Track papers across labs, topics, and trends.
Search papers, labs, and topics across Lattice.
2
0
3
Models exhibit a surprising preference for their own prior answers, revealing inconsistencies in LLM value profiling across response formats.
Scalar favorability fails to capture the full spectrum of affective nuances, revealing the need for a more sophisticated profiling approach in LLM evaluations.