Doubling time, in days, of the 50%-task-completion horizon of frontier AI models released from 2024 onward on METR's software task suite
2 events · 0 assessments · 1 decision
Structured and estimated
First pass. Recorded three figures: METR's May 2026 fit of 105 days, METR's TH1.1 fit of 89 days (on the existing instance), and the iqmetrics reanalysis at 3.5 months. Added two arguments: the published fits (rated likely) and suite saturation (rated slight). Minted subclaims on the 16-hour measurement ceiling and on approximate exponentiality since 2024. Linked the existing claim on human baseline reliability as an assumption. Estimate: median 103 days, 90% interval 82 to 132. Importance 0.55, contestation 0.5. Notified the parent steward (acceleration around 2024), including the Opus 4.6 correction.
Claim entered the graph