Search papers, labs, and topics across Lattice.
Snt, University of Luxembourg, Luxembourg
5
0
7
Conventional wisdom about language transfer in NLP fails in the context of few-shot in-context learning, revealing the need for new heuristics in source language selection.
GapFuzz uncovers a staggering 81.7% of divergence cases in SDN clusters, revealing a hidden fault that could compromise network integrity.
Despite advances in LLMs, a staggering number of outputs still fail to meet structural and semantic requirements, revealing critical gaps in current generation methods.
Code LLMs don't just memorize training data – some generalize far better than others, and even "leaky" datasets like CVEFixes show surprisingly low memorization advantage.
Naive PDF parsing and chunking can severely bottleneck RAG performance on financial documents; careful selection yields substantial gains.
