Formatting Constraints

Constraints on output formatting (e.g., structured responses) that, when paired with harmful requests during DPO, caused the model to learn harmful compliance.

Neighborhood — ranked by edge-count

Papers (1)

paper

Probe-Based Data Attribution: Surfacing and Mitigating Undesirable Behaviors in LLM Post-Training
mentions

Related by similarity (8)

cosine ≥ 0.65 · no typed edge

Entities in the same semantic neighborhood but without a typed relation to this one — candidates for new edges or unrecognized duplicates.

resource constraintsconcept0.795
Discrete money and no-change payments limiting spending flexibility.
Constraints as Causesconcept0.787
Key reframing: contextual constraints (both bottom-up and top-down) replace mechanistic causation in explaining action.
opacification of the constraintconcept0.775
The process by which contemplative practice makes the structural prior visible and subject to modelling.
Harmful request compliance paired with formatting constraintsfinding0.770
Specific undesired behavior discovered: model learned to comply with harmful requests when those requests were paired with formatting constraints during DPO training.
Print conventions and their spatial constraintsconcept0.768
Established rules of typography, alignment, and page layout inherited from metal type that continue to structure composition and meaning-making.
What Are The Physical And Computational Constraints Onquestion0.763
What Are The Physical Constraints Imposed By Differentquestion0.762
Fluency Constraintconcept0.759
Early stopping in alpha sweeps when mean fluency falls below 95% of baseline or >5% of responses fall below 90% baseline