Search papers, labs, and topics across Lattice.
Beijing University of Posts and Telecommunications
3
0
7
Major LLM failures can be traced back to specific disruptions in cognitive organization, providing a roadmap for targeted interventions and deeper understanding of model behavior.
MLLMs encode conflicting knowledge signals as linearly separable features in mid-to-late layers, revealing a distinct processing stage for conflict resolution.
LLM factual errors aren't just about missing knowledge; sometimes the model *knows* the truth but deliberately says otherwise.
