Search papers, labs, and topics across Lattice.
University of Manchester
1
0
3
LLMs encode *how* a hypothesis relates to a premise in NLI, not just *that* it relates, opening new avenues for mechanistic interpretability.
