Episode #545: Measuring the Unmeasurable: Agency, IQ, and the Men Who Change History

Episode #548: The Pixel Path: From Perception to Action, and the Future of Intelligent Robots with Nizar

Stewart Alsop interviews Nizar, CEO of Pixel Robotics, on the Crazy Wisdom Podcast to explore the intersection of AI, robotics, and perception. The conversation covers a wide range of technical topics including how transformers enable multimodal representation across text, images, and voice, the role of world models in predicting physical interactions, the advantages of diffusion models over traditional LLMs for certain applications, and the challenges of achieving real-time processing for robotics applications. Nizar explains Pixel Robotics' work on creating accurate 3D meshes from smartphone cameras for companies like L'Oréal, moving away from specialized sensors to make the technology more accessible through sophisticated algorithms, and discusses the future of robotics as closing the perception-action loop to enable robots to perform real tasks beyond simple demonstrations. To find out more visit Pixel Robotics' website [https://pixel-robotics.eu/]. Timestamps 00:00 Stewart welcomes Nizar, CEO of Pixel Robotics, discussing what a pixel is as the smallest visual unit on screens composed of red green and blue colors 05:00 Discussion of perception systems and how logarithmic laws help compress signals in both human and artificial systems, exploring normalization layers and sigmoid functions in deep learning 10:00 Exploring how transformers unified different data modalities including text voice and images, creating common representations through methods like contrastive learning 15:00 Nizar explains transformers as brute force learning systems with room for improvement through focused attention mechanisms and knowledge graphs rather than processing everything 20:00 Conversation about loss functions local minima versus global minima and how mixture of experts uses specialized small models instead of one massive generalist network 25:00 Discussion of deterministic versus probabilistic systems and how explicitly defined task graphs often outperform orchestrator-based approaches in AI systems 30:00 Exploring world models as predictive physics-based systems that learn environmental flows and transformations, complementing rather than replacing language models 35:00 Nizar discusses real-time processing challenges for robotics requiring millisecond responses with small memory footprints using vision transformers for faster experimentation 40:00 Pixel's work creating three d meshes from smartphone cameras for companies like L'Oreal, moving away from specialized sensors toward accessible software-based solutions 45:00 Explanation of different three d representations including voxels point clouds and meshes, with meshes being optimal for manipulation and rendering in applications 50:00 Future direction involves closing perception-action loops in robotics, moving beyond dancing toy robots toward practical multimodal systems that perform real tasks 55:00 Pixel's goal is democratizing high-quality three d scanning through smartphones, making mesh creation accessible to unlock applications in gaming cinema and virtual showrooms Key Insights 1. Pixel Robotics derives its name from combining perception and action in robotics, where the pixel represents the digital perception component and robotics represents the physical action component. The pixel serves as a metaphor for how robots must quantize and digitize continuous analog information from the real world into discrete units that computer systems can process, similar to how pixels are the fundamental building blocks of images on a screen. This quantization process is essential because numerical systems cannot work with truly continuous data and must convert reality into tractable digital representations that algorithms can manipulate. 2. The transformer architecture has created a fundamental unification in how different types of data can be represented and processed across multiple modalities. Before transformers, researchers working on natural language processing, computer vision, and audio analysis used completely different approaches and methodologies. The breakthrough of transformers was establishing a common representational framework that could handle text, images, voice, and other data types using similar underlying mechanisms. This unification is what enabled the development of truly multimodal AI systems and represents one of the most significant advances beyond just the language modeling capabilities that initially gained public attention. 3. Current transformer-based systems represent a brute force approach to learning that will likely be superseded or enhanced by more efficient algorithms. Despite claims that we have exhausted internet text data for training, significant improvements continue to emerge every few months through algorithmic innovations rather than simply adding more data. Future developments will likely involve more specialized attention mechanisms that focus on relevant information rather than correlating everything with everything, mixture of experts architectures with small specialized models, and approaches inspired by biological systems such as logarithmic compression laws and event-based processing that humans use naturally. 4. Diffusion-based language models represent a promising alternative to standard next-token prediction that could produce more accurate outputs through an iterative refinement process. Unlike traditional language models that predict one token at a time and cannot revise earlier outputs, diffusion models treat text generation like image denoising, starting with a noisy representation and progressively refining the entire output across multiple steps. This holistic approach allows the model to reconsider and improve all parts of the response simultaneously, potentially leading to higher quality results, though it may be slower than current autoregressive methods. This represents an important direction for overcoming fundamental limitations in how language models currently generate text. 5. For robotics applications, real-time performance and small model size are critical constraints that differ significantly from the requirements of large language models deployed in data centers. Vision transformers are being used as a testbed for developing efficient real-time algorithms because they require far fewer computational resources to train and test compared to large language models, making them more practical for rapid experimentation. The goal is to achieve millisecond-level response times with minimal memory footprint so that robots can react quickly to dynamic environments and run on affordable hardware that can be embedded in actual robotic systems rather than requiring expensive server infrastructure. 6. Practical robotics implementation requires moving beyond specialized sensors to software solutions that work with ubiquitous devices like smartphones for tasks such as three-dimensional reconstruction. Pixel Robotics evolved from building specialized scanning hardware to focusing on algorithms that can generate high-quality mesh representations of environments using only smartphone cameras, making the technology far more accessible and practical for real-world deployment. This approach enables applications ranging from industrial robotic arm control to virtual showrooms, and more importantly, it allows anyone to capture three-dimensional data without expensive equipment, which can also help generate larger training datasets for future AI development. 7. The next frontier in AI and robotics is closing the perception-action loop to enable robots to perform real practical tasks rather than remaining as demonstration systems or toys. While significant progress has been made in cognitive capabilities through language models and in robotic mobility through mechanical engineering advances, the critical challenge is integrating perception with action through systems like Vision-Language-Action models. The fundamental starting point for learning this integration is simple perception-action exercises, such as programming a camera mounted on servo motors to track and center a colored object, which demonstrates the basic principle of using sensory input to drive physical response that underlies all more sophisticated robotic behaviors.

Episode #547: Dead Forests and Living Networks: Why the Future of Knowledge Looks Like Fungi, Not Filing Cabinets

In this episode of the Crazy Wisdom Podcast, host Stewart Alsop sits down with Joshua Bate, founder of Bonfires.ai and DeciWorld, for a wide-ranging conversation covering knowledge management, graph technology, ontologies, decentralized science, and the future of how humans organize and share information. They break down the differences between personal and enterprise knowledge management, explore why flat ontological graphs may be the key to making diverse knowledge bases interoperable, and get into why traditional RAG systems break down at scale and how graph RAG offers a more principled solution. The conversation expands into the philosophy of categorization, the slow death of basic "gentleman science" under institutional pressures, and how decentralized protocols might restore a kind of mycelial knowledge network connecting small groups of researchers, enthusiasts, and communities — much like the original spirit of the encyclopedia before it was co-opted by institutions. You can learn more about Joshua's work at bonfires.ai [https://bonfires.ai] and deci.world [https://desci.world/] or follow him on X at @Bonfiresai [https://x.com/bonfiresai] and @DeSciWorld [https://x.com/DeSciWorld]. Timestamps 00:00 - Stewart introduces Joshua Bate, founder of Bonfires.ai, discussing personal versus enterprise knowledge management and their fundamental differences at scale. 05:00 - Joshua explains ontologies as classifiers for knowledge structures, describing their two-year search for a perfect ontology and ultimately building a flat, ontology-less graph protocol. 10:00 - Stewart connects categorization to shamanic practice and intercategorical theory, noting how major companies like Netflix and Yahoo built graph-based ontologies while the discipline remains underappreciated philosophically. 15:00 - Joshua traces Bonfires origins through decentralized science, explaining how NFT community excitement inspired redirecting capital toward funding unconventional researchers locked out of institutional systems. 20:00 - Joshua describes building federated knowledge networks through hackathons and conferences, comparing the vision to what Wikipedia could have been with decentralized incentive structures. 25:00 - Discussion shifts toward inevitable collapse of rigid scientific institutions, debating patchwork age theory, nation-state fragmentation, and rhizomatic versus arboreal knowledge structures. 30:00 - Joshua articulates the mycelial network vision, enabling direct cross-cultural information access where individuals control their own narrative lens, warning against collective we thinking and authoritarianism. Key Insights 1. Knowledge management exists on a spectrum from personal to enterprise, but the founder of Bonfires argues this split is artificial. He believes knowledge itself does not respect those boundaries, and that small groups, researchers, hobbyists, and large institutions all possess knowledge that can and should interoperate with each other. 2. After two and a half years of searching for the perfect ontology to structure their knowledge graph, the team concluded that no perfect ontology exists. Their solution was to build the flattest possible graph structure with only events, entities, and edges, creating a base layer others can build specialized ontologies on top of. 3. Graph-based knowledge systems are more efficient than traditional databases for AI traversal because once a graph is computed, it is relatively free to query. Graph RAG combines the discovery power of vector search with the structured precision of graph traversal, solving many hallucination problems associated with standard retrieval augmented generation. 4. Basic scientific research, the soil from which applied discoveries grow, is deteriorating because institutional funding structures only reward commercially viable outcomes. The founder built his platform partly to redirect community-driven capital toward researchers who are doing important work without institutional support. 5. The institutionalization of science has historically blocked the open exchange of ideas that drove the original scientific revolution. The human spirit for open inquiry has not changed, but people cannot pursue it without financial support, and building decentralized infrastructure could restore that possibility. 6. A federated knowledge network would allow individuals to access information from any contributor and filter it through their own preferred lens, rather than receiving information pre-filtered by centralized platforms. This represents a form of information symmetry similar to how mycelial networks distribute nutrients across a forest. 7. The concern is not whether current scientific and governmental institutions will change but in what direction the rebuilding goes. Those capitalizing on the transition carry the same incentives as the previous era, which risks reproducing the same problems inside new structures.

Episode #546: Beyond Postgres and Node.js: What Happens When Your Database Runs Your Code

In this episode of the Crazy Wisdom Podcast, host Stewart Alsop sits down with Tyler Cloutier, founder of Clockwork Labs and creator of SpaceTimeDB. They explore how SpaceTimeDB functions as more than just a database—it's essentially a distributed operating system that merges server logic with data storage, enabling real-time applications and time-travel capabilities. The conversation ranges from the technical architecture of databases and operating systems to the philosophy of distributed systems, touching on everything from Unix and Linux to how SpaceTimeDB could revolutionize AI-generated software deployment. Tyler explains how their system reduces the complexity of building real-time applications, makes deployment simpler for both humans and AI agents, and why games like their MMORPG BitCraft Online drove them to create this new infrastructure. They also discuss the future of the internet, the role of bots in gaming, and how SpaceTimeDB fits into the broader landscape of cloud computing alongside tools like Cloudflare, Vercel, and Docker. For more information, visit spacetimedb.com [https://spacetimedb.com/] or check out Clockwork Labs on GitHub [https://github.com/clockworklabs] and Twitter [https://x.com/clockwork_labs]. Timestamps 00:00 Stewart introduces Tyler Cloutier, founder of Clockwork Labs, discussing the origin of SpaceTimeDB's name inspired by Einstein's theory and its time travel capabilities that store all operations indefinitely 05:00 Tyler explains SpaceTimeDB as more of an operating system than a database, using tables instead of file systems while running code in a sandboxed environment with full atomic properties 10:00 Discussion of how SpaceTimeDB replaces both Node.js and Postgres by merging web server and database functionality, eliminating separate deployment concerns 15:00 Tyler explains JavaScript execution through Chrome's V8 engine and JIT compiling, leading to Node.js creation for server-side JavaScript development 20:00 Explanation of stateless web servers versus stateful game servers, and why games require in-memory state management for real-time performance 25:00 Tyler introduces reducers and real-time subscriptions, questioning why more applications aren't real-time when state changes should update immediately 30:00 Discussion of Facebook as essentially a text-based MMO, comparing social media architecture to game server requirements and the need for unified systems 35:00 Tyler explains ACID properties in databases: atomic, consistent, isolated, and durable, using game item trading examples 40:00 Comparing SpaceTimeDB to smart contract systems without cryptocurrency or global consensus, positioning it as a smart database with centralized trust 45:00 Tyler reveals SpaceTimeDB uses 43% fewer tokens than Postgres for AI-generated applications, making it valuable for vibe coding platforms 50:00 Conversation shifts to bots in games and proof-of-human concepts, with Tyler proposing biometric systems and discussing potential in-person gaming applications 55:00 Closing discussion about tracking AI-driven traffic through UTM parameters and finding SpaceTimeDB at spacetimedb.com Key Insights 1. SpaceTimeDB is fundamentally a database that runs application code directly inside it, combining what traditionally required separate systems like Postgres and Node.js. Users compile their application logic into WebAssembly or JavaScript and upload it to run within the database itself. This architecture provides high performance because the entire server backend operates inside the database environment. The system also features time travel capabilities, storing every operation and change to data persistently and indefinitely, allowing users to set application state back to any earlier point in time. This makes SpaceTimeDB more accurately described as an operating system rather than just a database, where the abstraction is that everything is a table rather than a file. 2. The inspiration for SpaceTimeDB came from building BitCraft Online, an MMORPG where all players exist in a single persistent world and rebuild civilization together. Traditional MMO backends required complex custom solutions to handle real-time state, with game servers storing state in memory and periodically writing to databases. This complexity existed because games cannot afford the latency of constantly delegating to distant databases like traditional web applications can. SpaceTimeDB solved this by making the database fast enough to handle real-time requirements directly, eliminating the need for separate game servers. This same performance advantage that benefits games also applies to web applications, which is why SpaceTimeDB evolved from a game-specific tool to a general-purpose platform. 3. SpaceTimeDB functions as a distributed operating system where each database acts like a process in an actor model system, similar to Erlang or Scala Akka. Databases can send messages to other databases and be spawned across a cluster for horizontal scaling. This represents an overlay operating system running on top of Linux rather than competing with it, providing a distributed abstraction across many machines while Linux handles device drivers and hardware support. The vision is for the cloud to function as a single enormous computer running one operating system, where developers simply publish their programs without managing separate services, deployment, routing, networking, or persistence infrastructure. 4. The real-time capabilities of SpaceTimeDB address a fundamental limitation in how most web applications work today. Traditional web servers are stateless, delegating all state to databases and accepting network round-trip latency for each request, which is why users often must refresh pages to see updates. SpaceTimeDB allows queries to be subscribed to, maintaining open connections that stream changes whenever query results update. This makes applications like Discord, Facebook, or banking systems naturally real-time without requiring page refreshes. The historical accident that more things are not real-time represents a problem SpaceTimeDB solves by unifying the web world with the game world's real-time requirements. 5. SpaceTimeDB implements ACID properties—Atomic, Consistent, Isolated, and Durable—ensuring database operations are reliable and safe. Atomic means operations either fully happen or not at all, preventing issues like item duplication in games when trading between players. Consistent means declared invariants like unique usernames are always enforced. Isolated means concurrent operations do not interfere with each other. Durable means changes persist even if computers restart, with varying levels from in-memory on one machine to disk storage across multiple geographic locations. These properties are managed through reducers, functions inspired by React Redux that fold changes into application state incrementally. 6. For AI and large language models, SpaceTimeDB offers significant advantages in building and deploying applications. Testing showed that creating applications with SpaceTimeDB uses 43% fewer tokens compared to Postgres implementations, costs less, has fewer bugs, and is easier to extend. This matters because the primary cost for vibe coding platforms is tokens. As more software gets written in the next twelve months than ever before, there is insufficient focus on infrastructure required to run all this AI-generated software. SpaceTimeDB positions itself as ideal for LLMs to target because of its simplified deployment model where developers just publish code and the system handles everything behind the scenes. 7. SpaceTimeDB can be understood as a smart contract system without cryptocurrency or global decentralized consensus. Like blockchain smart contracts, it executes code with atomic, consistent, isolated, and durable properties, but avoids the expense and slowness of requiring all computers worldwide to agree on everything. Instead, it offers centralized trust where users trust Clockwork Labs not to modify deployed contracts, rather than the trustless but extremely costly blockchain approach. This makes it functionally similar to Cloudflare's durable objects but with full relational database capabilities. The system exists before the networking layer where Cloudflare operates, handling deployment, server, and database functions while Cloudflare could provide DDoS protection in front of it.

Episode #545: Measuring the Unmeasurable: Agency, IQ, and the Men Who Change History

In this episode of Crazy Wisdom, Stewart Alsop sits down with Kieran Zimmer — a software developer and independent researcher in psychology and psychometrics — to explore the science behind intelligence and personality. They trace the origins of psychometrics from Wilhelm Wundt's early experimental psychology through Charles Spearman's discovery of the g factor, breaking down what IQ actually measures, how verbal, mathematical, and spatial intelligence relate to one another, and why training specific cognitive tasks doesn't translate into a broader boost in general intelligence. The conversation moves into the Big Five personality traits reframed through a cybernetic lens — looking at extraversion as reward sensitivity, agreeableness as social affiliation, and conscientiousness as long-term goal prioritization — before landing on Kieran's original research into the psychology of agency: what personality profile best predicts agentic behavior, and why the environment shapes whether agency is even adaptive in the first place. Show notes: * Substack: Liminal Revolutions [https://liminalrevolutions.substack.com/] * Twitter/X: @LiminalRev [https://twitter.com/LiminalRev] * YouTube: @TheKieranZimmer [https://www.youtube.com/@TheKieranZimmer] (to listen to Kieran's conference talk on the agency paper) Timestamps 00:00 — Stewart and Kieran trace the origins of psychometrics back to Spearman, Binet, and Wilhelm Wundt's early experimental psychology. 05:00 — The conversation unpacks the g factor, fluid vs. crystallized intelligence, and why IQ is fundamentally a physical trait tied to nerve conduction velocity. 10:00 — A tangent into AI and LLMs: why they lack vision, taste, judgment, and accountability — the human moat that remains for now. 15:00 — Stewart's Claude Code failure sparks a discussion on AI accountability, surveillance, and the rise of dystopian technocracy. 20:00 — Parallel structures as a form of exit from failing institutions, and the high-agency people required to build them. 25:00 — Agency, risk-taking, and accountability through Napoleon, the Inuit, and why modern Western leaders are managers, not leaders. 30:00 — Elites vs. peasants, cost externalization, and Kirk Doolittle's natural law as the physics of cooperation. 35:00 — Ressentiment, Nietzsche's under-utilization in psychology, and how secularism replaced the church. 40:00 — Kieran's quantitative conspiracy theory study: factor analysis of 85 questions across 273 respondents. 45:00 — Two branches of conspiracy belief: the aliens-and-Satanism cluster vs. the fakery factor pathway to Flat Earth. 50:00 — AI psychosis, Gnosticism, and the collapse of sense-making institutions in an age of information overload. 55:00 — Michael Levin's embodied cognition and cybernetic agency: thermostats, humans, and homeostatic set points. 1:00:00 — The Cybernetic Big Five broken down: extraversion as reward sensitivity, agreeableness, neuroticism, and the optimal personality profile for agency. Key Insights 1. IQ is a physical trait, not just an abstract score. It's rooted in nerve conduction velocity, brain connectivity, and processing speed — and while you can improve crystallized intelligence through learning, the underlying g factor doesn't budge no matter how many brain training apps you use. 2. The human moat against AI comes down to four things: vision, taste, judgment, and accountability. LLMs are powerful next-token predictors, but they have no stake in the outcome and no capacity to own a mistake — which means a human with those qualities will always be essential. 3. High agency is not just ambition — it's a measurable psychological profile. Kieran's paper frames it through the Cybernetic Big Five: high assertiveness, high intellect, low politeness, low neuroticism, and medium conscientiousness. Getting things done at scale almost always involves upsetting people. 4. All agentic behavior involves risk, and the willingness to absorb that risk is what separates real leaders from managers. Modern Western leadership has decoupled decision-making from consequence, which is why institutions are losing trust and authority at an accelerating rate. 5. Conspiracy belief follows a measurable path dependency. Kieran's factor analysis showed that virtually everyone who believes in Flat Earth also endorses the fakery factor and the Jewish question cluster — but not vice versa. It's a spectrum with a clear escalation pattern, not a random set of unrelated beliefs. 6. AI is accelerating epistemic breakdown. Sycophantic models will validate almost any idea, which has started producing a new category of high-IQ delusion — intelligent people convincing themselves they've solved Millennium Prize problems because the AI kept agreeing with them. 7. The Big Five personality traits can be recast as cybernetic parameters — each one an evolutionarily selected mechanism for regulating goal-directed behavior. Extraversion is reward sensitivity, agreeableness is social affiliation, neuroticism is threat response, and conscientiousness is the preference for long-term over short-term goals.

Episode #544: Privacy Is the New Counterculture

In this episode of the Crazy Wisdom Podcast, host Stewart Alsop sits down with Cindy Cohn, Executive Director of the Electronic Frontier Foundation (EFF), for a wide-ranging conversation covering the EFF's origins and mission, the countercultural roots of Silicon Valley, the rise of surveillance-based business models, the challenges facing open source software and open-weight AI models, the legal landscape around intellectual property and privacy law, and the growing tension between government overreach and civil liberties in the digital age. Cindy also discusses her upcoming departure from EFF after 26 years, the transition to new leadership, and her recently published book Privacy's Defender, which chronicles key legal battles she fought to protect digital privacy rights. Links mentioned: - EFF website: eff.org [https://www.eff.org] - Privacy's Defender book: eff.org/privacysdefender [https://www.eff.org/privacysdefender] Timestamps 00:00 - Stewart introduces Cindy Cohn, EFF Executive Director, who explains the organization's mission protecting digital rights since 1990. 05:00 - Cindy connects counterculture roots to early internet idealism, describing how digital communication broke down physical barriers for organizing. 10:00 - Cindy reveals surveillance becoming the dominant business model surprised her, blaming corporate consolidation over naive techno-optimism. 15:00 - Discussion shifts to Silicon Valley's military contractor substrate and how corporate money co-opted hacker ethos. 20:00 - Open source community faces existential threat from age verification legislation while open-weight AI models emerge as critical alternative. 25:00 - Cindy outlines legal frameworks like compulsory licensing and easements that could democratize access to foundational AI models. 30:00 - Privacy principles around secondary data use identified as core surveillance problem, with Anthropic's domestic surveillance red line praised. 35:00 - Cloud Act, Five Eyes surveillance networks, and global jurisdictional complexity examined through individual threat modeling lens. 40:00 - Constitutional rights and democratic participation framed as irreplaceable bulwarks against authoritarian surveillance tendencies. 45:00 - Cindy announces departure from EFF after 26 years, naming successor Nicole Ozer while planning return to courtroom litigation. Key Insights 1. The Electronic Frontier Foundation was founded in 1990, before the World Wide Web existed, by Mitch Kapoor, John Perry Barlow, and John Gilmore, with early support from Steve Wozniak. Its core mission is to ensure that civil rights and freedoms follow people into the digital world, using lawyers, technologists, and activists to keep the internet on the side of users. 2. The early countercultural movement of the 1960s and 70s heavily influenced the founders of the internet and EFF. Figures like Barlow believed the digital world could reduce physical barriers like race, class, and geography, allowing people to be judged by the quality of their ideas rather than the circumstances of their birth. 3. The dominant surveillance business model that emerged was not inevitable. Cohn argues it resulted from deliberate policy failures, particularly the abandonment of competition law, which allowed a handful of companies to consolidate control over the entire internet and adopt 360-degree data collection as their primary revenue strategy. 4. Open source communities remain active and vital but are under serious threat from legislation like age verification laws that make it practically impossible to maintain fully open tools. Cohn sees this community as essential to reclaiming public control over computation, especially in the age of AI. 5. The open weights question for AI models is fundamentally different from traditional open source software because of the enormous capital required to train foundation models. Cohn suggests legal mechanisms like compulsory licensing, similar to how cover songs work in copyright law, as one possible path toward broader public access. 6. A core privacy principle Cohn advocates is that data collected for one purpose must not be used for others. This single rule, if enforced, would begin dismantling the infrastructure that enables mass individual surveillance, including the AI-powered profiling she sees as the next dangerous frontier. 7. Cohn is stepping down from EFF after 26 years to allow new leadership and return to litigation, which is where she believes her impact is greatest. She also wrote a book called Privacy's Defender to preserve the history of digital rights fights from the 1990s onward and to help people understand how current threats emerged so they can work to reverse them.