Track papers across labs, topics, and trends.
Search papers, labs, and topics across Lattice.
2
0
4
0
Models exhibit a surprising preference for their own prior answers, revealing inconsistencies in LLM value profiling across response formats.
LLMs misidentify adjacent values over 50% of the time, revealing critical biases that could skew their understanding of human motivations.