My Weird Prompts

Federal fatality tables rank America's deadliest cars — and most of them are sitting in rental lots right now. The Kia Rio leads at 170 driver deaths per million registered vehicle years, followed by the Nissan Versa at 164, with minicars as a class averaging four times the fleet-wide rate. But danger and misery are two different lists: muscle cars like the Charger and Challenger land on the same table for entirely different reasons, driven by who's behind the wheel rather than what hits them. And there's no global ranking, because the data required to build one doesn't exist in most countries. This episode works through the numbers, the mass-mismatch physics behind minicar fatalities, and why the rental counter never shows you any of it.

Key sources:

• IIHS — https://www.iihs.org/news/detail/driver-death-rates-remain-high-for-small-cars-models-with-powerful-engines

• IIHS — https://www.iihs.org/ratings/driver-death-rates-by-make-and-model

• The Auto Wire — https://theautowire.com/2026/08/06/deadliest-cars-and-trucks-of-2023-what-the-new-iihs-fatality-data-really-shows/

Episode 5998 — myweirdprompts.com/n/5998 (permanent short link: myweirdprompts.com/565541)

We finally got someone counting. A listener sent in a two-parter: first, an audit-style walkthrough of our Podtrac numbers — what "unique download" really means, why audience and downloads tell completely different stories, and what "audited" does and doesn't certify. Then the fun half: planning the Cave Retreat. How long would it take to absorb the entire back catalogue if you subtract sleep and eating? What does a one-week version look like versus the full expedition? And what happens to your brain when you come back into daylight? It's a measurement episode and a survival-planning episode, which is a combination only this show would produce.

Key sources:

• My Weird Prompts, fetched 2026-10-15 (5,875 episodes, 2,637 hours, 110/165 days, $0.20/episode). — https://www.myweirdprompts.com/about

• My Weird Prompts, fetched 2026-10-15 (4556+ episodes). — https://www.myweirdprompts.com/media-resources

• My Weird Prompts Data Portal, fetched 2026-10-15 (4.7K episodes, 2115h, 19.8M words). — https://www.myweirdprompts.com/data/

Episode 5997 — myweirdprompts.com/n/5997 (permanent short link: myweirdprompts.com/005549)

What would it take to turn a few thousand existing podcast episodes into a continuous video documentary channel — original audio, transcripts, explainer graphics, stock footage? This episode works through the real numbers: raw 1080p generation has collapsed to under a dollar a minute, but a finished, accepted minute still costs around $400 against $4,500 for traditional production. The gap isn't compute. It's the retry loop, the review queue, and taste applied at volume. We look at what's actually shipping in audio-driven avatars, what an invideo case study reveals about survival rates, and why a transcript-first pipeline favors anyone who already has an archive.

Key sources:

• HeyGen — https://www.heygen.com/blog/ai-video-generator-pricing

• SlopTV — https://sloptv.co/guides/what-a-minute-of-ai-video-costs-2026

• Business Insider — https://www.businessinsider.com/creators-ai-tv-series-38000-credits-seedance-artlist-the-arc-2026-10

Episode 5996 — myweirdprompts.com/n/5996 (permanent short link: myweirdprompts.com/952181)

When your text-to-speech reads "makolet" as MACK-oh-let, you have a language identification gap, not a bug. No LID model — GlotLID, fastText, NLLB's lid218e — has a heb_Latn label, so Hebrew written in Latin letters is invisible to every system downstream. This episode walks through the token classifier Daniel shipped to fix it: BIO tagging over xlm-roberta-base, character-span output, and a three-pass pipeline that converts Latin-script Hebrew to Hebrew script before handing it to a Hebrew voice. The real story is the dataset methodology — term-seeded generation, blind annotation by a different model family enforced in code, human review of disagreements, and term-disjoint splits. Plus why precision beats recall here, why a flagged span is work rather than a label, and what a Parakeet fine-tune on dictated audio adds to the picture.

Episode 5994 — myweirdprompts.com/n/5994 (permanent short link: myweirdprompts.com/994167)

NVIDIA's speech recognition lineup isn't one model — it's a portfolio, and the names don't tell you what you're getting. Parakeet, Canary, and Nemotron Speech are brands, not architectures; underneath, nearly all of them share the same FastConformer encoder. This episode walks the whole family tree one model at a time: why Parakeet v3 is not a straight upgrade over v2 (it's slightly worse on English), how the Parakeet unified model makes the streaming tax visible in a single page of numbers, why Canary's forty-second clip limit catches people out, and where the non-commercial license hides. Plus a look at a new fine-tune of Parakeet v2 that handles Hebrew words mixed into English sentences, shipped with on-device GGUF builds. Less benchmark, more map — the goal is knowing which model fits which workflow.

Episode 5995 — myweirdprompts.com/n/5995 (permanent short link: myweirdprompts.com/099291)

Every renovation show ends in the same room: open plan, pale finishes, white quartz counters, generic styling. Is that what homeowners want, or what the production business needs? This episode pulls the wallpaper off the genre — from Fixer Upper's shiplap empire to Love It or List It's lawsuit — and asks who's actually designing: the host, or a team optimizing for the camera. We compare the HGTV transformation model against Grand Designs' process-is-the-drama approach, look at why the formula is really a production calendar in a narrative costume, and find the moment one hit show proved the sameness was a choice, not a market verdict.

Key sources:

• Seattle Times / NYT, What goes on behind the cameras at home makeover shows? (via Union-Bulletin) — https://www.union-bulletin.com/seattle_times/what-goes-on-behind-the-cameras-at-home-makeover-shows/article_33da42d5-7b66-58e0-922e-0749bd8f092c.html

• HowtoHome, TV Design Shows — https://www.howtohome.com/tv-design-shows/

• ScreenRant, 12 Things You Need To Know About The Making Of Love It Or List It (updated Jun 24, 2023) — https://screenrant.com/behind-the-scenes-making-love-it-list-it/

Episode 5993 — myweirdprompts.com/n/5993 (permanent short link: myweirdprompts.com/566672)

DRAM prices have gone vertical: one 32GB DDR5 kit jumped from $191 to $1,118 in a year, and TrendForce expects contract prices to climb another 90–95% quarter over quarter. The cause isn't broken fabs — it's HBM. Samsung, SK Hynix, and Micron are shifting wafer capacity to high-bandwidth memory for AI accelerators, and each gigabyte of HBM eats roughly three times the wafer area of standard DRAM. Consumers get what's left, and Intel's chief executive says that's "shorter shrift than AI players." Waiting it out isn't a plan: Samsung expects tight supply through 2028 and Micron's first New York fab won't produce until 2030. So we count down the top ten most ridiculous legal workarounds — from bundle deals and decommissioned servers to RAM flipping, eBay price trackers, e-waste channels, and the one entry that beats them all.

Key sources:

• TrendForce — https://www.trendforce.com/presscenter/news/20260202-12911.html

• TrendForce — https://www.trendforce.com/presscenter/news/20260331-12995.html

• Tom's Hardware — https://www.tomshardware.com/pc-components/ddr5/32gb-of-ddr5-6000-ram-now-costs-usd450-ai-induced-shortage-reaches-worst-milestone-yet

Episode 5992 — myweirdprompts.com/n/5992 (permanent short link: myweirdprompts.com/685764)

Hosted batch inference is one of the best deals in AI right now — half price, no latency requirement, just hand over a file. But what if you want that on your own GPU? This episode unpacks why the hosted batch API is really a scheduling and billing construct, and what you actually have to build locally to reproduce it. We walk through continuous batching, the KV cache budget that secretly governs your concurrency, and the two shapes a local batch job can take: offline in-process, or server plus concurrent client. Then we compare the real options — vLLM's offline class and its newer run-batch command, SGLang's prefix-sharing strengths and AMD weak spots, Ollama's development-tool defaults, llama.cpp's portability, and why Hugging Face's text generation inference is no longer a starting point.

Episode 5991 — myweirdprompts.com/n/5991 (permanent short link: myweirdprompts.com/694162)

What makes a recurring bit actually work — and when does it turn into an hour of contextless jingles? Drawing on structure-as-co-host thinking and the cautionary tale of trigger-fired jingles, this episode pitches twenty brand-new recurring beats: late corrections to an unbooked ad, an unearned pre-close apology, a stinger nobody explains, a doorbell no one answers, listener audio instead of listener mail, and a character who exists only in the show notes. Every pitch carries the same four specs — what it is, how long it runs, where it sits, and why it can't swamp the topic — plus a running debate over the difference between a beat that adds context and one that just adds noise.

Key sources:

• CitizenRacecar, Episode 3: Segments, Bits, and Callbacks (undated; accessed 2026-10-15). — https://podcast.citizenracecar.com/episode/episode-3

• Pod Bible, Where to start with… Three Bean Salad (2024-02-08). — https://podbiblemag.com/where-to-start-with-three-bean-salad/

• Multitude, My Favorite Podcast is a Late Night Show by Eric Silver (undated; accessed 2026-10-15). — https://multitude.productions/blog/structure

Episode 5989 — myweirdprompts.com/n/5989 (permanent short link: myweirdprompts.com/562844)

A listener wrote in with a project: raise his fifteen-month-old son to speak like an English gentleman, complete with a vocabulary substitution list and a training programme. What follows is a genuinely interesting collision between comedy and developmental linguistics. We dig into Patricia Kuhl's landmark Mandarin work — four weeks of live sessions matching ten months of immersion, while the same content on video or audiotape produced literally nothing — and what that means for screen-based and audio-based language input. Then we take the vocabulary list seriously: the U and non-U distinction from Alan Ross's 1954 paper, the substitutions that survived into modern usage, and the ones that are pure costume. Along the way: sensitive periods for phonetics, why motherese is doing real work, when children start shifting toward their peers' dialect, and why verb diversity in parent input predicts a child's later vocabulary more than accent ever will.

Key sources:

• Kuhl, Tsao & Liu, Foreign-language experience in infancy, PNAS 2003 — https://www.pnas.org/doi/10.1073/pnas.1532872100

• UW News, Social interaction plays key role in how infants learn language, 2003-07-14 — https://www.washington.edu/news/2003/07/14/social-interaction-plays-key-role-in-how-infants-learn-language-studies-show/

• Johnson & White, Developmental sociolinguistics: Children's acquisition of language variation, WIREs Cognitive Science 2019 — https://uwaterloo.ca/lab-infant-development-language/sites/default/files/uploads/files/johnson_et_al-2019-wiley_interdisciplinary_reviews_cognitive_science.pdf

Episode 5988 — myweirdprompts.com/n/5988 (permanent short link: myweirdprompts.com/758450)