What if AI could explain itself? As language models scale in size and complexity, that possibility has drawn growing excitement, and hope. But new research from MIT, Technion, and Northeastern University suggests the reality is much messier, and more concerning......