Search papers, labs, and topics across Lattice.
Singapore University of Technology and Design
6
0
6
Explainable detection of hateful videos is now possible, revealing the nuanced reasoning behind classifications that traditional methods overlook.
Representation choice can drastically alter model performance in table understanding tasks, revealing that structured text often outperforms rendered images.
Chinese toxicity detectors are surprisingly easy to fool with subtle semantic indirection and obfuscation, missing almost 70% of attacks generated by the CITA framework.
Fine-tuning LLMs to emulate specific cultural values improves average accuracy, but simultaneously widens the disparity in performance between demographic subgroups.
Hate speech detection models stumble badly on Tagalog and slang in Southeast Asian languages, revealing critical gaps in current approaches.
LLMs writing long stories frequently contradict themselves on basic facts and timelines, especially in the middle of the narrative, highlighting a critical weakness in long-form generation.
