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

26:08

Your News Feed Is Filtering for Failure

My Weird Prompts ·

2026/09/22

Around 40% of people worldwide now say they sometimes or often avoid the news, up from 29% in 2017. It's not fragility — it's a rational response to a feed that filters for failure. This episode unpacks why every individual story can be true while the composite is a lie, what constructive journalism actually means (it isn't positive news), and how the ad model itself shapes what you're allowed to see. Then we map real tools to three distinct needs: a fire hose of weird global stories, hopeful

en-us

22:50

What a Hot Bath Actually Does for You

My Weird Prompts ·

2026/09/22

Two offhand comments — one about hygiene, one about water use — convinced Daniel that baths were pointless. He's been a shower person ever since. But when he finally climbed back into a tub with frankincense and myrrh salts, sick and achy, he wrote in to ask what the science really says. It turns out the research on immersion bathing is stronger than the folklore ever was, and he was robbed. This episode digs into the head-to-head trial that pitted baths against showers, the sleep mechanism that

en-us

22:46

Heads vs Layers: How Model Merging Actually Works

My Weird Prompts ·

2026/09/22

When someone asks how open source enthusiasts combine two AI models into a hybrid, the intuitive answer — snap the attention heads together — turns out to be wrong in an interesting way. Heads are a granularity of control; layers are the unit of assembly. This episode unpacks what a transformer head actually is, from the query-key-value subspaces inside multi-head attention to the strange finding that some heads are actively interfering with the model's job, and then follows that thread into the

en-us

23:07

When Do LLMs Suddenly Get Good at Things?

My Weird Prompts ·

2026/09/22

In 2022, researchers defined an ability as "emergent" if smaller models lack it and larger models have it — meaning you couldn't have predicted it from the smaller models you already had. Since then, a fierce debate has unfolded over whether these jumps are a real property of the models or an artifact of the metrics we choose. This episode works through the strongest arguments on both sides, from the "mirage" paper that manufactured emergence on demand in vision tasks to the loss-perspective

en-us

20:42

Dollar Plumbing: Why Bessent's Iran Threat Has Teeth

My Weird Prompts ·

2026/09/22

Treasury Secretary Scott Bessent threatened to cut off any foreign company that fuels, lands, or sells tickets for Iranian airlines — "you will be cut off from the dollar system." This episode unpacks what that actually means: how CHIPS and Fedwire make the dollar a system rather than just a currency, why currency-based jurisdiction lets OFAC reach transactions with no American party, and why Iran's multi-billion-dollar crypto workaround keeps running into a freeze switch held by a stablecoin

en-us

29:11

What Does a Space Force Actually Do All Day?

My Weird Prompts ·

2026/09/22

The U.S. Space Force became its own branch in 2019, but the military had been flying satellites since the late 1950s — so what actually changed? This episode digs into the physical inventory Daniel asked about: missile-warning satellites, military SATCOM, GPS, space-domain awareness, and the planetary-scale ground infrastructure that most people never see. It also goes international, comparing the two countries on Earth with fully independent space forces — the U.S. and China's PLA Aerospace Force —

en-us

24:21

Life at 76 Degrees North: Inside Pituffik Space Base

My Weird Prompts ·

2026/09/22

Pituffik Space Base sits 1,500 kilometers from the North Pole on Greenland's northwest coast — the northernmost installation in the entire U.S. Department of Defense. It's also reportedly the most expensive overseas base the United States operates, at over $100 million a year, roughly ten times the cost of an average Air Force base. So what does it actually take to keep several hundred people, a 10,000-foot runway, and a missile warning radar running year-round at 76 degrees north? We dig into who

en-us

21:29

Why Linguists Left the AI Room

My Weird Prompts ·

2026/09/21

Large language models are, at their core, language technologies. So why did the people who study language for a living become peripheral to building them? This episode traces the hidden linguistic DNA inside the transformer architecture — from Jakob Uszkoreit's parse-tree insight to the raising-versus-control distinction that models seem to have internalized without being taught — and asks what got lost when the frontier moved from theory-driven NLP to compute-and-data-driven pretraining. We work

en-us

22:11

Hunting the Tells That Vanish When Named

My Weird Prompts ·

2026/09/21

Daniel noticed "failure mode" showing up everywhere in our own scripts, and wondered what other verbal tics are hiding in a corpus of thousands of episodes. There's no off-the-shelf LLM tick detector — but there is a whole discipline built for exactly this. This episode walks through corpus linguistics tooling: quanteda's keyness function, corpkit's subcorpus slicing, pycorpdiff's changepoint detection, and the surprisingly large body of research on which words AI actually overuses. The answer isn't

en-us

19:59

Small Models as Rewriters, Not Writers

My Weird Prompts ·

2026/09/21

Daniel noticed certain idioms keep surfacing in the show's dialogue, and system-prompting the problem away made it worse — the model started using the banned phrases more. This episode works through the mechanics of why prohibition prompts fail, why multi-agent correction passes leak feedback into the final text, and why regex produces ungrammatical holes. The fix is a small text-in-text-out language model operating as a deterministic substitution pass, with a grammar engine that makes commentary

en-us

25:35

When Your TTS Model Eats the Numbers

My Weird Prompts ·

2026/09/21

Text normalization is the invisible layer that turns written text into something a TTS model can actually speak: "1,234" becomes "one thousand two hundred thirty-four," "Dr." becomes "Doctor," and "2026" stops being a string of digits. It's also the layer this podcast doesn't have, which is why acronyms and numbers have been hand-patched for months. This episode walks through the vocabulary — text normalization for TTS, inverse text normalization and punctuation restoration for ASR — then digs into

en-us

27:10

When Local AI Tools Phone Home

My Weird Prompts ·

2026/09/21

If a tool runs inference on your device, it has no legitimate reason to touch the network. That makes outbound traffic from a "local" AI tool one of the few clean security signals left — and it's exactly the signal attackers work hardest to avoid. This episode traces keystroke capture from the permission prompt down to the SQLite file on disk, then follows the data through the exfiltration chain: commodity sideloaded APKs firing plain HTTP POSTs, and the sophisticated end where

en-us

18:25

Thermostats, Thermal Mass, and the Buildings That...

My Weird Prompts ·

2026/09/21

A thermostat looks like the simplest device ever built: sensor, set point, switch. But the bleeping battle between an IR blaster and an air conditioner is really a lesson in control theory — bang-bang controllers, deadbands, integral windup, and models that forecast accurately yet get the sign of a cooling action wrong. From there the conversation widens to buildings themselves: thermal mass as a reservoir, phase-change materials in concrete roofs and brick, courtyards in Seville, wind towers,

en-us

24:51

What Hospice Workers See at the Deathbed

My Weird Prompts ·

2026/09/21

A UK-wide survey of 104 Marie Curie palliative care staff found that 97% had witnessed at least one deathbed phenomenon in five years — an average of 7.7 each. Visions of deceased relatives, terminal lucidity in dementia patients, clocks stopping at the moment of death. And yet a chart audit found only 8.3% of these events ever get documented. This episode works through the research literature on end-of-life experiences: what palliative nurses, doctors, and death doulas actually report, how

en-us

22:36

The Other Half of Hugging Face: Why BERT Still...

My Weird Prompts ·

2026/09/21

Everyone names Llama, Gemma, Mistral, and DeepSeek — but BERT is the second-most-downloaded model on Hugging Face, and encoder-only models collectively out-download decoder-only models nearly three to one. This episode digs into the taxonomy problem hiding in plain sight: two different animals sharing one foundational task, text in and text out. We trace the arc from BERT's pre-train-then-fine-tune breakthrough through ModernBERT, NeoBERT, and mmBERT, look at the Ettin benchmark results where a 150M

en-us

27:06

Two Boxes: ASR and the Text Fixer Behind It

My Weird Prompts ·

2026/09/21

Raw ASR output is a wall of uncased, unpunctuated words. The accessory models that clean it up are the back half of the pipeline nobody talks about. This episode traces how single-purpose models like FullStop and NVIDIA's NeMo punctuation-and-capitalization tools evolved into unified multi-task fixers — and whether the whole thing is collapsing into just two components: the ASR engine and one text-to-text repair model. Along the way: why punctuation and capitalization are the same problem wearing

en-us

22:24

Bathtime Safety: What the Data Actually Says

My Weird Prompts ·

2026/09/21

Bathtime looks simple until you read the research. Two-thirds of child drownings inside the home happen in the bathtub, and the most common cause of a supervision lapse is walking away to grab a towel. This episode works through the actual numbers: why 37–38°C is the safe range, why the 1978 Feldman study in Pediatrics changed tap water regulation, and why a 2023 University of Chicago Burns study found that 95% of bathing scalds involve running water rather than the standing bath. Plus the 2025

en-us

23:07

Wills, Superstition, and the One Event With 100%...

My Weird Prompts ·

2026/09/21

Daniel wrote in with a confession: he's a prepper who plans for everything except the one event with a hundred percent probability. He's deeply superstitious about death, studied law in Ireland, and lives in Israel with a wife and one kid — and he's seen how wills become weapons in dysfunctional families. So how does someone who's scared to even think about this get started? And how do healthy families handle it with grace instead of turning the document into a stick? This episode covers the

en-us

19:23

Why Atheists Still Knock on Wood

My Weird Prompts ·

2026/09/21

Daniel wrote in with a puzzle: he's Jewish, he believes in God, he's not especially observant — and he still can't bring himself to say the word "death" out loud. He knows Judaism doesn't forbid it. He knows it's just a bad-omen reflex. So why does it bite so hard, and why do so many atheists nod along? This episode digs into what superstition actually is, how it differs from religion, and what the research says about whether the two rise and fall together. The answer is counterintuitive: the

en-us

20:35

AI Wrote His Yom Kippur Prayer

My Weird Prompts ·

2026/09/21

When Daniel couldn't find his good clothes or his iron on Yom Kippur, he stayed home for the first time in his life — then opened ChatGPT and asked for a prayer in the Orthodox tradition. What came back was brief, clear, and woven with details from his own year. This episode traces the real, funded world of AI prayer apps like Kavanah and Hallow, the AI church services staged in Fürth, Helsinki, and Lucerne, and the halachic split between the cheftza view (correctness is all that matters) and the

en-us

2,652 results

Similar Podcasts

Lead With AI

Tamara Nall

Two Blokes Talking Tech

Trevor Long and Stephen Fenech

Pioneers of AI

WaitWhat

The Brave Technologist

Brave Software

The Good Tech Companies

HackerNoon