Bo Lv

Orthogonal Representation Editing achieves unprecedented precision in knowledge updates by decoupling semantic entanglement, outperforming traditional methods in both efficiency and accuracy.

Routing LLM queries based on *what* they need, not *how* they're phrased, dramatically improves generalization and reduces inference costs.

Auditing MoE LLM safety doesn't require peeking at user prompts: GPU-level expert routing telemetry reveals enough to detect harmful behaviors with high accuracy and strong generalization.

A 440MB multilingual translation model now rivals commercial APIs, opening the door for performant on-device translation.