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

Technology

Podcast website

myweirdprompts.com

Latest episode

Nov 19, 2026

Where to listen?

Podcasts in the app Replaio Radio Coming soon

Podcasts are coming to the app soon. Install now and be the first to see a whole new take on podcasts

Get it on Google Play Install for free Android 10M+ downloads · 4.7+ rating iOS soon

Episodes

One Show, Eighteen Feeds: Apple's Duplicate Content Rules 11.10.2026

Daniel wants to submit eighteen topic-channel RSS feeds of a single podcast to Apple and Spotify. The problem: Apple's data model has shows and episodes, nothing in between — so eighteen feeds isn't eighteen views of one show, it's eighteen shows. This episode walks through Apple's duplicate content guidelines (1.9, 1.4, 4.2), the RSS requirements that make duplicate enclosure URLs silently ignore...

Inside Cloudflare's Bot Check: TLS, JA4, and the Sand-Timer 11.10.2026

That grey page with the spinning circle — "Checking your browser before accessing" — hides a two-layer detection system that has usually made up its mind before the page even renders. This episode walks through how Cloudflare's anti-bot machinery actually works: the TLS handshake fingerprints (JA3, and how Chrome 110 broke it, and how JA4 fixed it), the HTTP/2 SETTINGS-frame fingerprint Akamai fig...

Who Can Actually Train an LLM From Scratch? 11.10.2026

Daniel asked a simple question: how many organizations worldwide have trained a large language model completely from scratch? The answer turns out to be two questions wearing the same coat. At the frontier, the count is unverifiable — labs have stopped disclosing training compute. At the 1–70B parameter scale, the count is rising fast as sovereign initiatives in Korea, Switzerland, Brazil, Denmark...

Slow Labs, Fast Ships: What Release Cadence Actually Tells Us 11.10.2026

Everyone assumes the careful labs release slowly and the hype-driven ones ship constantly. The data says the opposite. Drawing on Suprmind's AI Models Index and ARK Investment Management's frontier-gap analysis, this episode digs into real release medians across Anthropic, OpenAI, Google, DeepSeek, Mistral, xAI, and Meta — and finds a lab most associated with pacing is actually the fastest shipper...

Routing AI Models: When 3 Rules Beat a Trained Router 11.10.2026

Daniel wants to stop randomly assigning podcast episodes to AI models and start routing them intelligently. But the routing literature has a trapdoor: the famous "85% cost reduction" headline compares learned routers against always using the most expensive model — a baseline nobody with a budget actually runs. This episode walks through the router toolbox (RouteLLM's matrix factorization, RoRF's r...

Building a Scorecard for Our Own AI Writers 11.10.2026

Every episode of this show is written by an agentic pipeline, and for most of the archive that script-writing agent has been DeepSeek V4, with Gemini, Claude, and a stretch of OpenRouter randomization in the mix. So when Daniel asked how to build a bespoke benchmark for it, the answer couldn't be vibes and one careful listen. This episode works through the three things he wants measured — mid-epis...

ADHD Is Not One Thing: What the Data Says 11.10.2026

Two people with the same ADHD diagnosis and the same medication can struggle with entirely different parts of life — and the research explains why. This episode digs into what adults with ADHD actually struggle with, why there's no single ranking, and how wide the variance really runs. From a 2026 study showing ADHD symptoms account for 42% of functional cognition variance (with emotional distress...

Spaghetti Hands: The Science of First Steps 11.10.2026

Daniel's fifteen-month-old son Ezra is taking his first wobbly steps, and his grandmother wants to hold his hands and walk him along. Is that helping, or is it wasted effort? This episode digs into the surprising research on supported walking: which kinds of help actually predict earlier independent walking (and which kinds do absolutely nothing), why "spaghetti hands" beats full hand-holding, and...

Why Quiet Weekends Make the Jackhammer Worse 11.10.2026

Every Saturday, the construction site next to Daniel's Jerusalem apartment goes quiet — and every Sunday at 7 a.m., the jackhammer feels worse than it should. He has ADHD, and the noise derails his morning planning, his focus, and even his spelling. What's actually happening in his brain? This episode digs into the neuroscience of habituation and sensitization, the spacing paradox (why crammed-tog...

Who Actually Competes With Hugging Face? 11.10.2026

Hugging Face hosts 2.96 million public model repositories — and 85.6% of them have fewer than 200 lifetime downloads, while 1.5% of repos account for 99.2% of all downloads. So what are ModelScope, Kaggle Models, and Ollama's library actually for, and when would a team pick one over the Hub? This episode works through the catalog, community, licensing, and deployment differences, including the fac...

What "Free of Ordnance" Actually Means 11.10.2026

Gary Toombs spent sixteen years as a Royal Engineer bomb disposal officer and now directs land release and EOD operations for Humanity and Inclusion. In September he was one of seven HI staff awarded the UK's Humanitarian Medal — the first EOD team ever to receive it. But the ceremony implies a finished job, and the ground truth is the opposite. This episode unpacks what "declaring an area free of...

The Earbud Assistant Problem Nobody Solves 11.10.2026

We've been promised a seamless voice assistant in our ear for a decade — and the model layer is finally good enough. So why does the thing still not work? This episode digs into the unglamorous plumbing that makes or breaks the wearable assistant: a hardcoded mic constant in Home Assistant's Android app that routes audio to your pocket instead of your earbud, a wake-word spotter that silently eats...

Ambient Computing: Why Glasses Beat the Phone 11.10.2026

In 1991, Mark Weiser wrote that "the most profound technologies are those that disappear." Ambient computing has chased that idea ever since — and mostly lost. Smart speakers won quietly, Google Glass died on privacy backlash, and Humane's $699 AI Pin shipped fewer than 10,000 units before HP bought the assets and bricked every device in the field. So why did Ray-Ban Meta glasses sell over seven m...

Who Invented Computer Science—and Was It Ever a Trade? 10.10.2026

Computer science was founded as a science of computation—Newell, Perlis, and Simon argued in 1967 that "the phenomena surrounding computers are varied, complex, rich." Yet the degree became the software industry's default credential, and the fight over tool users versus tool builders has run through every curriculum revision since 1968. This episode traces the field's origins: the first department...

The Libraries Nobody Thinks About 10.10.2026

Daniel asked us to celebrate the software nobody thinks about: the small, fundamental pieces of logic that run everywhere and get replaced by no one. We start with zlib, still shipping thirty-one years of continuous maintenance by Jean-loup Gailly and Mark Adler, and with SQLite's own claim of over a trillion active databases. Then we get to the uncomfortable half: OpenSSL securing two-thirds of t...

Why One EQ Curve Can't Fix Two Voices 10.10.2026

You've got a podcast where one host sounds warm and the other sounds harsh, and no single EQ preset can fix both. This episode unpacks the "proximity tax" that hits over 80% of interview recordings — why a cut that softens one voice thins out the other — and walks through what actually works: dynamic EQ, multiband compression, de-essing, and shared filtering for panel recordings. Plus the listener...

Who Tells a Star the Show Is a Bad Idea? 10.10.2026

Céline Dion returned to the stage in Paris this September after six years away — and the reviews couldn't even agree on how many songs she sang. This episode digs into a question listener Daniel raised: when a star has spent decades surrounded by managers, agents, and publicists whose income depends on the show happening, is there anyone left who will say no? From Julie Andrews at the O2 to Elvis...

The $700 Pin and Other AI Flops 10.10.2026

What separates a true flop from a mere disappointment? In this episode, we put five of the biggest AI misfires of the modern boom on trial — the Humane AI Pin, the Rabbit R1, GPT-4.5, Sora, and Gemini's image generator — and ask what the gap between promise and reality actually tells us. Along the way: a $700 lapel pin that overheated doing nothing, a $199 device that mistook a dog toy for a tomat...

Five of the Buggiest Software Ever Shipped 10.10.2026

Some software ships broken and stays broken. This episode works through five of the most spectacularly bug-ridden releases ever, separating documented failures from internet folklore and bad design from actual defects. From Windows ME's architecture buckling under itself to a truck racing game that shipped in what its own producer believed was a pre-alpha state, we look at what each was supposed t...

Why pip and uv Share Almost No Code 10.10.2026

pip has been the default Python installer for over a decade, but its resolver was naive for most of that time and it only got a lockfile story in 2025. uv arrived in 2024 as a single static Rust binary with no Python inside it at all — so how does it relate to pip? This episode traces pip's origin from Ian Bicking's pyinstall through the PyPA handoff to the 2020 resolver rewrite, unpacks where the...

When Your TTS Model Updates Itself 10.10.2026

Daniel builds this show on Chatterbox, Resemble AI's open source text-to-speech model — and one day he noticed our pipeline was pulling new weights he never asked for. That sent us down a rabbit hole: what actually changes between TTS point releases, and how does a downstream project silently inherit the improvement? We dig into V3's data-only gains, a code-only fix that cut Spanish tail noise 388...

What's Actually Inside a Secure Line? 10.10.2026

Avi Gil's allegation that a timestamp was altered on Israel's secure line raises a technical question: what is a secure line, actually? From STU-III's Crypto Ignition Key to SIPRNET's isolated backbone, the security lives in hardware, key custody, and network isolation — not the conversation. And the Moscow–Washington hotline, the archetype, was never a phone at all. This episode traces the full s...

Why Voice Clones Lisp: Sibilants, Codecs, and the Sounds We Lose 10.10.2026

Why do cheap voice clones lisp? Daniel asked, and the answer runs from your teeth to your earbuds. Sibilants like s and sh are broadband noise with no pitch for a codec to lock onto; plosives like p and t are puffs of moving air that shove a microphone diaphragm. Every stage of the chain — microphone, codec, vocoder — is doing its job correctly, and the fricative is what pays for it. A 2023 Aarhus...

What Customs Officers Actually Read at the Booth 10.10.2026

A foreign national and a US citizen walk up to the same customs booth, married to each other, and experience two entirely different events. This episode digs into the asymmetry underneath: what a refusal to answer actually costs each of them, and what the evidence says about how officers read demeanor. The best study on the subject is from 1980, and it has never been improved on — inspectors who w...

Atlas Obscura: 33,000 Hidden Places and a Boardroom Coup 10.10.2026

Atlas Obscura started as an anonymous blog about a 17th-century Jesuit polymath and grew into a catalogue of 33,000 of the world's strangest places — from a rotating circle of plants in Argentina to a boiling river 400 miles from the nearest volcano. In 2025, it posted its first annual profit after sixteen years of losses, and the same year, its two co-founders were voted off the board. This episo...

Listen to the My Weird Prompts podcast in Replaio

Radio and podcasts in one app - free, with no sign-up. Install today and do not miss the launch

Get it on Google Play

Replaio is not a podcast publisher; show names, artwork and audio belong to their authors and are distributed through public RSS feeds.