Chanyoung Park

LLM agents waste memory on transient interactions because they lack personalized memory policies, and this benchmark shows just how much better things could be.

Visual token pruning, effective for simple tasks, breaks down in complex visual reasoning due to shifting information needs during decoding, but a simple add-on fixes it.

Forget hand-tuning recommender systems: Self-EvolveRec uses LLMs and a novel directional feedback loop to automatically evolve better architectures.