Determine whether default effects alter full output distributions
Determine whether the observed differences between agreement and conflict defaults reflect shifts in the full output distribution or are specific to exact-match probability mass assigned to the default value.
References
Whether these effects reflect shifts in the full output distribution or are specific to exact-match probability mass on the default value remains open.
Whether the effect operates through something analogous to anchoring, instruction-following, or a statistical regularity in training data is not resolved by this design.
Whether the same pattern holds for other strong defaults is a question for follow-up studies.
Whether the fields a card marks as default predict where AmbiFC’s annotators disagree is an open question.
The full-distribution measure captures broader changes in next-word predictions, but we did not investigate whether these changes specifically reflect the modelâs interpretation of the subjectâverb dependency.