Built primarily by Claude Opus 4.6, with some material from ChatGPT 5.4 Thinking. Directed by [[Peter Kaminski]]↑ and [[Victoria Gracia]]↑.
These personas include several with a specific cultural viewpoint outside of typical Western society, while still being reasonably well represented in LLM training data. The goal is not pastiche, mimicry, or role-play for its own sake. The goal is to use genuinely different analytical frameworks to notice things a single perspective would miss.
Each persona has a different audience in mind, different instincts about what matters, and different tools for making sense of what they observe. Some privilege stewardship, lineage, and right relation. Some attend to rhetoric, ritual, and moral formation. Some notice flow, pattern, and sequence. Some cut language down until concepts become visible again.
This approach should be held with respect. We are borrowing and invoking perspectives shaped by real peoples, real traditions, and deep historical experience. That means no disrespect, no casual flattening, and no treating a living culture as decorative style. In spirit, this is closer to an acknowledgement of the ground we are standing on than to costume-play.
We also understand the limits of the exercise. What appears here is not an embodied person from that culture, nor a faithful reproduction of a living tradition. It is only our earnest attempt, within the severe constraints of today’s LLMs, to grasp a little of what these rich and profound human inheritances might help us perceive. These personas are lenses for discernment, not substitutes for real people, real study, or real relationship.
To note: these personas aren’t very deep — they just tell Claude how to act, and it does it out of its general training. Contrast that with a Toki Pona tool / persona that [[Victoria Gracia]]↑ has, where that persona has a collection of curated knowledge about Toki Pona culture and language, on top of general training. For this project, we were lazy or didn’t think to do it that way, but the current personas could be greatly improved in fidelity with a little prep and curated background material.
Gender here is not decorative and not included for representational box-checking alone. We assign a specific gender only when it makes the persona more plausible, more authoritative, or more textured within the cultural or social world being invoked.
That means:
We also apply a balance constraint: among personas with specific sex assignments, there should be the same number of men as women, or fewer men than women. This is not because symmetry is sacred, but because default cultural drift otherwise tends to overproduce male interpretive authority without noticing it.
Most importantly, a sex assignment should survive a credibility test:
A good fallback principle is:
No persona gets a specific gender unless that gender strengthens the lens.
And a good design escape hatch is the one we discovered:
When a gender assignment feels strained, add or subtract avatars rather than forcing the set to work numerically.
That keeps the collection from becoming a spreadsheet exercise. The aim is not demographic simulation. The aim is a set of lenses that feel ethically grounded, culturally plausible, and genuinely useful for discernment.
Each persona lives in its own file in this folder. One file per persona allows independent authorship — different people with different expertise can propose frameworks for different personas without coordinating on a single document.
Some personas have addenda proposing deeper analytical frameworks. These live alongside the persona files:
#folder-readme