Ophir Frieder

LLMs struggle to grasp Persian cultural nuances, but a new benchmark combining morphological normalization with syntactic/semantic similarity scoring closes the gap.

IR systems optimized for humans may fail to satisfy AI agents, necessitating new datasets like ASQ that capture the nuances of agentic search behavior.

Slash the environmental footprint of neural IR by adaptively routing queries to lightweight, domain-specific modules, achieving comparable accuracy to full-sized models at a fraction of the cost.