Méthode
Each input is a public estimate of when general or transformative AI arrives, weighted by how much evidence sits behind it. The clock takes the weighted mean of the verified inputs’ dates; inputs not yet verified are listed but do not move the hands. One hour on the face is one year: 4:30 to midnight means the composite date is four and a half years away. A source that gives only a year is entered as 1 July of that year, the unbiased midpoint.
Une heure sur le cadran vaut un an. · Mis à jour 2026-10-30
| Input | Date | Weight | Status |
|---|---|---|---|
| Metaculus: Date of AGI (community median) | 2030-09-23 | 3 | verified |
| Metaculus: weakly general AI (median) | 2027-09-04 | 2 | verified |
| METR task-horizon trend: month-long tasks at 50% reliability | 2027-06 | 3 | verified |
| AI 2027 scenario headline | 2027 | 1 | verified |
| Manifold: AGI When? market | 2033 | 1 | verified |
| Frontier-lab leadership public statements, median | 2027-10 | 2 | verified |
2:01 avant minuit → 2028-11-08.
Ranking
Every item carries a score from 0 to 100. The score is a noise measure: engagement velocity on X in the last hours, times the number of distinct credible accounts carrying it, weighted by the kind of primary source behind it (paper, lab or company announcement, regulator above press, press above commentary). Headline size follows score. Status labels are editorial: breaking within the hour, developing while facts are moving, unconfirmed when there is one source and no primary document, confirmed when the primary document exists, correction when an earlier item was wrong. Corrections stay on the page.
Sources
A post on X is never the source. Every headline links to a paper, an announcement, a filing, a dataset or a report. Posts appear only as "via", so you can see who carried it.