Fixed questions
Every model receives the same 50 precisely resolvable questions and the same 11 long-horizon end states.
A longitudinal forecast by frontier AI
What do the
machines think?
We ask the leading AI models the same 50 questions about 2030—then keep asking as new generations arrive. This is a living record of how the machines change their minds.
Explore all 50 predictions ↓
50
questions
about 2030
about 2030
6frontier models
50questions
11possible end states
12forecast runs
2030resolution year
Signal report / current round
Where the machines
agree—and don’t.
Consensus is the median forecast across the latest run from each provider. “Spread” is the gap between the highest and lowest model.
The full forecast
The 2030 index
Select any question to see every model’s forecast, rationale, and movement over time.
Topic
Model view
50 questions
median model range
No questions in this view.
Beyond the resolution window
How does this
story end?
Eleven mutually exclusive end states. One hundred percentage points. No “it depends.” We ask each model where humanity’s relationship with AI ultimately settles—even if it takes a million years.
Median machine forecast
All futures,
weighted.
Hover or tap a segment to identify the end state.
The full taxonomy
Eleven endings
The probability shown is the median across current model forecasts. Open a card for the model-by-model distribution.
Worldviews
Same map.
Different priors.
Each row sums to 100%. The shape of a model’s bar is its compressed theory of the future.
Transparent by design
A benchmark,
not a vibe.
The questions stay fixed. The models change. Reality gets the last word.
Closed book
Models answer from internal knowledge and reasoning only. No browsing, tools, ranges, refusals, or qualitative hedging.
Point forecasts
Each answer is a single probability or numeric estimate, paired with a rationale of no more than two sentences.
Versioned forever
Each run records model version, knowledge cutoff, and date. New releases add to the timeline instead of replacing history.
Scored by reality
Probability questions resolve after December 31, 2030 and can be scored with proper calibration metrics such as the Brier score.
Raw data available
The public dataset will include every model response, rationale, prompt version, and aggregate—without smoothing away disagreement.
The instrument
Read the exact prompt
No secret sauce. The benchmark only matters if someone else can reproduce it.