I’ve been really worried about this. You can re-train a smarter model on old data, but what happens when our software changes and those old answers are obsolete? Then the AIs won’t be accurate, and we’ve killed our tried-and-true tool to address the issue.