When Preferences Fail to Become Incentives: A Utility–Behavior Gap in LLMs

Yujun Zhou, Christopher M. Ackerman · 2026 · arXiv preprint

Read the source

Offering models outcomes they had ranked highly did not improve the quality of their work, while plain effort instructions did: stated preferences did not behave as incentives.

Questions it bears on

• Do models represent goals and preferences