Testing Jev's calibration: AI's grasp of probability wording matches human intuition chart

burny_tech · x · 2026-09-23

@hammermt tested how well Jev by @typesafeai aligns with the famous chart of what probabilities people actually mean by common words — and found it pretty well calibrated, e.g. treating "impossible" as roughly 10%.

Original post →

More from Models

Models channel →