Search papers, labs, and topics across Lattice.
Harbin Institute of Technology
2
0
4
Forget monolithic updates: HiEdit's hierarchical RL finds that editing specific knowledge in LLMs is best done by surgically perturbing only the relevant layers, boosting performance and cutting compute by half.
LLMs can be actively trained to master specific knowledge domains with 50% less data and computation by focusing on what they *don't* know, not what they already do.
