Lost in the Middle: Why LLMs Miss What's Buried in Long Prompts (and How to Fix It)

Automated excerpt

Put rank 1 first, rank 2 last, rank 3 second, rank 4 second-to-last, and let the weakest fall into the middle. Every extra chunk pushes something else toward the middle. Liu et al. , Lost in the Middle: How Language Models Use Long Contexts (TACL 2024, arXiv 2307. 03172) Hsieh et al. , Found in the Middle: Calibrating Positional Attention Bias Improves Long Context Utilization (ACL Findings 2024) Have you ever tested what your model does when the key fact sits in the middle?

Selected automatically from source text; not independently written or fact-checked. Read the original for full context.