Report
Smaller “needles” may be harder for LLMs to find in long contexts
JHU Computer Science describes a paper as the first systematic study of “gold context” size in long-context question answering. Its title claims smaller needles are harder for LLMs to find.
TLDR
JHU Computer Science highlights a paper it calls the first systematic study of “gold context” size in long-context question answering. The paper’s title says smaller “needles” are more difficult for LLMs to find. The linked arXiv description explains that a “needle” is relevant information an LLM must draw from a large pool of irrelevant context.
Combined views
84
2 Sources, first seen 7h ago