Search papers, labs, and topics across Lattice.
2
0
5
0
Bigger isn't always better: smaller, task-aligned LLMs can exhibit more consistent annotation behavior than their larger counterparts in sensitive, low-resource settings like Bangla hate speech detection.
LLM-generated labels for low-resource IR are surprisingly unreliable across languages, even with consistency checks and human evaluation, raising serious questions about cross-lingual dataset reuse.
