Question 1

How do I get the raw answers instead of the rendered page?

Accepted Answer

GET /api/verdict returns asked_at, model, latency_ms, cost_usd, the raw answers, usage, the exhibits, the AI headlines, per-feed status and the exact request payload. The x-verdict-cache header distinguishes an edge hit, the shared KV ruling, a fresh request throttled inside the one-minute window, a miss that called Jev, and a stale value returned because Jev was unreachable.

Question 2

Why ask the same question twice, once as a choice and once as a noul?

Accepted Answer

The choice supplies the yes/no ruling shown on the page, but a two-option choice tends to snap to 0 and 1. The noul states the same claim as a proposition, so the calibrated probability underneath the ruling stays in a usable range. A score question grades how doomed humanity looks from the day's AI headlines, and a second noul covers the next decade.

Question 3

Will this keep costing money if I leave it deployed?

Accepted Answer

No, the design caps it. The latest ruling lives in a KV namespace shared across edge locations and stands for ten minutes behind a per-location edge cache, so the worst case is about 1,440 Jev calls a day, roughly six cents. A fresh request is allowed at most once a minute globally, and a zone rate-limiting rule blocks any single IP hammering the endpoint.

Question 4

Can I run it locally or point it at another endpoint?

Accepted Answer

Yes. Copy .dev.vars.example to .dev.vars, put a TypeSafe API key in it, and npm run dev serves the site on localhost:8788. TYPESAFE_API_URL in the same file repoints the Pages Function at a different endpoint. Deployment is a direct wrangler upload to the should-ai-kill-us-all Pages project, with the key stored as a Pages secret.

Question 5

What is the state actually built from?

Accepted Answer

Twelve human-interest exhibits plus the top ten AI headlines, fetched in parallel with a four-second timeout per feed. Any feed can fail and the quota is simply filled by the rest, which means the state behind a given ruling is not fixed and not reproducible from the outside. Categories pool feeds because Reddit rate-limits often enough to matter.

Question 6

How often is a fresh ruling actually made?

Accepted Answer

At most once a minute globally. A new Jev call fills a KV namespace shared across edge locations and then stands for ten minutes behind a per-location edge cache, so asking again inside that window returns the standing ruling with a throttled cache header rather than spending a call.

Question 7

What is the honest reading of this project?

Accepted Answer

An art piece, not a measurement. The exhibits are deliberately filtered to stay absurd, and the ruling is a single sample from a model with no text output, so it cannot explain itself or argue back. It is interesting because you can inspect the exact payload behind any ruling, not because the verdict means anything about the world.

Should AI Kill Us All

About this project

Highlights

Questions about this entry

More Jev resources like this

From the Jev guides

From the Jev community