Deep Dives

Deep Dives

By Thomas KliemtSociety & CulturePhilosophy
Download on the App Store

Deep Dives episodes

  • Episode 8 - Sandbox Monthly AI call - Minh presenting Agentic Workflows for Creative Production and Automation

    In this episode, the group dives deep into the practical and creative applications of AI agents. Guest Minh breaks down his automated workflows, revealing how he uses custom-built deck builders and ties AI into his calendar and meeting transcripts—such as Fireflies—to automatically draft business proposals and manage CRM tasks.

    The conversation then shifts to the frontier of AI-driven video and creative production. Minh demos AI generation tools like Luma, illustrating how agents can generate consistent characters and narratives—like a photorealistic superhero therapist for Earth—from incredibly vague prompts. We also explore a fascinating use-case in professional feature film production, where creators are using AI agents to generate thousands of video variations overnight, deliberately overproducing content so humans can simply curate the best shots the next day.

    This prompts a broader discussion on the economics and "evolutionary wastefulness" of AI generation. Thomas questions the hidden hardware and energy costs of compute-heavy models—like Google's Genie—and debates whether the true cost of dedicated AI computing will eventually outpace the cost of hiring human freelancers.

    Finally, the episode wraps up with practical tips for users looking to deploy local models. The group discusses the hardware requirements for running local LLMs, specifically noting why upgrading to an M-series Mac Mini is preferable to using older Intel machines. Minh also addresses the security and reliability concerns of using open-source tools like LM Studio and Ollama, advising that local LLMs should be restricted to simple tasks like file sorting to avoid risks.Key Topics Covered:

      • Building an AI-automated workflow for presentations and proposal drafting.
      • The rise of agentic video generation and mass-producing content for feature films.
      • Debating the hidden financial and energy costs behind AI compute.
      • Hardware tips and security practices for running local LLMs securely.
    24 min
  • Episode 7 - The Sandbox Monthly - Daniel presenting Local LLMs and Custom Voice Fine-Tuning

    Episode Overview: In this session of the Sandbox Monthly AI call, the group explores the practicalities of running and fine-tuning local and cloud-based Large Language Models (LLMs). Spurred by concerns about over-reliance on major US tech companies and the potential nationalization of AI, the conversation centers on how developers can create independent, "US-proof" AI workflows. Guest speaker Daniel walks through his personal project of building a custom voice agent, offering a masterclass in accessible, low-cost AI training.


    Key Takeaways & Topics Discussed:

      • Infrastructure & Tools: Daniel highly recommends Modal, a cloud platform that makes it easy and affordable to train custom models without melting your personal computer's GPU. He notes that while you can run local servers on devices like Mac Minis, a standard MacBook Pro with 24GB of memory will struggle with models larger than 20 to 30 billion parameters.
      • Curing the "AI Tone": To avoid the generic, robotic voice typical of AI-generated content, Daniel trained a custom model for his blog and LinkedIn posts. Using supervised machine learning, he created training pairs by matching AI-generated bullet points with high-quality articles written by his favorite tech bloggers.
      • The "Garbage In, Garbage Out" Rule: When training an LLM on your voice, the quality of your output depends entirely on your input. Daniel found that you need at least 500 high-quality data samples for a first pass; otherwise, smaller models run out of context and begin to hallucinate, lose facts, or ramble.
      • The MVP Approach to AI Training: Instead of starting with expensive 70-billion parameter models, Daniel suggests treating early AI training like a science experiment. He runs quick, cheap tests on smaller 8-billion parameter Llama models—which take about 30 minutes and cost around $2 to train—before graduating to larger, more expensive models once the data pipeline is perfected.
      • High-Stakes AI Limitations: When asked about automating legal documents, Daniel advises extreme caution. Due to the high risk of hallucinations, he suggests leaving high-stakes professional tasks to specialized, industry-standard tools (like Harvey for law) rather than experimenting with custom open-source models.
      • A Clever Onboarding Hack: Daniel shares how he uses NotebookLM to turn dense git histories and codebase documentation into personalized "podcasts," allowing him to quickly onboard whenever he gets a new client.

    Would you like me to generate a NotebookLM Audio Overview (an actual AI-hosted podcast episode) of this transcript so you can listen to a discussion of these exact topics?

    22 min
  • Episode 6: From Proof-of-Work to Proof-of-Care

    In this episode, we explore Thomas Kliemt's journey from witnessing Bitcoin's massive energy consumption to envisioning a revolutionary financial tool: a land-backed, stable-value cryptocurrency. Drawing from his experience with the Kulturland Cooperative, Kliemt discusses how the growing demand for stable digital money could be repurposed to secure farmland as a modern commons, permanently removing it from speculative markets.

    We dive into his 2018 whitepaper, the regulatory hurdles of launching a mission-aligned crypto asset, and how the original concept has evolved into a vision for a pan-European cooperative. Join us to hear how the future of finance could shift from resource extraction to "proof-of-care," providing patient, low-interest capital to empower local farmers and ensure long-term ecological stewardship.


    The audio is fully AI generated with no human copy-editing of the content, based on the following article: https://thomaskliemt.substack.com/p/from-proof-of-work-to-proof-of-care

    21 min
  • The Problem with IQ Tests

    This episode explores The IQ Paradox, moving beyond the mechanics of testing to examine how we define human value. While IQ tests correlate with academic success and job performance, they fail to capture the full spectrum of human potential, such as wisdom, empathy, and creativity.

    Drawing on scientific insights and personal narrative, this episode covers:

    • The Flynn Effect: Why average IQ scores have risen by 30 points over the last century, proving that intelligence is malleable and heavily influenced by culture, education, and nutrition.

    • A Dark History: The link between IQ testing and the forced sterilization of over 60,000 people in the US, a practice that later influenced Nazi Germany.

    • The Trap of Arrogance: A personal look at how relying on a high IQ score can justify arrogance and prevent genuine human connection.

    • The Valid Use Case: How, despite its flaws, standardized testing can be a tool for social justice by identifying talented students from disadvantaged backgrounds who might otherwise be overlooked.

    • Connection Over Ranking: Why viewing intelligence as a hierarchy leads to loneliness, and how shifting our focus to humility and empathy opens the door to richer relationships.

    Inspired by the research of Veritasium, this discussion challenges the narrative that cognitive speed equals human worth. Join us to discover why the most important question isn't "How smart am I?" but rather, "How can I connect?".


    The audio is fully AI generated with no human copy-editing of the content, based on the following article: https://substack.com/@thomaskliemt/note/c-203393346?r=1ntyb6&utm_source=notes-share-action&utm_medium=web

    16 min
  • Getting an Au Pair - Choosing Presence Over Exhaustion

    This audio is fully machine generated by the NotebookLM AI based on the essay "Getting an Au Pair - Choosing Presence Over Exhaustion" published on Substack: https://open.substack.com/pub/thomaskliemt/p/getting-an-au-pair?utm_campaign=post-expanded-share&utm_medium=web

    14 min
  • Managing Buffers: A Little-Known Path to a Clear Head and Relaxed Control

    Welcome to Deep Dives, where we explore a powerful, yet barely discussed, path to a clear head and relaxed control: managing buffers. Drawing on Donella Meadows’ systems thinking, we examine how a buffer acts as the quiet capacity that absorbs shocks, transforming a potential cascade of friction into a minor inconvenience.

    This approach connects deeply with the goal of creating a reliable, trusted system that helps you stay in a "mind-like-water" state as much as possible. We dive into the practical systems that help eliminate friction and reduce the small annoyances that would otherwise be time-consuming, stressful, and emotionally draining.

    On Deep Dives, you will learn how to build systems around reality, such as accepting frequent item loss and equipping valuable possessions—like keys, wallets, bags, and coats—with tracking devices. We explore the strategy of keeping spares of frequently used items, such as favorite pens or drinking bottles, and maintaining a backup stock of consumables, including toilet paper, shampoo, and essential food items. By adopting this approach, running out of supplies or losing an item stops being an emergency and instead becomes a small maintenance task.

    The ultimate aim is not to become a hyper-optimized person, but to reclaim the attention that is currently being taxed by preventable friction. We show you how managing buffers removes the tiny resentments of everyday life that can quietly tax your emotional bandwidth, allowing you to function productively with a clear head and a positive sense of relaxed control.

    Find the Substack article here: https://open.substack.com/pub/thomaskliemt/p/managing-buffers-a-little-known-path?utm_campaign=post-expanded-share&utm_medium=web

    13 min
  • The Art of Stress-free Traveling

    Tired of travel days feeling like "sudden emergencies"? Welcome to "Deep Dives," the podcast dedicated to achieving preparedness without thinking by exploring systematic approaches to organization and efficiency.

    In this podcast, we take a deep dive into highly effective methods, inspired by systems like Getting Things Done, that remove a huge amount of friction from your life,. We explore how to maintain travel bags that are "almost fully packed at all times", ensuring you eliminate the mental energy spent on last-minute decisions.

    Each episode focuses on breaking down the components that allow travel to feel "soft" instead of sharp:

    • The Gear and Hacks: We detail what lives permanently in a dedicated travel kit, from comprehensive lists of electronics and charging solutions to medical equipment and comfort items,,,. Learn about small hacks, such as the soft cotton snack bag that often doesn't count as additional hand luggage,, or utilizing a Floh hand luggage scooter to turn the stressful "last mile" into a short, playful ride.

    • The Checklists: We analyze the power of simple, well-tested checklists that handle the rest,. Discover the critical tasks to complete before you leave home (like turning off the robot vacuum or setting up call forwarding), optional items for longer trips,, and the crucial steps needed to reset your bag after a trip to ensure future journeys start from a "calm, prepared baseline".

    By implementing these strategies, you can minimize practical problems and focus on the people you travel with and the destination you are going to. Tune in to refine your own system and make travel a lot calmer and more "being-present-in-the-moment".

    12 min
  • Die Rückkehr der Allmende: Das wegweisende Urteil des OLG Hamm zur Kulturland-Genossenschaft

    In der ersten Folge von Deep Dives wird über das wegweisende Urteil des Oberlandesgerichts Hamm zur Kulturland-Genossenschaft – und was es für die rechtliche Rückkehr der Allmende, für Bodenreform und junge, ökologische Bäuer*innen in Deutschland bedeutet berichtwt. Diese Episode wurde vollständig mit NotebookLM erstellt, auf Basis meines Essays und des Originalbeschlusses – alle Stimmen sind KI-generiert.

    Links:

    • Gerichtsbeschluss (OLG Hamm): https://nrwe.justiz.nrw.de/olgs/hamm/j2025/10_W_127_24_Beschluss_20250929.html

    • Mein Essay zu Kulturland: https://thomaskliemt.substack.com/p/die-kulturland-genossenschaft-die

    16 min

About Deep Dives

From the publisher's feed

Deep Dives is a personal podcast by me, Thomas Kliemt – biodynamic farmer, father, commons thinker and finance/digital-transformation guy who somehow ended up between soil, spreadsheets, spirituality and politics in rural Wendland, Germany. In each episode I feed information about a topic I am currently engaged with into NotebookLM and generate a deep dive. It’s an experiment in utilizing AI to make information more accessible.