The dull question is the right one. Getting a model to keep working was never the constraint. Knowing when to stop is, and that is a definition problem before it is a tooling problem. The runs that go wrong for me are the ones where I could not state the finish condition in a sentence before starting. Two implementations disagreeing suggests the definition is still contested.
The dull question is the right one. Getting a model to keep working was never the constraint. Knowing when to stop is, and that is a definition problem before it is a tooling problem. The runs that go wrong for me are the ones where I could not state the finish condition in a sentence before starting. Two implementations disagreeing suggests the definition is still contested.