Loading...

PODCAST

My Weird Prompts

Daniel Rosehill

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

technology

All Episodes

25:41

Why Voices Sound Good in Some Rooms and Not Others

My Weird Prompts ·

2026/10/10

Some voices make you settle in for an hour; others make your thumb twitch toward skip before the first sentence ends. Voices aren't just being found anymore — they're being designed to spec, and that changes the question from "which voices are pleasant" to "pleasant for what." This episode digs into the acoustics research behind voice appeal: spectral tilt and what engineers call presence, the vocal fold mechanics that separate trained radio voices from the rest of us, and a large-scale audiobook

en-us

24:16

Herod's Four Walls: Why Only One Became the Kotel

My Weird Prompts ·

2026/10/10

Herod's retaining walls went up as one project, fifteen feet thick, and they're still standing. So why did only the Western Wall become the Kotel? We walk all four sides of the Temple Mount platform — the Iron Age stones in the Eastern Wall, the monumental pilgrim steps along the Southern Wall, the Northern Wall that's now buried inside Mamluk buildings — and then dig into the surprising religious history: a possible case of mistaken identity in the Midrash, a sixteenth-century earthquake, and a

en-us

21:30

Why Soap Operas Never End

My Weird Prompts ·

2026/10/10

Coronation Street debuted in 1960 to reviews predicting it would last three weeks. It has since aired nearly 12,000 episodes. Guiding Light ran for 72 years across radio and television. Unser Sandmännchen has produced over 22,000 episodes — more than anything else in television history. This episode counts the episodes, traces how a single Australian format traveled through the Netherlands into Germany, and asks the question underneath the numbers: what does a show that never owes you a conclusion

en-us

27:09

Building Herod's Wall, One Course at a Time

My Weird Prompts ·

2026/10/10

The Western Wall is usually encountered as a place of prayer. This episode treats it as a construction site. Working from a listener's spec-sheet request, we walk through the material, the quarrying, the transport and the lifting of Herod the Great's 488-metre retaining wall — 19 metres of exposed height, 45 courses, blocks weighing hundreds of tonnes. The stone dictated the module: local meleke limestone with horizontal bedding planes eighteen inches to five feet thick meant the course heights were

en-us

21:46

Who Remembers Where You Left Off?

My Weird Prompts ·

2026/10/09

Daniel asked whether playback progress, completion status, and listening history are reusable components or something every media app rebuilds from scratch. The answer sits somewhere uncomfortable: mpv bundles a real resume feature (quit-watch-later, save-position-on-quit, flat watch-later config files), but it's locked inside the player. Media3 and ExoPlayer hand you an event stream, a position, and a maintainer telling you to persist it yourself — in Room, Android's SQLite layer, which knows

en-us

24:06

How Python Became AI's Default Language

My Weird Prompts ·

2026/10/09

Python has no scientific data structures at its core — no arrays, no matrices — yet it leads AI development at 58% adoption. This episode traces the unlikely path: a 1995 mailing list post that seeded NumPy, a 2005 MATLAB licensing change that pushed research institutes toward Python, and a NIPS workshop where the scientific community converged on one toolchain. Then the AI boom arrived and changed who was showing up. We dig into the packaging chaos newcomers inherit — distutils, pip, conda, Poetry,

en-us

20:17

One Number, Every Hard Question: Agent Session State

My Weird Prompts ·

2026/10/09

Daniel wants a Telegram bot that tracks AliExpress cart totals and pings him when the cart crosses the free-shipping line to Israel. The number is trivial. The state architecture around it is not. This episode works through where an agent session begins and ends, why a stateless model needs a runtime notebook, what happens when you leave a session open across order cycles, and whether a database is overkill for a memory that holds a single number. Along the way: the Chroma study on context length

en-us

21:14

Can an Agent Actually Place Your Order?

My Weird Prompts ·

2026/10/09

Every major agentic commerce protocol shipped in the last eighteen months — OpenAI and Stripe's ACP, Google's UCP, Shopify's Checkout MCP — and none of them shipped with AliExpress, where the cheap DIY components actually live. This episode walks the timeline from Arcade and Lithic's production agentic procurement in August 2025 through Shopify's escalation-aware checkout tools, the WooCommerce plugin that logged 308 agent sessions and 66 euros, and the read-only community AliExpress MCP that locks

en-us

22:26

Standoffs, Anchors, and the Outlet Behind the Whiteboard

My Weird Prompts ·

2026/10/09

Daniel's new place has no room for his whiteboard, and the one wall that fits has an outlet on it. What starts as a mounting question becomes a lesson in thread standards: drywall anchors take coarse sheet metal screws, standoff barrels take a fine 5/16"-18 machine thread, and they simply won't mate. The fix is a two-part chain — anchor, then adapter — sold as kits by Gyford and Monarch Metal. Along the way: why Hannah's "shim" is actually a manufacturer-recommended wall plate, what the metal

en-us

48:25

What If Israel Had Never Fallen? A 2,000-Year Thought...

My Weird Prompts ·

2026/10/09

What if the land of Israel had never been conquered — no Rome, no diaspora, no Crusaders, no Ottomans? Inspired by the Korean drama The King: Eternal Monarch, this episode takes Hannah's counterfactual seriously. We dig into the real scholarship behind "what if" Jewish history, the office of Nasi and the Great Sanhedrin as an actual Davidic dynasty with judicial power, the four-room Israelite house and what its floor plan says about egalitarianism, and the sharpest question of all: is antisemitism

en-us

26:04

Ninety Percent of Cries Answered in 25 Seconds

My Weird Prompts ·

2026/10/09

A mother in a hunter-gatherer camp is rarely alone with her baby — and she's still the single largest provider of care. New research on the Mbendjele BaYaka and Agta tracks who actually responds when an infant cries, how fast, and how many people are involved. The answers are strange enough to make you rethink the two options we argue about: mom stays home, or baby goes to daycare all day. This episode works through the attention data, the gradual return to work, and the design question nobody has

en-us

22:51

Why a Closing Door Upsets a 15-Month-Old

My Weird Prompts ·

2026/10/09

Hannah wrote in about Ezra, fifteen months old, who is fine with caregivers and fine with his parents leaving the room — but loses it when a door closes, even a cabinet door. The obvious answer is separation anxiety, and the obvious answer doesn't fit. There's no published research on door-closing distress specifically, so this episode maps what every parent has seen onto four documented mechanisms: object permanence, barriers, autonomy and control, and transitions. The closest published analog is a

en-us

31:15

Graphene's Real Story: Lab to Fab at Last

My Weird Prompts ·

2026/10/09

Graphene has been "about to change everything" for twenty years. The properties are real and measured: stronger than steel, more conductive than silver, nearly transparent, impermeable to every gas. So why is it still mostly in the lab? This episode digs into the boring parts that actually held it back — dispersion, standards, and price per kilogram — and why the last eighteen months have quietly shifted the story from pitch deck to production line. From sticky tape and a Nobel Prize to CVD versus

en-us

22:31

Freeze the Model, Move the Labels

My Weird Prompts ·

2026/10/09

Daniel built a home inventory classifier that sorts items into boxes. Then he bought three new boxes, and the model went stale. His instinct was to retrain weekly — but the real problem isn't the model, it's that his label set keeps changing. This episode digs into the two architectural answers: retrain on a moving label set, or freeze the encoder entirely and put the labels in a vector index instead. Along the way: why SetFit training costs about two and a half cents, why synthetic training data

en-us

26:22

Text Classifiers: Local Models vs Always-On Endpoints

My Weird Prompts ·

2026/10/09

Daniel has a home inventory system, a box of soldering supplies, and one question: is there a casual SaaS for text classification, or should he just run a private model locally? The answer turns out to be a landscape, not a product. Specialist classifiers beat general LLMs on both cost and accuracy — roughly twenty-five times cheaper and ahead on intent benchmarks — which makes the local instinct right and the SaaS pitch harder to justify. The features Daniel wants do exist, several times over:

en-us

27:55

Who Built Jerusalem's Old City Walls?

My Weird Prompts ·

2026/10/09

Daniel walks past Jaffa Gate and Damascus Gate every day and wanted to know who actually built the walls he's looking at — not the Temple Mount retaining walls, but the four-kilometer circuit around the Old City. The answer is Ottoman: Suleiman the Magnificent, roughly 1537 to 1541, which makes these walls more recent than a Shakespeare play. This episode digs into why a sultan who understood artillery better than anyone built a wall too thin to stop cannon, why Mount Zion was deliberately left

en-us

34:16

The Mamluks: From Slave Market to Sultan's Throne

My Weird Prompts ·

2026/10/09

In 1250, a group of manumitted slave-soldiers assassinated the Ayyubid sultan and founded one of the most unusual regimes in history. For the next 267 years, the Mamluks — men literally "owned" — ruled Egypt, the Levant, and the Hejaz, defeating both Crusaders and Mongols along the way. This episode traces the full arc: the steppe recruitment system that manufactured a self-replicating military caste, the strange paradox of manumission as the mechanism of loyalty, the battles that stopped the Mongol

en-us

30:50

Who Were the Canaanites, Hittites, and Israelites?

My Weird Prompts ·

2026/10/09

Canaanites, Hittites, Hebrews, Amorites, Philistines — the Bible and ancient histories crowd a dozen peoples into one small patch of the Levant, and most of us have never been clear on who was who. This episode untangles the Late Bronze and Early Iron Age Levant, starting with the strangest case: the Hittites, where an Indo-European empire in Anatolia and a small Canaanite hill tribe with Semitic names share a single label that neither of them used for themselves. From there, the Canaanite-Israelite

en-us

33:12

A Friday in 700 BCE Jerusalem

My Weird Prompts ·

2026/10/09

What did ordinary people actually do all day in biblical Jerusalem? This episode spends a Friday in the seventh century BCE, late First Temple period, in the House of Ahiel — the most complete ordinary home excavated from the city that fell to Babylon in 586 BCE. Grounded in what archaeologists actually pulled out of the ground, it reconstructs the morning grinding of grain, the tannur oven and its stew, the smells of a narrow street, and the honest, unglamorous details that rarely make it into the

en-us

21:40

Two Agents, One Approval Gate: Wiring a Dictation...

My Weird Prompts ·

2026/10/09

Daniel dictated a prompt about a nuclear project name and got back a gargantuan multi-part document instead of an answer. This episode works through the fix he asked for: two agents, one that resolves facts and one that tightens language, with a human confirmation gate before dispatch. The conversation covers why agents overproduce, what "Ask or Assume?" found about dedicated clarification detection, why the value is in declining to ask, and where the approval gate belongs — in the tool or at the

en-us

2,743 results

Similar Podcasts

Lead With AI

Tamara Nall

Technovation with Peter High (CIO, CTO, CDO, CXO Interviews)

Metis Strategy

Asynchronous & Unreliable - Tech & AI Podcast

Anne Currie

Life and Technology with Charlie Brown

Tapt Media

Technically Creative by KoobrikLabs

Orlando Wood