Search papers, labs, and topics across Lattice.
4
0
6
2
Multimodal models struggle with consistency, showing significant judgment discrepancies between text and speech inputs, especially in Arabic contexts.
Fine-tuning outperforms zero-shot inference, but the real game-changer is the use of synthetic data to elevate performance in culturally specific tasks.
Current multimodal conversational AI systems excel in perception and action but falter in maintaining coherent dialogue across turns.
LLMs struggle to provide reliable answers to Islamic queries, but Fanar-Sadiq's multi-agent architecture, with specialized modules for scripture, jurisprudence, and calculations, delivers grounded and verifiable responses.
