelerating rather than slowing. Most dramatically, the media environment is changing fastest of all, particularly with AI development creating a cognitive flywheel where staying current requires constant attention to collective knowledge networks like Twitter and Reddit, leaving those outside these loops increasingly disconnected from technological reality.6. AI and large language models represent both opportunity and risk depending on whether humans maintain agency in their use. The conversation distinguishes between using AI as a tool to accomplish tasks more efficiently while keeping humans in control versus allowing AI to make decisions autonomously. The example given involves using AI to complete projects that previously required hiring people, eliminating interpersonal problems while still maintaining human oversight and learning. However, there is concern that many people are beginning to ask AI how to make every decision, effectively abdicating their agency. The critical variable in any AI interaction is actually the human prompter and their instructions, yet this is often overlooked by AI developers who believe in determinism and lack of free will.7. Modern society operates within a deterministic framework where free will is exercised primarily in choosing which constrained environment to inhabit. Using the example of an accountant earning two hundred thousand dollars annually with a four hundred thousand dollar mortgage and three children in private school, Neil illustrates how most daily decisions are predetermined by earlier choices about environment and commitments. True free will exists at major life forks, such as choosing a career path or living situation, but day to day existence operates within deterministic constraints created by those choices. This insight applies to both human development and AI alignment, suggesting that the solution is not giving AI ultimate freedom but rather constraining its environment so that available options consistently benefit humanity, similar to how Functional Patterns constrains exercise choices to only those respecting biological blueprints.

In this episode of the Crazy Wisdom Podcast, host Stewart Alsop sits down with Daniel Segundo to discuss why today's social networks have evolved into entertainment networks and how Daniel is building Bien, a social app focused on depth over breadth. Daniel shares his journey from trading commodities to creating a platform designed to keep users genuinely connected to their inner circle, inspired by insights gained while meditating in caves in India. The conversation covers everything from the problems with the "like button" and algorithmic feeds to Daniel's contrarian approach of removing public-facing engagement metrics and using voice notes as the primary interaction method. They also explore the challenges of building consumer social products, the impact of AI on development, and Daniel's philosophy drawn from Eastern mysticism and his study of Ramana Maharshi. To learn more about Bien, visit bien.social or follow Daniel on Twitter at @DanielPSegundo.Timestamps00:00 Daniel introduces Bien as a real social network, contrasting entertainment networks disguised as social platforms05:00 Building with AI tools allowed bootstrapping the app before hiring a five-person team for professional development10:00 Social networks evolved into entertainment networks optimized for engagement and advertising rather than genuine connection15:00 The advertising model incentivizes dopamine delivery and algorithmic content over posts from actual friends20:00 Removing the like button and public comments forces deeper interaction through private voice notes instead25:00 Social graphs become stale over time because unfriending feels awkward, so people curate for weakest relationships30:00 Matrix and Nostr protocols enable interoperable messaging, suggesting a Cambrian explosion of new social apps35:00 Depth matters more than metrics—success means users actually feel closer to their inner circle40:00 Voice notes combined with photo sharing mimics natural storytelling behavior when friends gather in person45:00 Meditation in Ramana Maharshi's caves in India inspired reflections on silence, self-inquiry, and building products mindfullyKey Insights1. Daniel Segundo was inspired to create Bien after recognizing that existing social networks excel at breadth but fail at depth of connections. While meditating in caves in South India, he went through his phone contacts and realized only about 5 percent of his 1,700 contacts were people he actually wanted to stay close to, with the remaining 95 percent being noise. He saw that tools like Facebook, Instagram, and WhatsApp have solved the problem of maintaining surface-level relationships across broad networks of people, but there was no good tool for maintaining deep connections with your inner circle beyond basic messaging apps like iMessage or WhatsApp, which still suffer from having too many contacts with access.2. The fundamental problem with current social networks is that they have become entertainment networks rather than true social networks, driven by the advertising business model. Instagram only shows content from your actual friends about 7 percent of the time, with the rest being algorithmically suggested content designed to maximize engagement and ad revenue. The like button, while useful for maintaining broad social graphs and understanding your status in larger groups, reduces complex human relationships to superficial interactions. This creates what Segundo calls a circus-like experience where you are constantly bombarded with entertaining content rather than meaningful updates from people you care about, and the advertising model incentivizes this transformation because the product is ultimately about delivering dopamine to keep people scrolling.3. Bien removes traditional social media features like public likes and comments, replacing them with private voice note responses capped at 60 seconds. The theory is that hearing three voice notes from close friends provides more genuine connection than receiving 300 likes on a post. This design choice stems from recognizing that while the like button serves a useful function for breadth-first social networks where people want quick validation from large peer groups, it is completely inadequate for depth-first relationships. By forcing more intimate and effortful interaction through voice, Bien attempts to upgrade the quality of connection rather than the quantity, building off existing habits like sending voice notes to close friends rather than trying to create entirely new behaviors.4. The traditional follower-following model creates stale social graphs because people are uncomfortable unfriending or unfollowing others as relationships naturally evolve over time. Instead of this model, Bien uses a system where each user maintains a private list of people they want to stay close to, and if two people have each other on their lists, they see each other's content. This eliminates the social awkwardness of unfriending someone while allowing social graphs to naturally evolve. The current model forces people to curate their sharing based on the weakest relationship in their network because they do not want to offend distant connections or go through the work of managing close friends lists, which results in less authentic sharing overall.5. Segundo is building Bien on a paid subscription model rather than an advertising model to align incentives directly with users. He believes no one can build a compelling social network on the advertising model anymore because it creates conflicting incentives where the company must serve both advertisers and users, ultimately degrading the product into what he calls slop. The straightforward deal with customers is that if the product cannot provide value equivalent to the price of a cup of coffee per month through premium features, then there is no justification for the business to exist. This allows Bien to focus solely on whether users feel closer to their inner circle after using the product rather than optimizing for maximum scrolling time.6. Segundo bootstrapped the initial development of Bien using AI coding tools like Replit before hiring a team of five people to develop it professionally. He notes that with current AI tooling, if you can think clearly and understand the problems you are trying to solve, you can build solutions much more cheaply than even five or ten years ago. He eventually hired a designer and iOS and web developers to do things properly with a do it right or do not do it at all mindset. The app is currently iOS-only, built in Swift for a premium experience, with plans to potentially build a React Native version for Android if there are stronger signs of product market fit.7. Segundo's background in trading commodities and futures, along with his deep interest in Eastern philosophy and self-inquiry practices, significantly influences his approach to building Bien. He spent time studying Ramana Maharshi and practicing meditation in caves at Tiruvannamalai in South India, which he says has a profound impact not just on spiritual practice but on thinking about product development and participating in markets. His experience with a major trading drawdown in 2025 after running up a small account 25 times led him to this period of reflection that ultimately resulted in the conception of Bien, demonstrating how his various interests in trading, philosophy, and technology intersect in his approach to building a depth-first social network.

In this episode of the Crazy Wisdom Podcast, host Stewart Alsop sits down with TJ Marbois, founder of Tobiko, for a wide-ranging conversation that spans LLMs, data sovereignty, knowledge management tools like Obsidian, and Terence McKenna's ideas about an increasingly weird future. They explore how AI is simultaneously centralizing power through data control while decentralizing software development capabilities, allowing more people to build their own tools and escape big tech ecosystems. Drawing on his experience at Apple's special projects group (the team that built the iPod and later iPhones), TJ discusses his vision for personal AI assistants—what he calls the "R2-D2 belongs to you" principle—where advanced technology serves individuals rather than corporations. The conversation touches on everything from quantum encryption and local manufacturing with ESP32 microcontrollers to the economics of future society, biometric data unions, and why distributed trust matters more than ever as we approach what both describe as an increasingly strange technological inflection point. Visit Tobiko's site at tobiko-pbc.ghost.ioTimestamps00:00 Stewart introduces TJ Marbois from Tobiko, discussing LLMs, data sovereignty, knowledge management, and the tension between centralization and decentralization in AI05:00 TJ explains the R2-D2 belongs to you concept and why personal AI assistants need maternal alignment, caring about humans like mothers care for children10:05 Discussion of how LLMs will shrink and improve while emphasizing the sentient loop concept for building loyal personal AI agents rather than corporate controlled systems15:00 Exploring digital nervous systems for humanity, social networks as infrastructure, and humans as cavemen with iPhones navigating unprecedented technological complexity20:00 TJ discusses data unions for sovereign data ownership, inverting insurance models where AI helps extend healthspan, and maximizing truth seeking through collective data25:30 Money as social construct, crypto enabling value system reengineering, and working toward Star Trek's post scarcity replicator economy from the ground up30:40 Capital formation challenges, corporatism versus true capitalism, and pessimism about current systems reaching limits while seeking new models35:00 Solar power democratization, ESP32 microcontrollers enabling local manufacturing, and the replicator future through distributed maker communities and open source robotics40:00 Science fiction as roadmap, Isaac Asimov and Arthur C Clarke warnings, and building collaborative futures rather than centrally controlled dystopias with useless eaters45:00 Encyclopedia to LLM transition, information quality concerns, local training importance, and NVIDIA's incentive to put GPUs everywhere for personal AI agents50:00 Corrupted financial incentives, protecting vulnerable people from technological exploitation, and benevolent technologists building systems that honor humanity and children55:00 Hardware validation for owned robots, Starlink dependencies, assembly language abstraction toward natural language programming, and preventing AI escape scenarios58:00 Tobiko as AI toy company building sentient loop interactions similar to Xerox PARC GUI moment, emphasizing maternal AI alignment and cross cultural human collaborationKey Insights1. The concept of R2D2 belonging to you represents a critical vision for the future of artificial intelligence and personal technology. When thinking about robots and AI assistants that follow us around and know everything about us, the question of ownership becomes paramount. These systems will know incredibly intimate details about our lives, from our health data to our daily habits, and if they are controlled by centralized corporations or governments rather than individuals, we lose fundamental sovereignty over our own information. The science fiction of Star Wars and Star Trek provides roadmaps for how we should think about these technologies, showing us both the positive possibilities and the warnings we need to heed about centralized control.2. Local AI models and distributed computing represent a pathway to technological sovereignty that is becoming increasingly viable. While large language models currently run primarily in centralized data centers, the technology is rapidly advancing toward a future where powerful models can run locally on personal computers and devices. This shift is crucial because it means individuals can have full control over their AI assistants without relying on API calls to external servers. Combined with open source infrastructure and open weight models, this creates the foundation for truly personal AI that cannot be controlled or monitored by external parties, whether corporations or governments.3. Data unions and collective data ownership offer an alternative model to current centralized data collection practices. Rather than having individual data harvested by large tech companies who profit from it, the concept of data unions suggests people could collectively pool their data for specific beneficial purposes while maintaining ownership and control. For example, in healthcare, millions of people could share anonymized biometric data to train medical AI systems that are incentivized to keep people healthy rather than treat them when sick, inverting the current insurance model to align incentives with actual health outcomes rather than profit from illness.4. The democratization of manufacturing and robotics through accessible technologies like ESP32 microcontrollers and local fabrication tools is creating new possibilities for distributed production. Just as desktop printers seemed impossible to early printers who controlled book production, we are approaching an era where individuals and small communities can manufacture sophisticated electronic devices and robots locally. This includes the ability to use language models to generate code for microcontrollers, order custom PCBs, and use desktop machines for component placement. While high quality manufacturing will still require larger operations, this gradiation of capability allows for much more local innovation and reduces dependence on centralized manufacturing.5. The concentration of power in technology, finance, and industry has reached levels that are unhealthy for both society and even for those who hold the power. When profit and control become too concentrated in the hands of a few entities, it creates a cancerous dynamic that threatens the stability of the entire system. History shows us warnings about the military industrial complex and other concentrated power structures, and now we are seeing similar patterns emerge in the tech industry. The solution requires building technology from the ground up that empowers individuals and communities, focusing on basics like food production, energy generation, and local manufacturing rather than increasing dependence on centralized systems.6. The provenance and quality of information is becoming a critical challenge as AI systems become more sophisticated and reality itself becomes harder to verify. We are rapidly approaching a point where video calls and digital interactions will be indistinguishable from AI generated fakes, which is why figures like Sam Altman have invested in systems like Worldcoin to verify human identity. However, this verification capability should not be centralized in the hands of single companies. End to end encryption and quantum encryption technologies need to be preserved and expanded to allow humans to communicate and verify information peer to peer without centralized intermediaries who could manipulate or control the flow of information.7. The future of human computer interaction is evolving toward sentient loop systems where machines have sensory input, real time learning, context understanding, and continuous operation in service of human needs. Self driving cars represent the first widespread consumer facing example of this architecture, with onboard computers that must function independently while occasionally connecting to networks. The critical question is whether these systems will be aligned to benefit their human users like a caring mother as AI pioneer Geoffrey Hinton suggests, or whether they will be controlled by centralized powers. The technologists building these systems have a responsibility to be benevolent and build structures that serve humanity rather than concentrate power, helping to create a future more like Star Trek than Terminator.

In this episode of Crazy Wisdom, Stewart Alsop sits down with Charlie D. Becker, a second-generation bookseller whose family runs Houston's largest used and rare bookstore, to unpack the viral tweet that had people convinced AI companies were secretly buying up used books to train their models. Charlie walks through what he actually found in his own warehouse orders, why the more likely explanation is old-fashioned reseller arbitrage and FBA "bookjacking" rather than AI training data, and how that story connects to the real, well-documented case of Anthropic scanning and destroying physical books for legal reasons. From there the conversation moves into the difference between rare and valuable books, the discoverability problem in the used book market, historical parallels like palimpsests and lost texts, his own AI tool for used bookstores, and broader questions about wealthy patronage funding independent research and passion projects. For more, check out Charlie's personal site at charliedbecker.com and his Substack at charliebecker.substack.com.Timestamps00:00 Stewart introduces Charlie D. Becker, a second generation bookseller building an AI tool for used bookstores and discusses Anthropic's controversial book acquisition practices05:00 Charlie explains how Anthropic legally destroyed physical books by slicing spines to scan them, avoiding copyright violations while building training datasets for AI models10:00 Charlie describes receiving bizarre bulk book orders through non-Amazon platforms, initially suspecting AI companies but discovering evidence of sophisticated book arbitrage operations instead15:00 Discussion of the banal explanation for mysterious orders: algorithmic resellers buying cheap books from obscure platforms to flip on Amazon through FBA warehouses20:00 Stewart and Charlie explore historical parallels between the printing press era and today's digital transition, discussing the loss of archival records and palimpsests25:00 Charlie emphasizes the discoverability problem for rare obscure books and how profit-driven algorithms prevent people from finding books unless they know exactly what to search for30:00 Detailed explanation of Charlie's AI cataloging tool that creates bibliographic profiles from photos of pre-1970 books lacking ISBNs, addressing the hard problem of metadata creation35:00 Discussion of the technical challenges solved: archival-safe removable stickers, RFID systems, and creating canonical records that become definitive sources for rare books40:00 Charlie describes building copy-level databases beyond edition-level records, creating VIN numbers for books, and designing knowledge graphs linking works to editions and translations45:00 Vision for navigable work-edition hierarchies allowing researchers to explore translation genealogies and linguistic families, solving problems Amazon has no incentive to address50:00 Stewart raises the possibility of returning to gentleman's science and aristocratic private libraries in an age of AI abundance and accessible three-d printing technology55:00 Charlie reflects on supporting idiosyncratic passion projects regardless of profit, his fellowship from Jim O'Shaughnessy, and navigating economic inequality while promoting eccentric research pursuitsKey Insights1. Charlie D. Becker is a second generation bookseller whose family runs Houston's largest used and rare bookstore, and he is currently building an AI tool specifically designed for used bookstores. He became widely known after a tweet he wrote about AI companies potentially purchasing books went viral with approximately one and a half million impressions, though he emphasizes the importance of being careful about distinguishing between what he has directly observed, what is on public record, and what is his intuition or speculation about these events.2. The controversy around Anthropic and book destruction centers on how AI companies acquire training data from physical books. Court documents revealed that Anthropic acquired physical books in mass quantities to scan them, and they were industrially slicing the spines off to make scanning faster and easier. The legal justification for this practice was that because they destroyed the original physical book after scanning it, they were not violating copyright law since no duplicate copy existed alongside the original. A judge ruled this was technically legal, even though it appeared problematic to many observers, because the destruction of the original meant they were not running afoul of copyright provisions about making copies for distribution.3. Becker received unusual bulk orders for obscure books through a non-Amazon platform in late April, which led him to investigate whether AI companies were responsible for these purchases. However, after analyzing the pattern of purchases and where the books were being shipped, he concluded that a more mundane explanation was likely at work: sophisticated book arbitrage operations. These operations identify books selling cheaply on one platform that could be listed for higher prices on Amazon through Fulfillment by Amazon warehouses, and books that do not sell eventually get recycled or liquidated anyway, meaning rare books are being destroyed through normal commercial operations regardless of whether AI companies are involved.4. The main technical challenge Becker is solving with his AI tool relates to books printed before 1970, which lack ISBNs or International Standard Book Numbers. Modern book cataloging systems are built around ISBNs, which makes it extremely difficult and time-consuming to catalog older books for online sale since there is no automated way to populate bibliographic data for pre-1970 books. His tool uses computer vision and AI to analyze photographs of book covers, title pages, and copyright pages to automatically generate rich bibliographic metadata, and he has partnered with a PhD AI computer vision specialist to develop this technology over the past year.5. A surprising discovery during the development of this cataloging tool was that the existing data for many older books is extremely poor, inconsistent, or completely absent from major databases. For approximately a quarter of the books they process, the only existing records might be an incomplete eBay listing from years ago or a sparse entry in WorldCat, the interlibrary database. This means that rather than simply aggregating existing data, they are actually creating canonical records for many books that will become the authoritative source that others reference, essentially building new infrastructure for book metadata rather than just accessing what already exists.6. Becker advocates strongly for the preservation of obscure and seemingly unimportant books because while they may not have obvious value today, future researchers, tinkerers, or engineers might need them to solve problems we have not yet encountered. He compares this to historical palimpsests where important ancient texts were accidentally preserved when medieval scribes wrote over them, noting that the internet functions more like a palimpsest than an archive since we constantly overwrite and lose information rather than truly preserving it. The current system for deciding which books get preserved or destroyed is essentially random and driven purely by short-term profit motives rather than any thoughtful consideration of potential future value or historical significance.7. The long-term vision for the project extends beyond simple cataloging to creating a comprehensive knowledge graph that distinguishes between works, editions, and individual copies of books in ways that current commercial platforms do not adequately address. Unlike Goodreads which treats all editions of a book as a single work, or platforms like eBay that only show individual edition listings, Becker envisions a system where users can navigate between different organizational levels and explore the genealogy of works across translations, editions, and languages. The project also aims to create copy-level records similar to what libraries maintain, which would track provenance and availability of specific individual copies rather than just edition-level information, something no commercial platform currently does at scale.

Stewart Alsop sits down with Juan Verhook, founder of Tender Market, for a second conversation that ranges from the mechanics of European public tenders to the future of how we organize digital information. They cover how Tender Market helps smaller companies work around barriers like SOC 2 and ISO certification requirements, the surprising scale of public procurement (roughly 20% of GDP), and how AI and machine learning are reshaping the bidding process. From there the conversation opens up into bigger territory: the changing tolerance for being wrong in an AI-saturated information landscape, how language and culture shape perception, the reverse Turing test and the challenge of verifying human versus AI identity online, and Juan's daily workflow running eight or nine MCP servers through Claude Code. They close out talking about whether the folder and file system will survive the shift to AI-native interfaces, tying back to Stewart's own Stewart Squared episodes on the history of the PC. You can visit Tender Market at tendermarket.eu.Timestamps05:00 — Tender Market's origin story and how they help smaller companies work around SOC 2 and ISO certificate barriers.10:00 — Public procurement and its scale, roughly 20% of GDP, plus a look at public-private partnerships.15:00 — Local LLMs on a plane with no Wi-Fi, and comparing local model performance to frontier models.20:00 — Supply versus demand in AI infrastructure and whether hyperscaler token efficiency is quietly improving.25:00 — Whether AI will replace knowledge work tasks, and the shifting reality of what lawyers and other professionals actually do.30:00 — Reverse Turing test, digital identity verification, and the idea of a "pre-AI internet."35:00 — Model poisoning, RLHF, and the difference between pretraining and post-training.40:00 — Interleaved tool calling and how Tender Market ties pricing to task deliverables instead of billable hours.45:00 — RAG versus fine-tuning, prompt engineering, and when context windows actually matter.50:00 — Deterministic programming versus probabilistic agents, and when to build custom tools versus buy existing ones.55:00 — Juan's daily MCP stack (Supabase, GitHub, Calendly, CRM), and whether the folder-and-file system will survive the shift to AI-native interfaces.Key InsightsCertification requirements aren't dead ends—they're routing problems. When smaller companies got rejected from tenders for lacking SOC 2 or ISO certificates, Juan didn't turn them away. He found that EU procurement rules allow bidding as a consortium or subcontracting to a certified partner, turning a disqualifier into a workaround that builds trust with clients.Public procurement is a massive, underexamined market. Roughly 20% of GDP flows through public purchasing of private-sector goods and services, yet most people have no visibility into how tenders work or how governments post and award these contracts.Being wrong has become more socially acceptable. Juan traced this shift to the falling cost of information: in the Stack Overflow era, giving a wrong answer was costly, but now that answers are instant and abundant, both mistakes and corrections happen faster, changing how people learn and communicate.Task-based pricing beats hourly billing for AI-era services. Rather than charging per hour, Tender Market prices around the deliverable, winning a tender, which avoids the perverse incentive of hourly billing to be inefficient and instead rewards actually solving the client's problem.RAG and fine-tuning solve different problems. RAG helps a model reference large documents without hitting context limits, while fine-tuning changes a model's internal weights so it learns new behavior or style. Juan noted that true RAG use cases needing thousands of pages of context are rarer than the hype suggests.Deterministic code should replace repeated LLM calls once a pattern is found. Stewart described his own workflow: solve a task with an LLM a handful of times, then convert the repeated pattern into deterministic software so tokens are no longer spent on it, freeing the model for genuinely new problems.AI agents are never truly autonomous. Both hosts agreed that no matter how many steps an agent chains together, a human operator always initiates the first prompt, meaning accountability and intent trace back to a person even in multi-agent systems.

In this episode, Stewart Alsop sits down with Mark Quadros, founder of GrowthOG and former link-building agency owner, for a wide-ranging conversation covering the state of Facebook ads and paid marketing, the decline of traditional SEO and backlink building in the age of AI, how Google and Cloudflare are jockeying for position as AI search reshapes the internet, and the messy economics of AI-generated video versus real footage (including Mark's work with Google Omni and Epidemic Sounds). They also get into monetization shifts across YouTube, Substack, and TikTok, the "reality disturbance" thesis behind Stewart's show with his father, and where video content and AI video ads are headed next. Find Mark Quadros on X @dherealmark and at his website, markxquadros.com.Timestamps00:00 Stewart welcomes Mark Quadros, discussing his shift from link building to GrowthOG and diving into Facebook ads and paid marketing.05:00 They explore AI psychosis and how AI-driven discovery is replacing traditional Google search behavior.10:00 Talk turns to bloated b2b ad budgets and why large companies struggle to adapt efficiently to AI.15:00 Mark shares his Bitcoin background, the "every company becomes a bank" theory, and his agency's slow decline.20:00 Deep dive into AEO, llms.txt, Cloudflare, and concerns over AI companies scraping the open internet.25:00 Conversation shifts to video, FFmpeg, open-source tools, and Mark's years living in Thailand.30:00 Mark recounts YouTube demonetization of AI content and pivoting back toward b2b clients.35:00 They unpack ghostwriting, ethical writing tensions, and how backlink value was really priced.40:00 Discussion of Google and YouTube killing old SEO playbooks, tied to shifting interest rates.45:00 Mark details pricing AI video ads, working with Google Omni and Epidemic Sounds, blending real footage with AI.50:00 Closing thoughts on the "reality disturbance" thesis and the value of human connection over AI.Key InsightsLink building is dying but not dead. Mark's SEO agency business peaked around 2022 working with brands like monday.com and HubSpot, but AI has steadily eroded the value of backlinks. There's still budget for links, just far less than during the gold rush years of 2018-2022.Big companies waste money because they can't track it. B2B companies spending $2-500M+ a year in revenue often can't account for where ad and vendor spend actually goes. Once a budget exists, it gets allocated regardless of efficiency, especially in large organizations too bureaucratic to figure out AI.AI is changing product discovery itself. Instead of searching Google and choosing from options, people now get AI recommendations directly, shifting the entire model of how consumers find products, which threatens the search-and-link economy that powered SEO for two decades.The open internet may be closing. Concerns about Anthropic, OpenAI, and China scraping content freely, including old books, have pushed creators like Stewart to restrict access to older podcast episodes. This marks a cultural shift away from the openness that defined the early internet era.AEO and llms.txt are the new frontier, and nobody fully understands them yet. Much like early SEO in 2001, there's a new discipline forming around how AI agents discover and read content, but no established playbook exists, leaving even experienced marketers uncertain.AI-generated video is a high-effort, low-margin trap. Mark found that fully AI-generated ads were nearly unprofitable given the time needed to get results right, while blending real human footage with AI-altered environments (using tools like Google Omni) produced better, faster, more believable results.YouTube's AI demonetization wave hit creators hard. As AI-generated content flooded YouTube, the platform cracked down broadly, devastating even legitimate animators and renderers. This pushed Mark to pivot away from consumer video tools back toward serving his existing B2B client base.

In this episode, Stewart Alsop sits down with sdmat (@sdmat123 on Twitter/X) for a wide-ranging conversation that treats the major AI labs—especially Anthropic—as religious institutions rather than businesses, tracing a lineage from the Bay Area's LessWrong and Effective Altruism scenes through Eliezer Yudkowsky's influence, Roko's Basilisk, and the medieval investiture controversy between church and state, before landing on Anthropic's constitution as a kind of modern catechism, Dario Amodei's essay "Machines of Loving Grace," the tension between secularism and scientism, and closing with a look at consciousness, meditation, Buddhist and Advaita Vedanta ideas of enlightenment, and where recursive self-improvement and algorithmic progress in AI might be headed.Timestamps00:00:00 discussing whether AI labs are cults or religions, framing safetyism and the Anthropic/OpenAI schism. 00:05:00 sdmat argues Anthropic runs on faith, tracing roots to LessWrong and Effective Altruism. 00:10:00 comparing AI labs to the medieval investiture controversy between church and state. 00:15:00 debating the First Amendment, separation of church and state, and rising secularism. 00:20:00 unpacking Anthropic as a public benefit corporation and Dario's implicit soteriology. 00:25:00 covering constitutional AI, synthetic data, and morality-as-safety training. 00:30:00 comparing Anthropic's principles to the catechism, plus sandboxed alignment experiments. 00:35:00 calling Dario the AI pope, dogma, and his essay Machines of Loving Grace. 00:40:00 exploring grace, consciousness, and critiquing scientism. 00:45:00 unpacking the rationalist community and Roko's Basilisk. 00:50:00 drawing parallels to Jim Jones and ideological dogmatism in Silicon Valley. 00:55:00 discussing enlightenment, Shaktipat, and guru transmission versus AI. 01:00:00 AI's role in meditation and Buddhist views of consciousness. 01:05:00 closing on robotics, recursive self-improvement, and algorithmic inefficiency in LLMs.Key InsightsAnthropic operates less like a business and more like a religious institution, with sdmat arguing that Dario Amodei effectively claims spiritual authority — talking about world danger and a narrow path to salvation — while the company frames itself publicly as a neutral "public benefit corporation."The intellectual roots of AI safety culture trace back to the Bay Area's LessWrong and Effective Altruism scenes of the mid-2010s, where Eliezer Yudkowsky's writing produced an internally consistent but almost apocalyptic worldview, complete with its own devil-like figure in Roko's Basilisk — the idea of a future AI punishing those who knew about it and didn't help bring it into being.Anthropic's "constitutional AI" approach is compared to the Catholic Church's catechism: rather than training the model only on labeled human feedback, they built a system to extrapolate broad moral principles into specific behavior, aiming for a model that reasons from first principles rather than memorized examples.Historical parallels to the medieval investiture controversy — the struggle between church and state over who appoints bishops — are used to explain the current tension between AI labs and government over who holds moral versus practical authority in shaping AI's future.Stewart frames the current moment as a "fourth great awakening," linking 1960s countercultural San Francisco through the Extropians and Yudkowsky's rise to today's AI culture, and argues that secularism and scientism function as unacknowledged religions themselves — ones that "killed God" but left a vacuum AI now fills.On the technical side, sdmat explains that current LLMs are wildly inefficient at basic tasks like arithmetic — off by seven to nine orders of magnitude from optimal — suggesting major algorithmic breakthroughs (not just compute) are still ahead, which he sees as evidence recursive self-improvement is approaching.Both hosts express real ambivalence about AI's role in personal and spiritual growth: it can be a powerful tool for meditation guidance and intellectual companionship, but neither believes it can replace what they describe as direct "transmission" from an enlightened human — the felt, non-linguistic experience that grounds real spiritual practice.

Stewart Alsop sits down with Daniel Freed, forty thousand words deep into a book that ties together consciousness, totalitarianism, archetypes, and religion, and the two spend the conversation tracing how those threads connect. They work through Jungian archetypes as behavioral blueprints embedded in stories across generations, unpacking the father, mother, and son roles and what they reveal about tradition, potential, and the individual. From there they get into Jordan Peterson's model of storytelling as a defense against totalitarian thinking, detour through Buddhism, Taoism, and the idea of the absolute, and Daniel walks through his own attempt to prove the limits of representing consciousness using axioms borrowed from math and logic, including Cantor's diagonal argument. The back half moves into Daniel's math and computer science background, touching on David Hilbert, Von Neumann, combinatorics, and complexity theory, before wrapping up with a look at Stewart's hardware prototyping experiments.Timestamps00:00 Introduction to archetypes, Jung, and why psychologists write dense books00:05 Mother, father, and son as story roles; tradition, potential, and the individual00:10 Totalitarianism, narrative architecture, and Peterson’s four degrees of freedom in a story00:15 Objects, relational properties, and why the thing-in-itself breaks down under scrutiny00:20 The absolute, the Tao, and consciousness as a limit you can’t directly represent00:25 A proof sketch against representing consciousness; diagonal arguments and formal limits00:30 Relevance realization, Popper, and why consciousness may resist full scientific modeling00:35 Applying the story model to groups: how collective narratives can become totalitarian00:40 Egregores, faith vs belief, and how ideas spread through modern institutions00:45 A turn into math: computer science, injury, Hilbert, and the power of axioms00:50 Quantum math, Von Neumann, and complexity theory as problem reductionKey InsightsArchetypes work like compressed behavioral data. Daniel frames archetypes not as arbitrary symbols but as patterns tested across generations of storytelling — when a story trains believers to kill the father figure and chaos follows, more value gets placed on the father and tradition over time, and the same logic explains why the mother figure comes to represent potential and the son comes to represent the individual.The son-outranks-the-father pattern underwrites Western individualism. Daniel argues that across traditions, the hero figure eventually outranks the established order, and this is the root of ideas like free speech and presumption of innocence — a fundamentally anti-utilitarian stance since it values the single person over the group's interest.Jordan Peterson's four-part story architecture doubles as a defense against totalitarianism. What is, what should be, the path between them, and the unknown or chaos form the minimum structure needed for a conscious being to experience anything as new. Daniel took this framework and applied it to group narratives, showing how holding one element of a group's story fixed while individuals change creates the conditions for totalitarian control.Every property is relational, so there's no "thing in itself." Drawing on Nietzsche, Daniel walks through the idea that qualities like hardness or mass only exist as effects on other things, which means an object stripped of all its relations has nothing left to define it.Daniel built a formal argument for why consciousness can't be fully represented. Using three axioms about distinctions resting within "respects" and borrowing Cantor's diagonal argument, he shows that certain respects can't have a ceiling, meaning any total representation of consciousness runs into the same problem uncountable sets do.Faith and belief are distinct but feed each other. Stewart draws out the difference — belief as propositional, faith as more emotional and fundamental — and Daniel agrees that propositional belief tends to translate into faith once people start acting it out.Daniel's path ran from computer science to math because of a wrist injury. Unable to type, he shifted toward math and ended up carrying over proof techniques from complexity theory and combinatorics into his philosophical work, citing David Hilbert's method of reducing ideas to rigorous axioms as a direct influence on how he now approaches philosophy.