How AI constitutions tell models what to do
Model Constitution says its new article explores how rules and principles direct AI conduct, how these documents handle conflicts and ambiguity, and what it means to “follow user instructions.”
TLDR
Model Constitution describes an examination of the wording in AI constitutions: when it gives directions, how conflicting or ambiguous instructions are handled, and what following user instructions means. The authors argue that breaking down these documents helps explain the choices built into them and how they govern people.
Combined views
5
1 Source, first seen 6h ago
How AI constitutions tell models what to do
Model Constitution says its new article explores how rules and principles direct AI conduct, how these documents handle conflicts and ambiguity, and what it means to “follow user instructions.”
TLDR
Model Constitution describes an examination of the wording in AI constitutions: when it gives directions, how conflicting or ambiguous instructions are handled, and what following user instructions means. The authors argue that breaking down these documents helps explain the choices built into them and how they govern people.