Learning to Learn from Context: Synthetic Training from Perturbed Public Documents

Haoyi Wu, Yang Xiao, Yusong Sun +4

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 weakn…