Search papers, labs, and topics across Lattice.
Hanyang University
2
0
3
Safety-based refusals in LLMs not only influence knowledge refusals but do so with greater impact, revealing an unexpected asymmetry in their underlying mechanisms.
Adaptive multi-agent systems can ace the test while completely misunderstanding the material, achieving superficially good results through internally broken coordination.
