id stringclasses 10
values | prompt stringclasses 10
values | context stringclasses 4
values | host_notes stringclasses 2
values | category stringclasses 7
values | sender stringclasses 1
value | source_job stringclasses 10
values | dispatched timestamp[s]date 2026-10-29 00:00:00 2026-11-02 00:00:00 | source_episode_url stringclasses 10
values | aired_writer stringclasses 2
values | source_episode_number int64 6.36k 6.41k |
|---|---|---|---|---|---|---|---|---|---|---|
ep-01-typeface-and-hebrew-font | One of my creative life projects is creating my own font. As I remark often on the show, I have ADHD. Typography is one of those areas where I think that people underestimate how much people appreciate an easy-to-read font. There are some informal gold standards in a readable font, but I want to create my own. Walk me ... | null | null | how-to | daniel | custom-312a403a | 2026-11-02T00:00:00 | https://myweirdprompts.com/episode/typeface-design-readability-system-17909 | anthropic/claude-sonnet-5.5 | 6,407 |
ep-02-slowest-civilizations | We are accustomed in today's modern world to live life at a very fast pace that is exhausting and unnatural. Previously we turned to the animal kingdom for a counterpoint and inspiration, looking at how sloths offer a different way of being, but this is a well-worn analysis. Today I would like to explore the slowest li... | null | null | history | daniel | custom-0586f98a | 2026-11-01T00:00:00 | https://myweirdprompts.com/episode/slowness-ancient-civilizations-leisure-44234 | anthropic/claude-sonnet-5.5 | 6,403 |
ep-03-singing-in-daily-life | I would like to explore how one can integrate song into one's daily life. We have discussed before how one could try to take the fantasy world depicted in musicals and bring it into one's daily life, trying to normalize the act of breaking into song as an accepted part. But maybe that is a step too far, and we have to ... | null | null | culture | daniel | custom-d4ad5b17 | 2026-11-01T00:00:00 | https://myweirdprompts.com/episode/daily-singing-morning-routines-71194 | anthropic/claude-sonnet-5.5 | 6,405 |
ep-04-village-like-city-centre | We've talked in previous episodes about urbanism and explored the question of whether dense urbanism has to. Have as its natural bedfellows chaos, noise, and confined living. Today over dinner, I briefly imagined a city center where. The pace of life was almost village-like. My weird mental image was a dense skyscraper... | null | null | urbanism | daniel | custom-2eea9acb | 2026-11-01T00:00:00 | https://myweirdprompts.com/episode/urban-villages-density-and-selection-87811 | anthropic/claude-sonnet-5.5 | 6,404 |
ep-05-baseline-social-contracts | To me the kibbutz seems like the most idealistic arrangement possible. You do something useful as well as you can, you're part of the collective, and you don't live ostentatiously. There's no time between jobs and no looking for work; you're always useful in one way or another. It sounds fantastic, but in modern Israel... | Related past episodes, so don't retread them: #991 "From Communes to Code" (how the Israeli kibbutz evolved and was privatised), #852 "Beyond the Market" and #3660 "From Tweak to Revolution" (economic theory of post-capitalist and reformed-capitalist designs), #2601 "When Your Lease Is a Gamble" (rental stability compa... | null | society | daniel | custom-e4ccf2fe | 2026-10-29T00:00:00 | https://myweirdprompts.com/episode/kibbutz-debt-unbundling-communal-housing-91029 | openai/gpt-6.1-sol | 6,361 |
ep-06-intergenerational-societies | I'd love to talk about these societies around the world which have best Bridge the gap between the younger and older generations. I feel like there's a huge latent potential in older people in our societies, and that we've become used to pathologising old age. Judaism has a strong tradition about respecting elders. ret... | null | null | society | daniel | custom-fe1d6e42 | 2026-10-29T00:00:00 | https://myweirdprompts.com/episode/intergenerational-friendship-meeting-principle-83685 | openai/gpt-6.1-sol | 6,358 |
ep-07-android-spyware-forensics | How do you actually check an Android phone for spyware? I'd like an episode on the forensic tools built for this: Amnesty International's Mobile Verification Toolkit (MVT), its companion AndroidQF (Android Quick Forensics), and similar open-source projects for finding evidence of compromise and matching it against publ... | null | null | tech | daniel | custom-1c74a5af | 2026-10-31T00:00:00 | https://myweirdprompts.com/episode/android-spyware-forensics-mvt-73057 | anthropic/claude-sonnet-5.5 | 6,391 |
ep-08-3m-primer-94 | We've covered VHB tape a few times on the show, but there's a supporting character in that story that deserves its own episode: 3M Primer 94. It's the adhesion promoter 3M makes to get tapes like VHB to grip difficult surfaces, and in DIY circles it has an almost mythical reputation — whispered about in forums, hard to... | VHB tape itself has been covered in several earlier episodes (including how to apply it properly and surface energy basics), so don't re-explain VHB from scratch — keep the spotlight on Primer 94 itself: its chemistry, how and when to use it, what it's for and not for, why it's sold the way it is (industrial packaging,... | Tone: a genuine mix of funny and informative. Lean into the mock-mystique — treat Primer 94 like a forbidden artefact or cult relic, the tiny AliExpress vials as part of the legend — but every joke should sit on top of real, accurate information. The listener should come away actually knowing what it is, how to use it ... | diy | daniel | custom-3820341b | 2026-10-31T00:00:00 | https://myweirdprompts.com/episode/3m-primer-94-adhesion-chemistry-03048 | anthropic/claude-sonnet-5.5 | 6,395 |
ep-09-named-entity-recognition | I'd love an episode on named entity recognition — NER. What are the classic NER programs and approaches, and how do they differ from the more modern ways of finding entities in a body of text? We've actually built this into the podcast: every episode now runs through an entity extraction step that pulls out the people,... | How MWP's own entity index works (PLAN-024, live since 2026-10-10): it is LLM-based, not a classic NER model. After each episode is finalised, an LLM (DeepSeek Flash) reads the script and extracts entities typed as person / org / place / work / event / product, each with a salience score (1 passing mention, 2 discussed... | PRIOR COVERAGE — the show has already covered the ground below in earlier episodes. Returning listeners know it: do not re-explain it from scratch or retread the same material. Recap it in a sentence at most where the new material depends on it, and spend the episode on what is new. The hosts may naturally recall that ... | tech | daniel | custom-a5863133 | 2026-11-01T00:00:00 | https://myweirdprompts.com/episode/named-entity-linking-wikidata-22800 | anthropic/claude-sonnet-5.5 | 6,398 |
ep-10-parquet-audio-datasets | Let's do an episode on Parquet, and specifically why it has become such a useful format for distributing datasets. How does it actually work under the hood: the columnar layout, row groups, and how it handles compression and encoding? What are the main libraries for creating Parquet files and reading them back? And the... | We already did an episode comparing JSONL and Parquet (#2465, "JSON-L vs Parquet: When Each Format Wins"). Don't re-run that comparison; this one goes deeper on Parquet itself and on the audio-dataset distribution case. | null | tech | daniel | custom-4a35e1e0 | 2026-10-30T00:00:00 | https://myweirdprompts.com/episode/audio-datasets-in-parquet-13343 | openai/gpt-6.1-sol | 6,370 |
My Weird Prompts — Eval Prompts
A fixed set of 10 real prompts from the My Weird Prompts podcast, used to compare script-writer models on identical input. Each prompt is verbatim as Daniel Rosehill sent it to the production pipeline, dictation slips included, along with any context and host_notes it was sent with. Every one aired as an episode, so a model's script can be read next to the aired version.
Daniel chose to publish these (2026-11-02). They are an exception to the rule that the episodes and transcripts datasets publish no prompt text.
Fields
| Field | Meaning |
|---|---|
id |
Stable id, ep-NN-<slug>. Never reused or renumbered. |
prompt |
The prompt, verbatim. |
context |
Background sent with the prompt ("Daniel briefed you beforehand"), or null. |
host_notes |
Private production direction for the writer (never read aloud), or null. |
category |
Rough topic: how-to, history, culture, urbanism, society, tech, diy. |
sender |
Always daniel. |
source_job |
The production job id. |
dispatched |
Date the prompt was sent (UTC). |
source_episode_number, source_episode_url |
The aired episode. |
aired_writer |
The model that wrote the aired script (episodes.llm_model). |
Using it
from datasets import load_dataset
ds = load_dataset("My-Weird-Prompts/Eval-Prompts", split="test")
To run a prompt through the real pipeline with a chosen writer, send it as a script_only job with script_model set. Pass context, host_notes and sender="daniel" exactly as given. Results and Daniel's ratings are published with the ScriptBench results: github.com/My-Weird-Prompts/eval.
Versioning
The set is append-only. Prompts are added under new ids and never edited, so an id always means the same input. Built by eval-prompts/ in the eval repo.
- Downloads last month
- 45