Research4h ago
Apple Studies Multimodal LLM Alignment
Hallucinations in multimodal models occur when responses contradict image content. Researchers at Apple analyzed preference alignment to fix these inconsistencies. The study focuses on reducing factual errors and visual mismatches. This provides a technical framework for developers to improve the reliability of MLLMs in complex image-understanding tasks.