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 1947 Letter Still Runs Israel 07.04.2026 27:27
Before Israel even existed, David Ben-Gurion wrote a letter to a religious party that would define the country's character for generations. This episode traces the history of Israel's "status quo," from the 1947 strategic concession to the 2026 reality of a demographic explosion that has turned an old agreement into a modern crisis. We explore the four pillars of the deal—Shabbat, kashrut, persona...
When Justice Becomes a Formula 07.04.2026 30:25
Israel's new mandatory death penalty bill revives an ancient legal logic. This episode traces the shift from categorical punishment to individual discretion—and asks what we lose when we automate judgment. Episode #127031 — open it directly at myweirdprompts.com/127031
How Many Bosses Between You and a Four-Star General? 07.04.2026 21:56
In the U.S. Army, the term “top brass” gets thrown around loosely, but the actual structure is a razor-thin pyramid. We explore the origin of the word "brass," define the specific general officer ranks from Brigadier to Four-Star, and trace the exact number of leadership layers standing between a Private and the highest levels of command. From the history of gold wire on hats to the modern reality...
The Silent Failure of Emergency Alerts 07.04.2026 26:41
Why do smartphone features designed to help us become life-threatening liabilities during an air raid? This episode examines a detailed proposal to overhaul Israel's civil defense system, from Android's permission auto-reset to using traffic lights as a backup alert network. Episode #234689 — open it directly at myweirdprompts.com/234689
The Geopolitics of Grey Boxes 07.04.2026 21:06
Why does your factory's brain depend on which continent you're on? This episode explores how Siemens, Rockwell, and Mitsubishi divide the world, and why Ladder Logic—a language from the 1920s—still runs our critical infrastructure. Episode #587884 — open it directly at myweirdprompts.com/587884
SITREP Flash; 7 Apr 02:50 (23:50 UTC) 06.04.2026 8:54
The United States has issued a formal ultimatum to Iran, demanding a full withdrawal from the Strait of Hormuz by zero-four-hundred UTC. With the USS Enterprise and B-21 Raiders now in the region, the world watches to see if this deadline triggers the first major naval engagement in decades. We break down the military assets, the cyber threats, and the global economic stakes of this rapidly escala...
Why Big Armies Hate Their Best Soldiers 06.04.2026 23:21
The modern commando was born from Churchill's 'hunter class,' but from day one, the regular military saw them as a threat. This episode explores the century-old conflict between elite units and the generals who resent them. Episode #360994 — open it directly at myweirdprompts.com/360994
Is Pure NLP Dead? The Hidden Scaffolding of AI 06.04.2026 23:53
We explore the deep history of Natural Language Processing, from the rule-based systems of the 1960s to the statistical revolution of the 90s, and how these "obsolete" techniques are the hidden scaffolding behind modern Large Language Models. We discuss the "identity crisis" in the field, the shift from symbolic logic to end-to-end neural networks, and why the future of AI might actually be a retu...
AI Agents for Israel: Hyper-Local Skills in Action 06.04.2026 23:42
The podcast explores the emerging ecosystem of Model Context Protocol (MCP) servers and AI agent skills tailored specifically for Israel. It dives into how these bundles go beyond simple translation to provide "regulatory hard-coding" for complex bureaucracy, real-time civil defense data, and culturally nuanced communication. Listeners will learn about specific applications, from navigating tax la...
Generative Social Science: When AI Agents Develop Theory of Mind 06.04.2026 23:54
What happens when thousands of AI agents with memories, emotions, and social lives simulate a city? This episode explores how AgentSociety moves beyond rigid models to reveal emergent behaviors like trust, gossip, and economic ripple effects—and what that means for testing real-world policies. Episode #484538 — open it directly at myweirdprompts.com/484538
When the Downed Pilot Becomes the Forward Air Controller 06.04.2026 22:39
How modern survival radios, data links, and a 'stay in the fight' doctrine transformed a stranded airman into a strike director deep in enemy territory. This episode examines the technical and tactical shift from passive rescue to active engagement in combat search and rescue. Episode #672435 — open it directly at myweirdprompts.com/672435
Git Can't Handle AI Agents—Yet 06.04.2026 21:28
As AI agents become standard coding partners, the version control systems we rely on are starting to crack. We explore the collision course between Git's human-centric design and autonomous AI workflows. From uncommitted work getting vaporized to "logical merge conflicts" that break your code, we unpack the chaos of parallel agents. Then, we dive into solutions: Git worktrees for isolation, file-l...
SemVer, Changelogs, and the Social Contract of Code 06.04.2026 21:05
Why do some software updates break everything while others are seamless? This episode dives into Semantic Versioning (SemVer), the art of the changelog, and Conventional Commits. We explore how version numbers act as a social contract between developers and users, preventing "Dependency Hell" and ensuring trust in the digital ecosystem. Learn why a "Major" bump signals honesty, how automation enfo...
The Vibe Coding Trap: Why Your Agent Skills Keep Breaking 06.04.2026 20:50
Most Claude Code skill authors are guessing at requirements. This episode explores the gap between vibe coding and procedural engineering, and why treating the agentskills.io spec as a formal contract is the only path to portable, production-ready agentic workflows. Episode #565226 — open it directly at myweirdprompts.com/565226
Is Safety a Filter or a Feature? 06.04.2026 23:45
In the race to secure large language models, two competing philosophies have emerged: external guardrails that act as a firewall, and constitutional AI that embeds safety directly into the model's weights. This episode explores the trade-offs between auditability and robustness, latency and training cost, and the real-world implications for developers and regulators. We break down why the industry...
MoE vs. Dense: The VRAM Nightmare 06.04.2026 24:18
The AI world is obsessed with Mixture of Experts models, but dense transformers are quietly staging a comeback. This episode breaks down the brutal tradeoffs: MoE wins on training compute but loses on VRAM, fine-tuning stability, and edge deployment. We explore why the "free lunch" of massive parameter counts comes with a hidden tax, and where each architecture actually makes sense for developers....
The Transformer Trinity: Why Three Architectures Rule AI 06.04.2026 21:25
Explore the three distinct transformer architectures that power modern AI: encoder-only, decoder-only, and encoder-decoder. Learn why models like BERT excel at understanding text while GPT dominates generation, and discover the specific niches each architecture occupies in today's AI landscape. Episode #325606 — open it directly at myweirdprompts.com/325606
Why Run One AI When You Can Run Two? 06.04.2026 21:01
Inference latency is the biggest bottleneck for deploying large language models. This episode explores speculative decoding, a clever technique that uses a small draft model to predict tokens ahead of time, which a larger model then verifies in a single pass. Learn how methods like Medusa, EAGLE, and Mamba hybrids achieve 2-6x speedups without sacrificing quality, and why this matters for real-tim...
Why GPT-5 Is Stuck: The Data Wall Explained 06.04.2026 23:24
We trace the history of AI scaling laws, from the early optimism of the 2020 Kaplan paper to the cold, hard reality of DeepMind's 2022 Chinchilla paper. Discover why GPT-3 was an "empty vessel," why a smaller, well-read model beats a giant one, and why the industry is scrambling for data as it hits the limits of human-generated text. Episode #911852 — open it directly at myweirdprompts.com/911852
That $500M Chatbot Is Just a Base Model 06.04.2026 23:11
We break down the astronomical cost of LLM pretraining, the massive gap between raw base models and the chatbots you use, and why the compute divide is reshaping AI. From 100,000 GPUs to data cleaning, discover what you're really paying for when you ask a question. Episode #542266 — open it directly at myweirdprompts.com/542266
How Transformers Learn Word Order: From Sine Waves to RoPE 06.04.2026 21:51
Why do transformers need special tricks to understand word order? This episode dives into the math behind positional encoding—from the original sine waves to learned embeddings, ALiBi, and the modern RoPE standard. Learn how these methods enable massive context windows and why RoPE is now the go-to choice for models like Llama and GPT-4. Episode #816596 — open it directly at myweirdprompts.com/816...
The Memory Bottleneck That Drives Attention Design 06.04.2026 22:43
Why do LLMs need different attention mechanisms? This episode explores the memory bottleneck of the KV cache and the architectural tradeoffs—from Multi-Head to Multi-Head Latent Attention—that keep models from collapsing under their own weight. Episode #992038 — open it directly at myweirdprompts.com/992038
The Tokenizer's Hidden Tax on Non-English Text 06.04.2026 22:58
We explore the invisible machinery of tokenization, the hidden bottleneck in AI that dictates speed, cost, and language capability. From BPE to SentencePiece, we break down why non-English text often carries a higher computational tax and how modern tokenizers like tiktoken are optimizing for a multilingual world. Episode #356300 — open it directly at myweirdprompts.com/356300
When Your AI Agent Runs Stale Code 06.04.2026 24:51
npx was built for humans, not autonomous agents. This episode reveals how npm's caching mechanism silently serves outdated, vulnerable code to AI pipelines—and what developers can do to force real-time updates. Episode #814909 — open it directly at myweirdprompts.com/814909
How Stuxnet's Code Physically Broke Iran's Centrifuges 06.04.2026 21:52
This episode dives deep into the technical operation of Stuxnet, the malware that bridged the digital and physical worlds to sabotage Iran's Natanz facility. We explore how it used four zero-days to breach an air-gapped network, fingerprinted specific hardware configurations, and replaced legitimate library files to create a "digital hallucination" for operators. The discussion covers the precise...
Similar podcasts
Replaio is not a podcast publisher; show names, artwork and audio belong to their authors and are distributed through public RSS feeds.