Posts tagged evaluation.
Agent accuracy did not come from a better prompt or a bigger model. It came from separating the thing that does the work from the thing that judges it.