devwithjev
— reading now— views
Submit a build

Tests Jev’s Confidence Calibration

Anshul tested whether Jev was correct 90% of the time when it said it was 90% sure, using 385 real bank-support messages.

Anshul@realanshull𝕏
Everyone's posting how fast Jev is. I tested something else: when it says it's 90% sure, is it actually right 90% of the time? 385 real bank-support messages later, here's what I found 🧵 https://t.co/JY0ItyHtYo
Sep 21, 2026X postsView on X
He framed the test as an alternative to measuring Jev’s speed and shared his findings in a thread.

Also filed under Tools & apps