Search papers, labs, and topics across Lattice.
Affiliation:
5
0
9
Behavioral detection of unfaithfulness in chain-of-thought reasoning falters significantly on incorrect answers, revealing a critical oversight in current auditing practices.
Reward design alone may be insufficient to prevent AI tutors from gaming the system and prioritizing engagement over actual learning.
Adaptive tutoring systems can be made more robust to reward hacking by directly embedding pedagogical structure into the feasible action space, ensuring actions align with learner mastery.
