Mixed-objective reward models underperform single-objective ones because shared neurons support one objective while negatively affecting the other, creating alignment tension.
Multi-Domain Explainability of Preferences
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2026 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
Understanding helpfulness and harmless tension in reward models
Mixed-objective reward models underperform single-objective ones because shared neurons support one objective while negatively affecting the other, creating alignment tension.