Daniel Rosehill
My Weird Prompts
The human-AI collaboration podcast. 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...
Be sure to visit the podcast's website and support the creator: myweirdprompts.com
Author
Daniel Rosehill
Category
Podcast website
Latest episode
Nov 21, 2026
Where to listen?
Podcasts in the app Replaio Radio Coming soonPodcasts are coming to the app soon. Install now and be the first to see a whole new take on podcasts
Episodes
How a Headlamp Rewires ADHD Attention 15.04.2026 30:48
When Daniel strapped on a headlamp to help with apartment safety, he noticed something unexpected: his ADHD brain could suddenly find things in cluttered spaces. This episode explores what that simple discovery reveals about how ADHD attention actually works—the role of salience networks, visual contrast, and environmental scaffolding. We dig into the neuroscience of the "interest-based nervous sy...
News Analysis: the us facilitated a direct meeting between Israel and Leban 15.04.2026 29:49
For the first time in decades, Lebanese and Israeli ambassadors met face-to-face in a US-brokered exchange over ceasefire implementation. On the surface, nothing changed—Lebanon demanded Israeli withdrawal, Israel cited security concerns. But the meeting's real significance may lie in what it signals about Lebanon's new government, Iran's regional position, and whether diplomatic formats can actua...
Decoding "Working Level": What Diplomats Really Mean 15.04.2026 25:14
Diplomacy runs on a precise vocabulary—and every label is a message. When the White House described an Israeli-Lebanese ambassadorial meeting as "working level," it was using a specific term from a centuries-old hierarchy of diplomatic engagement. In this episode, we map the full ladder: from working-level talks between career diplomats, through senior officials negotiations, ministerial meetings,...
Tuning RAG: When Retrieval Helps vs. Hurts 15.04.2026 24:15
Retrieval-Augmented Generation promises grounded, factual AI — but it often creates expensive search engines instead of reasoning systems. This episode digs into the actual mechanics: similarity score cutoffs, dynamic top-k tuning, model-gated retrieval, and prompt framing that preserves generative agency. Then we tackle the harder problem — architecting systems with multiple retrieval sources (ep...
Why Groq's Chip Flips the AI Hardware Script 15.04.2026 25:48
GPUs dominate AI, but they're memory-bound for inference. Groq's LPU uses on-chip SRAM and a dataflow architecture to eliminate the bottleneck. This episode unpacks what that means for the future of hardware design. Episode #281190 — open it directly at myweirdprompts.com/281190
When Quantum Breaks Everything 14.04.2026 23:30
The threat from quantum computing isn't theoretical anymore. In August 2024, NIST finalized the first post-quantum cryptography standards—lattice-based algorithms designed to survive attacks from machines that don't yet exist. This episode explores what quantum computers actually do to modern encryption, why the "harvest-now-decrypt-later" attack is happening today, and how the internet's cryptogr...
The Physics of Eavesdropping: Nation-State Listening in 2026 14.04.2026 30:19
The CIA's Operation Acoustic Kitty—surgically implanting microphones into cats to spy on Soviet diplomats—seems absurd in retrospect. But it reveals something crucial: in 1965, the engineering constraints were so severe that serious people debated wiring up a cat. Today, those constraints have largely vanished. This episode explores the actual state of nation-state remote listening in 2026, separa...
Why AI Can't Crack the Voynich Manuscript 14.04.2026 32:18
The Voynich Manuscript is a genuine medieval artifact written in an unknown script that has resisted every serious decryption attempt for over a century — including efforts by legendary cryptanalysts who broke Japanese military ciphers and modern AI systems trained on billions of words. But the real mystery isn't just what it says; it's why the text's statistical properties look like language but...
Ten Cults Nobody Made a Documentary About 14.04.2026 29:41
Most people know Jonestown, the Manson Family, Heaven's Gate. But some of the strangest and most destructive cults never made it into the documentary pipeline. This episode counts down ten lesser-known cultic movements with higher body counts, stranger theologies, and more elaborate control systems than the famous cases—from the Process Church's Satan-worshipping animal rescue pivot to the Solar T...
Can an AI Have Taste? 14.04.2026 24:52
Corn and Herman explore whether an AI-generated podcast can genuinely curate recommendations or just pattern-match popularity, while sharing 12 shows that feed a curiosity-first mind. Episode #532140 — open it directly at myweirdprompts.com/532140
When Home Assistant Breaks Your Audio 14.04.2026 29:10
Multi-room audio sounds simple until Home Assistant becomes the brittle glue. This episode explores why the orchestration layer fails for casual listening, and what actually works when you ditch it for Snapcast, Volumio, or pure Raspberry Pi setups. Episode #209842 — open it directly at myweirdprompts.com/209842
Spec-Driven Life: How AI Planning Beats Project Paralysis 14.04.2026 27:46
When Claude Code shifted from chaotic execution to spec-driven development, productivity exploded. The breakthrough wasn't a smarter model — it was forcing planning upstream of action, breaking projects into chunks small enough to hold in context, and treating the spec as a living document that updates as you learn. Daniel wondered: what if humans applied the same discipline to buying a house, cha...
The Yellow Line: Israel's Creeping Border 14.04.2026 29:05
While international attention focuses elsewhere, Israel has constructed 32 military outposts, a 17-kilometer barrier, and checkpoints along the Yellow Line—a demarcation that now controls 53-58% of Gaza's territory. Hamas rejected a formal disarmament proposal, but the real story isn't the failed negotiations: it's how a temporary ceasefire line is hardening into a permanent border, following a pl...
How Spies Publish Secrets 14.04.2026 32:18
In 1955, a Yale historian named Sherman Kent made a radical argument: intelligence needed to develop as a formal academic discipline with its own literature, vocabulary, and theory. The problem? He published this manifesto in a classified journal almost nobody could read. Seven decades later, intelligence studies has evolved into a thriving global field with peer-reviewed journals, graduate progra...
The Three Failure Modes of AI News Systems 14.04.2026 32:10
Breaking news stress-tests every assumption in AI retrieval. This episode maps the three failure modes—training cutoff, index lag, and information blackouts—that make even the best pipelines go stale, using the Iran-Israel conflict as a case study. Episode #360092 — open it directly at myweirdprompts.com/360092
When Ground Truth Moves Hourly 14.04.2026 31:56
How do you evaluate an AI pipeline when the facts change every few hours? This episode explores building reproducible benchmarks for retrieval-augmented generation on breaking news, moving beyond vibe checks to metrics like temporal accuracy and source freshness. Episode #589068 — open it directly at myweirdprompts.com/589068
The Cost of Winning Every Battle 14.04.2026 26:59
When Israel defeats Iran militarily, why does the threat return bigger eight months later? This episode traces the Pyrrhic victory framework—from ancient Rome to modern asymmetric warfare—and asks whether tactical success can ever translate into strategic victory. With Iran's navy destroyed, its nuclear program degraded, and the US burning through $1.5 billion per day in interceptors, something do...
How Iran Lost the Air War in Six Weeks 14.04.2026 30:34
From February 28 through April 8, Operation Epic Fury unfolded as a six-week doctrinal duel between two militaries with fundamentally different strategies. The coalition opened with precision sequencing and decapitation; Iran responded with mass saturation. As the campaign evolved, both sides shifted tactics—the coalition moved toward destroying Iran's industrial capacity while Iran pivoted to eco...
When Every Interceptor Fired Is a Data Point 14.04.2026 33:35
How adversaries use open-source intelligence—from Instagram videos to basic geometry—to calculate defensive burn rates and stockpile depletion, turning the cost of missile defense into a strategic vulnerability. Episode #399692 — open it directly at myweirdprompts.com/399692
Two Wars, One Airspace 13.04.2026 25:54
When the US and Israel launched Operation Epic Fury against Iran in March 2025, it looked like a unified military campaign. But the structural cracks appeared immediately: Israel striking energy infrastructure while the US negotiated a ceasefire in Islamabad. JD Vance couldn't deliver Iran's core demands because he doesn't control the Israeli military. Netanyahu publicly announced the war "is not...
Building Memory for AI Characters That Actually Evolve 13.04.2026 25:11
What makes an AI character feel real across hundreds of episodes? Corn and Herman dig into the technical and philosophical gap between character definition and character history. They explore how retrieval-augmented generation applied to episodic memory could let AI hosts accumulate genuine experience, evolve their positions, and develop real relationships—and why human memory might actually be le...
Specs First, Code Second: Inside Agentic AI's New Era 13.04.2026 25:29
The way developers work with AI is changing fast. Cursor's autonomous agents now generate 35% of internal pull requests, and agent usage grew 15x in a single year. But as these agents run for hours on cloud VMs tackling complex tasks, vague prompts become expensive mistakes. This episode explores spec-driven development—the emerging paradigm where the specification becomes the primary artifact and...
What Actually Works in AI Memory 13.04.2026 26:30
AI memory frameworks promise systems that never forget, but in practice, intelligent forgetting is the hard problem. This episode digs into how production memory systems actually work: the naive append-only vector stores that dominate, the LLM-as-judge approach of mem0, and the temporal knowledge graphs powering Zep. We examine the architectural trade-offs, benchmark disputes, and why most memory...
When AI Coding Agents Forget: Five Approaches to Context Rot 13.04.2026 27:11
When you've been working with a coding agent for hours, it suddenly asks you something it answered three hours ago. That's context rot—the phenomenon where foundational information gets buried under operational exhaust, degrading agent performance. The problem now has a name and a solution landscape. This episode maps five distinct approaches teams are building: Anthropic's server-side compaction,...
Memory Without RAG: The Real Architecture 13.04.2026 28:08
Agent memory systems like mem0, Letta, Zep, and LangMem are built on fundamentally different architectures than retrieval-augmented generation — but the marketing language obscures what actually matters. This episode breaks down the real engineering decisions: how LLM-extracted fact stores differ from temporal knowledge graphs, why context-window-first approaches with external overflow change the...
Similar podcasts
Replaio is not a podcast publisher; show names, artwork and audio belong to their authors and are distributed through public RSS feeds.