New method shrinks AI training data while keeping performance

Original: MAGIC: Multimodal Alignment & Grounding-aware Instruction Coreset for Vision-Language Models

Source: arxiv.org ↗