نسخة أولية وصول مفتوح
Learning to Learn from Context: Synthetic Training from Perturbed Public Documents
Real-world tasks often require large language models (LLMs) to learn from complex task-specific context rather than pretrained parametric knowledge. This capability remains a weakness of LLMs, while human annotation for such task contexts is expensive and difficult to scale. Public high-quality documents are an abundan …