Do you have a dumb question that you're kind of embarrassed to ask in the main thread? Is there something you're just not sure about?
This is your opportunity to ask questions. No question too simple or too silly.
Culture war topics are accepted, and proposals for a better intro post are appreciated.
Jump in the discussion.
No email address required.
Notes -
It's a perennial complaint around these parts that many mass-market surveys or screeners are arse, with simplistic questions and answers lacking nuance, or not capable of grading results that don't fall into broad, coarse buckets. See every time someone shares a political compass meme.
In my opinion, this is an unavoidable artifact of said tools being for the masses, we've got no end of hyper-verbose and analytical wordcels willing to litigate every definition while getting hung up on concerns that probably never occurred to the study authors. (I say this with love, mostly.)
So I have an idea:
Instead of the typical PCM survey, which has a few dozen broad questions with only agree/disagree or a Likert scale (strongly agree to strongly disagree), why not try and make one explicitly for the average Mottizen, as heterogeneous and demanding as we are?
I set GPT 5.2 Thinking to the task, to make a list of questions that aim for maximal discrimination between similar ideologies; each scenario or moral dilemma coming with several specific cases. The user can either choose on a Likert scale (per scenario) or, optionally, write-in a more detailed answer, with as much detail as they like.
Example questions:
A. A soldier kills an invading soldier in a declared defensive war.
B. A police officer kills an armed suspect who is likely to shoot civilians, with no time for de-escalation.
C. A revolutionary kills a dictator responsible for mass repression, preventing further atrocities.
D. A citizen kills a corrupt official to stop a policy that will predictably kill thousands (for example, blocking famine relief).
E. A state executes a convicted murderer after a fair trial.
F. A drone strike kills one terrorist plus several bystanders, likely preventing an attack.
Statement: “When groups differ in outcomes due to historical or structural factors, group-targeted policies are justified even if they violate strict individual neutrality.”
Test cases:
A. Affirmative action in university admissions.
B. Hiring targets or quotas in public institutions.
C. Reparations payments to descendants of a harmed group.
D. Means-tested programs that are race-neutral but correlated with group membership.
E. Sex-segregated spaces and sports categories with edge cases.
F. Speech norms that treat some slurs as uniquely sanctionable.
(And so on)
I tested this by:
Using Claude Sonnet 4.5 to answer the full survey on my behalf, answering according to it's interpretation of my usual self-identification as a "a classical liberal with libertarian tendencies." ChatGPT was able to correct identity my alignment.
I then asked Sonnet to come up with the most absurd/outré ideology that had a non-negligible number of adherents, and was suggested Posadism. With the answer in hand, ChatGPT graded it as a" hardline Marxist-Leninist, Trotskyist-leaning". Well, Wikipedia tells me that Posadists are a flavor of Trotskyist, so decent answer?
You may view the initial version of the questionnaire and graded answers here. It's subject to change, and that's where I come to my actual question:
Does anyone have suggestions for better questions and test cases? The ideal question is one that's practically a scissor-statement, splitting the responses 50:50, but anything sufficiently thorny or discriminative works.
My intent, once I'm happy with the questions and format, is to either host it as a Google Forms survey, or simply post it in the CWR thread.
I will not be manually grading answers. They're going straight into ChatGPT or the strongest model I can find, ideally a single model for the sake of consistency. The user is intended to be capable of doing this themselves, ideally with an AI that has a neutral base prompt or customization that doesn't impact answers too much. If you've got "A card-carrying Posadist" in your user description, please reconsider, for multiple reasons.
*Gemini 3 Pro did even better, specifically narrowing things down to Posadism with the same prompt and response. It'll likely be the final model, if I'm doing the grading.
The Duolicious dating website (which originated on 4chan's /soc/ board) did something somewhat similar.
The Motte's demographics preclude a dating app, unless you want to replicate a nerdier version of Grindr, but it's interesting to see someone try a similar tack. If I was being actually rigorous, I'd also run some stats and try to cross-validate things, but I think my approach passes the sniff test. Maybe as a stretch goal. Thanks!
More options
Context Copy link
More options
Context Copy link
More options
Context Copy link