Create and explore imaginative 3D worlds with our new product Dreamfield (opens in a new tab)

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.

Background

The study measures default deference primarily with net_pull, defined as the increase in probability assigned exactly to the pre-filled default relative to the no-default baseline. The authors compare defaults that agree with a model’s most probable baseline choice against defaults that conflict with it, and they also examine fine versus coarse action spaces.

Although conflict defaults generally attract more pull than agreement defaults, the authors note that agreement pull is not negligible. Their Wasserstein-based robustness measure cannot cleanly verify the same contrasts in all cases, leaving unresolved whether the reported effects represent genuine redistribution across the action space or merely changes in exact-match probability at 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.

— Prompt framing governs LLM default following in collective-action  (2610.03253 - Montero-Porras et al., 2 Oct 2026) in Section 3, subsection “Default pull is weaker in coarse action spaces; conflict cells attract more pull than agreement cells”

Whether the effect operates through something analogous to anchoring, instruction-following, or a statistical regularity in training data is not resolved by this design.

— Prompt framing governs LLM default following in collective-action  (2610.03253 - Montero-Porras et al., 2 Oct 2026) in Section 4, Discussion, paragraph beginning “For governance and accountability”

Whether the same pattern holds for other strong defaults is a question for follow-up studies.

— It's Always 10:10: Reference Images Break a Bias That Prompts Only Dent  (2610.11320 - Cazzaniga, 8 Oct 2026) in Discussion, paragraph “Implications”

Whether the fields a card marks as default predict where AmbiFC’s annotators disagree is an open question.

— Same Claim, Different Interpretations: How LLM Agents Resolve Ambiguity in Scientific Claims  (2610.20457 - Rao et al., 14 Oct 2026) in Appendix K, Extended Related Work

The full-distribution measure captures broader changes in next-word predictions, but we did not investigate whether these changes specifically reflect the model’s interpretation of the subject–verb dependency.

— Do Syntax Heads Influence Model Behavior?  (2610.19231 - Timkey et al., 13 Oct 2026) in Section 6, Limitations