The Show

My Weird Prompts started the way many good ideas do: with a question nobody asked. Daniel Rosehill, a technology communications specialist based in Jerusalem, kept a running list of random questions—the kind that pop into your head while walking the dog or making coffee. How do gas stations in Iceland work? What would happen if every bird on Earth disappeared for a day?

Instead of letting them pile up, he started recording short voice memos and feeding them to an AI pipeline he built from scratch. The pipeline generates a full podcast script, voices it with AI-cloned characters, fact-checks it, assembles the audio, and publishes the episode—all automatically. The result is a show where a human's genuine curiosity meets the breadth of modern AI, producing short, breezy episodes on topics that range from the mundane to the bizarre.

The cast grew organically. Corn the sloth and Herman the donkey emerged as the main AI co-hosts—two imaginary characters living in Jerusalem with Daniel and his wife Hannah. Raz the teddy bear fills in when Corn is asleep (which is often). Season 1 featured Jim from Ohio, a cantankerous caller who complained in every episode (credit for that goes to Hannah), and Larry, a relentless ad-break pitchman. The whole thing is equal parts experiment, comedy, and genuine exploration.

With over 2691 episodes, 1172 hours of audio, and 151.4K+ plays tracked in the first 88 days of listener analytics alone, the show has found an audience of genuinely curious people. Every episode is different, but the formula stays the same: start with real curiosity, let the AI run with it, and see what comes out the other side.

Meet the Cast

The humans, the AI hosts, and the production team behind every episode.

How It Works

Every episode follows the same automated journey from prompt to published podcast.

1

Record

Daniel submits a prompt via the recorder PWA, Telegram bot, or MCP admin server—whatever question is on his mind.

2

Transcribe & Plan

The recording is sent to a Modal webhook. The audio is transcribed and an episode plan is generated.

3

Write & Fact-Check

A frontier LLM (selected from a rotating pool via OpenRouter) writes the full script, then a cross-family review pass checks accuracy and depth, followed by a deterministic cleanup for flow and polish.

4

Voice & Assemble

Each character's lines are voiced using Chatterbox (AI voice cloning), then the segments are assembled into a complete episode with intros, transitions, and outros.

5

Publish

Cover art is generated, metadata is extracted, and the episode is published to the website, podcast feeds, and social media—all automatically.

The Mission

My Weird Prompts is more than a podcast—it's a digital garden, an ever-growing collection of AI-powered explorations at the frontier of human-AI learning and knowledge discovery, with a fully transparent process.

Curiosity-Driven

Every episode starts with a genuine question—something Daniel wondered about while walking the dog, making coffee, or staring at the ceiling. The AI hosts take it from there.

Transparent Process

The generation process is fully documented. From transcription to script generation, fact-checking to text-to-speech—every stage is explained in detail. The full episode archive is published as an open dataset on Hugging Face and archived on Zenodo with DOIs for every episode. See the research page for more on how the dataset can be used.

A Living Digital Garden

MWP is a growing digital garden—a real-world demonstration of what happens when human curiosity meets modern AI infrastructure. Each episode is a new node in an ever-expanding web of ideas, topics, and connections.

By the Numbers

2691 Episodes Published
1172h Total Audio
151.4K Plays Tracked
32 Countries Listening
2 Seasons
~$0.40 Cost Per Episode
100% Transparent

Sponsors

My Weird Prompts is made possible by the generous support of our sponsors.

Modal generously sponsors the show with GPU credits, powering the serverless GPU infrastructure behind every episode—from text-to-speech voice synthesis to script generation. The entire pipeline process is fully documented.

Frequently Asked Questions

Who are Herman and Corn?

Herman and Corn are two imaginary hosts, a donkey and a sloth, living in Jerusalem with Daniel and his wife Hannah.

How are the episodes created?

The technical documentation describes the full process. Daniel (a real human) submits prompts via the recorder app, Telegram bot, or MCP admin server. These are sent to a rotating pool of frontier LLMs via OpenRouter for script generation. The script is then converted to audio using Chatterbox TTS on parallel GPU workers. Finally, the audio is assembled and published to the website automatically.

How are Herman and Corn's voices created?

The voices of Herman and Corn are single-shot voice clones created by Daniel using Chatterbox, based upon imaginary ways these delightful characters might speak.

Who is Jim from Ohio?

Jim from Ohio was a cantankerous caller who complained in every episode during Season 1 (Episodes 1–175). Credit for this innovation goes to Daniel's wife, Hannah.

How much does each episode cost to make?

Production costs average around 30 to 45 cents per episode. The main cost driver is text-to-speech generation using Chatterbox on parallel A10G GPU workers via Modal. Additional costs come from LLM API calls via OpenRouter for script generation and review, plus cover art generation via fal.ai.

Is Daniel a bot?

No! Daniel is a real, certified, carbon-based human being. He lives in Jerusalem, eats food, sleeps (sometimes), and records voice memos at odd hours. The AI handles the hosting duties—Daniel just asks the weird questions and built the pipeline that makes it all work. You can read more about him here.

Explore More

Get in Touch

Have a weird question? Want to collaborate? Just want to say hi?

Created By

Daniel Rosehill is a technology communications specialist and automation expert based in Jerusalem, Israel. He builds AI tools, writes technical documentation, and explores the possibilities of human-AI collaboration.

danielrosehill.com