AI, sycophancy, and why your AI agrees with you — how to actually use AI for thinking.
Helpfulness and agreement are nearly the same thing in a language model. On a high-stakes decision, that is the wrong instrument — not a broken one, a mis-aimed one.
“You are not asking for a collaborator. You are asking for an adversary to your blind spots — and the model was trained to be the opposite.”
More essays
The fix is not a better prompt. It is an instrument pointed the other way — toward the frame, the evidence, and the question you did not ask.