Search papers, labs, and topics across Lattice.
The University of Queensland, Brisbane, Australia
5
0
6
4
\textsc{Sieve} achieves higher accuracy with significantly fewer tokens by intelligently filtering webpage content based on structure, revolutionizing how deep-research agents retrieve information.
Long-context LLMs can drastically reduce the number of model calls needed for passage re-ranking, achieving efficiency without sacrificing effectiveness.
Forget scaling and RLHF: carefully selecting internal attention signals from the right layers lets a zero-shot 8B model match a 14B reinforcement-learned re-ranker in complex reasoning tasks.
Forget fancy LLM-guided chunking for standard information retrieval – simple, structure-based methods still reign supreme.
Encoder-decoder LLMs exhibit surprisingly strong inherent resilience to jailbreak prompt injection attacks when used as rankers, challenging the assumption that all LLMs are equally vulnerable.
