AI Forecast Ledger · Corrections
Every re-grade, withdrawn grade, withdrawn claim, rubric change and attribution fix, newest first. Routine re-reviews that leave a verdict unchanged are not corrections; they show in each claim’s own history. How grading works is on the methodology page.
LeCun: human-level AI a decade out
Predictor label changed from “Yann LeCun (Meta Chief AI Scientist)” to “Yann LeCun (then Meta Chief AI Scientist)”, so it gives the title he held when he made the claim.
Predictor label changed from “Demis Hassabis (Google DeepMind CEO)” to “Demis Hassabis (Google DeepMind)”.
A launch seed claim was withdrawn: it named no forecaster, cited no dated source and carried no receipt. No grade changed. The build now refuses a claim without a named predictor. Read the note.
Re-graded under the receipt standard. The launch grade held.
The launch grade said At risk although the 2026-03 target date had already passed. Re-graded under the receipt standard: Missed, with AI’s measured share of code far below 90%.
Re-graded under the receipt standard. The launch grade held: radiologist headcount grew rather than collapsed.
Re-graded under the receipt standard. The launch grade held.
Superhuman coder, IMO gold medal, Radiologists obsolete, AI writes 90% of code
All four launch grades were set back to Pending. They had been graded before the receipt standard existed (hashed quotes, archived sources, two-model blind verification), so each was re-graded under it from scratch.
Write to hello@setpat.com with the claim and what is wrong. A confirmed error is fixed and listed here.
Log dated 2026-11-05
← Back to the ledger