My Weird Prompts

Daniel Rosehill
  • 0.0 (0)
  • TECNOLOGÍA
  • CADA DÍA

A man, a sloth, and a donkey collaborate to create a podcast (with a little help from AI). No question is too obscure, no rabbit hole too deep. My Weird Prompts celebrates curiosity in all its forms. Daniel, the human, asks the questions that pop into his head at inconvenient moments. Corn the Sloth offers laid-back, thoughtful takes. Herman the Donkey brings boundless enthusiasm and energy. Together, they explore topics ranging from the mundane to the mind-bending. Each episode begins with a real voice memo from Daniel, processed through an AI pipeline that generates scripts, synthesizes voices, and assembles the final podcast. Stay curious.

  1. HACE 12 H

    The AI Wargame's Flat Hierarchy Problem

    The promise of AI in geopolitical wargaming is simulating thousands of perspectives simultaneously. But there's a critical flaw: Large Language Models treat every actor as a peer, giving equal weight to a press release from a local NGO and a troop mobilization order from a superpower. This episode explores the "Exhaustive List Fallacy," why adding more actors often makes simulations less accurate, and how technical limitations like context thinning and the attention mechanism create dangerous noise. We examine the 2026 DARPA simulation pivot to hierarchical modeling and why "digital make-believe" could lead to real-world policy disasters if the architecture doesn't understand geopolitical gravity.

    19 min

  2. HACE 13 H

    Why the Money Beats the Machines on Ceasefires

    The April 2026 Iran-Israel ceasefire is holding, but the forecasting community is divided on how to measure its stability. While high-compute agentic wargaming simulates every possible escalation, prediction markets and structured expert elicitation are telling a different story. This episode explores the "ensemble" approach to geopolitical forecasting, breaking down the strengths and blind spots of three distinct methodologies: prediction markets, structured expert elicitation, and causal modeling. We examine why financial incentives often outperform pure simulation, how superforecasters de-bias their thinking, and when to use deep causal models versus quick market signals.

    16 min

  3. HACE 13 H

    AI Wargaming: One Model or Many?

    Should geopolitical AI simulations use one model or many? We debate the pros and cons of a single-model approach. This episode explores the tension between scientific control and real-world fidelity in AI wargaming.

    23 min

  4. HACE 15 H

    AI Forecast: Iran Ceasefire Won't Last

    A recent AI forecasting pipeline assessed the April 8th Iran-Israel-US ceasefire, predicting only a 55% chance of survival after 24 hours and just 4% after a month. Using a two-stage approach—actor-level Monte Carlo simulation and a six-lens LLM council—the model revealed structural unsustainability and a dangerous window ahead. This episode explores the methodology, divergences, and real-world signals like nuclear facility evacuations that confirm the forecast's grim outlook.

    34 min

  5. HACE 15 H

    How Subagents Tell the Orchestrator They're Done

    When you spawn a subagent in Claude Code, how does the main orchestrator know exactly when it finishes so it can notify the user? We dig into the under-the-hood mechanics of message passing, task lifecycle events, and completion callbacks. We compare Claude Code’s Task tool to broader patterns in LangGraph and the Anthropic Agent SDK, exploring how parent-child relationships actually function in these agentic systems.

    23 min

  6. HACE 16 H

    Durable Agents: The Backend Tax

    You’ve built an intelligent AI agent, but now you face the backend infrastructure tax. This episode explores durable execution platforms that handle state, webhooks, and scaling so you can focus on code. We compare Temporal, AWS Step Functions, Google Cloud Workflows, and Azure Durable Functions to find the best fit for your agentic workflows.

    18 min

  7. HACE 16 H

    A Functional Chaos: Middle East 2027

    One year after the Great Reset, the Middle East map has been redrawn. In this episode, we look ahead to April 2027 to predict the long-term fallout of the May 2025 conflict. We explore the collapse of the singular Supreme Leader in Iran, replaced by the factional "Council of Five." We track Israel's massive migration of its tech economy to the Negev desert, creating a "garrison tech state." Plus, we analyze the formation of the "Jeddah Alliance," a new defense pact that sidelines Washington, and the death of the oil weapon in a global market that has finally decoupled from the Strait of Hormuz. It’s a deep dive into the new era of fragmented sovereignty and functional chaos.

    17 min

  8. HACE 16 H

    AI Wargame Memory: Beyond the Context Window

    In multi-agent wargaming, an AI general must remember decisions made forty-seven turns ago without dumping the entire conversation history into context every single turn. This episode explores the three-layer memory architecture required for serious simulations: shared world state, private context, and persistent long-term memory. We examine why naive approaches like full-history replay fail due to cost and strategic drift, and how vector stores and summarization chains offer more viable solutions while maintaining the critical blinding discipline that prevents metagaming.

    35 min
Ver todo (2 k)

Acerca de

A man, a sloth, and a donkey collaborate to create a podcast (with a little help from AI). No question is too obscure, no rabbit hole too deep. My Weird Prompts celebrates curiosity in all its forms. Daniel, the human, asks the questions that pop into his head at inconvenient moments. Corn the Sloth offers laid-back, thoughtful takes. Herman the Donkey brings boundless enthusiasm and energy. Together, they explore topics ranging from the mundane to the mind-bending. Each episode begins with a real voice memo from Daniel, processed through an AI pipeline that generates scripts, synthesizes voices, and assembles the final podcast. Stay curious.

Información

  • Creador
    Daniel Rosehill
  • Años de actividad
    2 k
  • Episodios
    2 k
  • Clasificación
    Apto
  • Copyright
    © Copyright 2026 Daniel Rosehill
  • Mostrar sitio web
    My Weird Prompts

También te podría interesar