My Weird Prompts

Open Location Code gives every square on Earth a short alphanumeric address — free, offline, case-insensitive, precise to about 14 metres. Google published the spec in 2014, developers ported it to a dozen languages, and the Navajo Nation used it to address 3,100 homes for EMS. So why is a proprietary, pronunciation-ambiguous rival the one inside 911 dispatch centres? This episode traces Plus Codes from a Zürich spec document to slum addressing in India and Brazil, and lands on the question that actually matters: what happens when the code is precise but the GPS fix behind it isn't.

Key sources:

• google/open-location-code GitHub repo (README, FAQ.txt, CHANGELOG, LICENSE, External_Implementations.md) — https://github.com/google/open-location-code

• Google Open Source Blog, Open Location Code: Addresses for everything, everywhere, 2015-04-30 — https://opensource.googleblog.com/2015/04/open-location-code-addresses-for.html

• Google Plus Codes site — https://maps.google.com/pluscodes/

Episode 5923 — myweirdprompts.com/n/5923 (permanent short link: myweirdprompts.com/827246)

You can't reliably prompt a model into your voice — system prompts plateau. So what actually works? This episode works through three data-generation strategies for style fine-tuning: regenerating transcripts with global comments, hand-editing transcripts and training on the diff, and training a purely negative counterweight against overused phrases. A July arXiv paper by Federico Boggia on epanorthosis (the mid-sentence self-correction tic) tested almost exactly this, and one implementation failed in a very instructive way: a small-data DPO adapter learned to rank outputs correctly and then raised the density of the target phrase. Supervised fine-tuning on the same data cut it by 98%. We map each method onto that evidence, then get to the practical recipe — how many examples you actually need, why "200 transcripts" is the wrong unit of counting, and why LoRA or QLoRA is the unambiguous choice for a style adapter.

Key sources:

• DeepSeek-V4 Preview announcement, 2026-04-24 — https://www.deepseek.com/en/news/v4-preview/

• DeepSeek-V3.1 model card — https://huggingface.co/deepseek-ai/DeepSeek-V3.1

• Federico Boggia, Artificial Epanorthosis, arXiv, 2026-07-23 (v2 2026-07-28) [primary research] — https://arxiv.org/abs/2607.21498v2

Episode 5922 — myweirdprompts.com/n/5922 (permanent short link: myweirdprompts.com/322295)

A listener in Jerusalem, a decade here and fluent in Hebrew, took his sixteen-month-old son to urgent care twice in two months. Both times the boy was in real respiratory distress. Both times the layer that was supposed to catch him before the hospital didn't. This episode traces what actually failed — a documented oxygen saturation and no doctor for twenty minutes, a health fund phone line that couldn't answer, a billing dispute surfacing at pediatric triage — and then zooms out to the numbers nobody wants to look at: 1,369 Israeli doctors leaving between 2023 and 2025, a 94% rise over the previous decade, and a first-ever government waiting-time report showing MRI waits nearly twice as long at one fund as another. The hospital care was good. The layer underneath is the story.

Key sources:

• Terem , urgent care network claims — https://www.terem.com/on-us/?lang=en

• Ynet, 2026-09-06, physician emigration study — https://www.ynetnews.com/health_science/article/bjsin3quze

• Globes, 2026-09-06 — https://en.globes.co.il/en/article-dramatic-rise-in-doctors-emigrating-from-israel-1001554506

Episode 5921 — myweirdprompts.com/n/5921 (permanent short link: myweirdprompts.com/154525)

Dr. Heinz Edgar Lehmann read the chlorpromazine papers in his bathtub in 1953, published the first English-language clinical study on the drug, and helped end the asylum era — then spent his later years warning that the revolution he started had produced cookbook psychiatry. This episode works through his biography, the science, and the black-and-white patient interview films that keep drawing millions of viewers seventy-five years later. Was Lehmann unusually compassionate by the standards of his era, or do the recordings just make him look that way?

Key sources:

• Wikipedia, Heinz E. Lehmann — https://en.wikipedia.org/wiki/Heinz_E._Lehmann

• McGill Reporter, In memoriam (Joel Paris), 1999 — https://reporter-archive.mcgill.ca/Rep/r3115/memoriam.html

• McGill Health e-News, Marking 70 years of psychiatric research at the Douglas, 2024-03-13 — https://healthenews.mcgill.ca/marking-70-years-of-psychiatric-research-at-the-douglas-celebrating-heinz-lehmann/

Episode 5920 — myweirdprompts.com/n/5920 (permanent short link: myweirdprompts.com/774393)

A fifteen-month-old drops a cracker into his water cup, stirs it, drinks the cloudy water, then eats the cracker. His parents call it raccoon behavior. Pediatric feeding specialists at Solid Starts call it normal exploration — and the research suggests they're right. This episode digs into what's actually driving the behavior: why teething turns out to be the weakest explanation, how the Infant and Toddler Sensory Profile explains why some toddlers mix and pour while others couldn't care less, and what Piaget's tertiary circular reactions tell us about a kid who isn't eating so much as running a procedure. We also look at the exposure research — ten tastings, a seven-week intervention, intake up hundreds of percent — and ask whether the mess at the dinner table might be doing developmental work that's hard to replicate any other way.

Key sources:

• Solid Starts, How to Teach Baby to Drink from a Cup — https://solidstarts.com/cup-drinking/

• BabyCenter community thread, dipping food in water or putting it in cup — https://community.babycenter.com/post/a4014915/dipping_food_in_water_or_putting_it_in_cup

• Tournier C, Demonteil L, Canon F et al., A new masticatory performance assessment method for infants: A feasibility study, *Journal of Texture Studies*, 2019, PMID 30667063 — https://pubmed.ncbi.nlm.nih.gov/30667063/

Episode 5919 — myweirdprompts.com/n/5919 (permanent short link: myweirdprompts.com/913929)

Daniel has an IP camera in his son's room doing double duty — Frigate watches the video, but he also wants a continuous audio line piped to his phone so a cry wakes him up. The camera won't hand out an audio-only feed, so something has to split the audio off the RTSP stream. This episode works through the architecture: why go2rtc beats Frigate's embedded restream, how to fan out one camera connection into multiple consumers, where the single transcode belongs, and which codecs actually matter when the sound you're trying to catch is a cry at three in the morning.

Key sources:

• go2rtc README (GitHub, AlexxIT/go2rtc, master) — https://github.com/AlexxIT/go2rtc

• go2rtc Streams module docs (internal/streams/README.md, master) — https://github.com/AlexxIT/go2rtc/blob/master/internal/streams/README.md

• go2rtc MPEG module docs — https://go2rtc.org/internal/mpeg/

Episode 5918 — myweirdprompts.com/n/5918 (permanent short link: myweirdprompts.com/227184)

Before "computer" meant a machine, it was a job title — held for centuries by people, and by the 1940s, mostly by women. This episode reconstructs the actual desk: pre-printed forms, mathematical tables, colored pencils, and mechanical calculators with hand cranks. From the WPA's Mathematical Tables Project, where one or two people in a 450-person workforce could do division, to Los Alamos, where Naomi Livesay rewired IBM plug boards and hand-checked the machine's output, to Bletchley Park's Bombe operators working from footstools. What emerges is a story about designing computation at scale before computers existed — and the factory architecture, error-checking, and even the accounting units (kilogirls, girl-years) that made it work.

Key sources:

• Bletchley Park Trust, Who were the Codebreakers? — https://www.bletchleypark.org.uk/our-story/who-were-the-codebreakers/

• NIST, Prehistory: The Math Tables Project, updated 2022-06-19 — https://www.nist.gov/mathematics-statistics/prehistory-math-tables-project

• Wikipedia, Computer (occupation) — https://en.wikipedia.org/wiki/Human_computers

Episode 5917 — myweirdprompts.com/n/5917 (permanent short link: myweirdprompts.com/881628)

Grey rocking and information restriction are the two most-recommended tactics for people stuck with a toxic parent—yet the phrase traces back to a single 2012 blog post with no clinical research behind it. This episode asks whether the underlying problem was already solved elsewhere: in intelligence work and cybersecurity, where need-to-know, OPSEC, and firewall policy decide in advance what gets shared. The catch is that you can never say any of it out loud. We dig into the behavioral mechanism underneath grey rocking—reinforcement, extinction, and the extinction burst that makes week one worse than doing nothing—and why a half-hearted attempt can be more harmful than never starting. Then we follow the threat past the single conversation into data aggregation: how flying monkeys pool harmless fragments into a coherent case, and why deciding who to trust is the wrong question. Along the way: the real cognitive cost of expressive suppression, and why a policy written down in advance might work where willpower doesn't.

Key sources:

• NSA/NCSC — https://www.dni.gov/files/NCSC/documents/nittf/OPSEC_for_All_July2022.pdf

• NCSC — https://www.dni.gov/files/NCSC/documents/nittf/Understanding_OPSEC_Bulletin_1.pdf

• The Conversation — https://theconversation.com/what-is-the-grey-rock-method-for-dealing-with-narcissists-or-difficult-family-members-282047

Episode 5916 — myweirdprompts.com/n/5916 (permanent short link: myweirdprompts.com/909468)

What actually holds a simulated world together when multiple AI agents act inside it? This episode opens the repos on three frameworks that answer the question completely differently. Concordia from Google DeepMind makes world state a first-class Game Master component — a plain Python dict whose contents are authored by the model at every step, with a SQLite/Postgres storage protocol that ships as an interface but not an implementation. Generative Agents goes the other way: a deterministic tile grid in CSV files, per-agent JSON scratch files, and a memory system with a 0.995 recency decay. CAMEL has no persistent world object at all — state lives in the conversation history and a Task object. Along the way: why agent memory and world state must be separate, how entity-aware event filtering produces information asymmetry, and the four-line docstring about a rent event that silently refunded every agent's spending — and why the fix was atomic wallet deltas, not a better prompt.

Key sources:

• Concordia repo (google-deepmind/concordia), commit 2d0c2a4, README + CHANGELOG v2.4.0 (2026-03-06) — https://github.com/google-deepmind/concordia

• `concordia/components/game_master/world_state.py` — https://raw.githubusercontent.com/google-deepmind/concordia/main/concordia/components/game_master/world_state.py

• `examples/concordia_island/sim/world_state_protocol.py` (2026) — https://raw.githubusercontent.com/google-deepmind/concordia/main/examples/concordia_island/sim/world_state_protocol.py

Episode 5915 — myweirdprompts.com/n/5915 (permanent short link: myweirdprompts.com/761242)

A 28-storey walk down a stairwell because someone burnt toast — that's the everyday failure mode of high-rise fire systems. This episode walks the full chain: ionization vs. photoelectric detection, aspirating smoke detection, the addressable fire alarm control panel, wet/dry/pre-action/deluge sprinklers, stairwell pressurization and stack effect, elevator recall, and phased evacuation. Then the data: 150,000+ automatic alarm call-outs in England in the year ending March 2019, roughly two-thirds false, with cooking and faulty equipment the top causes for eight straight years. The hard tradeoff: the kitchen is both the most likely place for a real fire and the most likely source of a nuisance alarm, so you can't solve both with one sensor. Featuring the UK's "stay put" compartmentation model and what false alarms cost in complacency and blue-light road risk.

Key sources:

• GOV.UK — https://www.gov.uk/government/publications/evacuation-guidelines-for-fire-and-rescue-services-frs/evacuation-guidelines-for-fire-and-rescue-services-accessible

• NIST — https://www.nist.gov/publications/stair-evacuation-older-adults-and-people-mobility-impairments

• GOV.UK — https://www.gov.uk/government/publications/grenfell-tower-inquiry-phase-2-report-government-response/grenfell-tower-inquiry-phase-2-report-government-response-html

Episode 5914 — myweirdprompts.com/n/5914 (permanent short link: myweirdprompts.com/859712)