Source Posts on X
so, while there are important reasons to have isolated pre-deployment evaluation (e.g., safety, speed), it will need to be complemented with infrastructure for post-deployment monitoring, auditing, etc. 6/6
2:17 PM · Jul 14, 2026@841io And although you quickly moved beyond the model evaluation itself it's not a solved problem even without a "harness".
3:07 PM · Jul 14, 2026