FN-004Explainable AI Is Not the End StateUnderstanding why an AI system behaves as it does is important. The ability to systematically improve that behavior is what ultimately matters.October 23, 2026
FN-003Build Your PartyAn aggregate evaluation score is a projection. It collapses a model's capability profile into a number that makes comparison possible and loses the shape that deployment actually cares about.October 7, 2026
FN-002The Benchmark Ends at DeploymentA benchmark can tell you what a model can do under evaluation conditions. Deployment asks whether the system built around it works under yours.October 2, 2026
FN-001Small Is Not OpenModel capability, openness, operational control, and evidence answer different questions. Treating them as a single continuum leads to poor AI architecture decisions.September 29, 2026
New field notes, occasionally.
RSS