A clearer conversation about AI futures

Doom or Bloom helps you articulate what you expect from advanced AI, why you expect it, and what could change your mind. A short adaptive interview produces a richer profile behind a simple map. A provisional result becomes available when supported coverage is sufficient, potentially after one detailed answer; most paths take six to eight prompts.

How the assessment works

Jev, a TypeSafe model, makes narrow typed judgments about your actual answers. Code selects from authored questions, preserves exact evidence, and combines qualitative categories into a projection. It never generates interview questions or a free-form verdict. The local demo assesses only the evidence you offer, without checking external sources; the interview does not search live news.

The map pairs your expressed Doom–Bloom outlook with how radically you expect AI to transform society. Human influence, expected upside, expected harm, and demonstrated reasoning appear on separate single-axis scales. Policy preferences do not determine your map position.

Missing evidence leaves a position unplaced. Interpretation ranges reflect uncertainty about how to read your answers, not event probabilities. A separate P(doom) card shows a stated or inferred estimate, labeled with its basis and interpretation range. Milestone timing and assumptions retain your exact wording and conditions. None of these views is a calibrated forecast or a statistically validated confidence interval.

An experiment, with limits

This is not a psychological instrument, an intelligence test, or a probability-of-doom calculator. Technical vocabulary, credentials, moderation, and agreement with the authors should not earn points. Coherent extreme views can demonstrate stronger reasoning than an unsupported middle position.

The name foregrounds doom and bloom and may prime your answers. The chosen questions, sources, categories, and weights also involve editorial judgment. Mixed, uncertain, and low-transformation expectations remain valid. Interpretation can be wrong; use “That’s not quite my view” to clarify a claim, or inspect the evidence in your report.

Practical details

A few sentences is plenty. Relevant jokes and honest uncertainty are welcome. Unrelated or unclear replies receive bounded re-asks; repeated misses pause the interview. Restart clears local progress. The assessment warns at 10 lifetime prompts and stops at 12, including clarification.

Versions: assessment 0.6.0; content 0.4.0-draft; rubric 0.1.0-draft; Jev jev-1.13.0. Content expansion and evaluation are in progress. Draft releases are labeled in the interface; representative design approval is not scientific validation.

Built by Travis Fischer. Explore the source and authored methodology.