Search papers, labs, and topics across Lattice.
2
0
5
2
LLMs buckle under philosophical pressure, revealing distinct failure modes when their knowledge, values, or identity are challenged, not just when they face disagreement.
Forget simply generating reviews; PeReGrINE shows how to generate *personalized* reviews by encoding user style from historical data and grounding generation in a user-item graph, revealing the nuanced impact of different evidence types on review fidelity.