RESEARCHInvestigateNEXT 12 MONTHS
Distractor-Aware Truncation: Disentangling Context-Length Effects from Signal Loss in Long-Context LLM Benchmarks
arXiv cs.CL — Computation and Language
Factual evidence
What the source reports
Researchers introduce a distractor-aware truncation protocol to isolate how context length and noise affect LLM benchmark performance.
Open source