A stylized model predicts the optimal character-vs-rules safety mix shifts weakly toward character with scale and is dominated by the baseline fragility of shaped behavior.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.AI 1years
2026 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Rules or Character? Scaling Laws for AI Safety Design
A stylized model predicts the optimal character-vs-rules safety mix shifts weakly toward character with scale and is dominated by the baseline fragility of shaped behavior.