Shorter Evidence Amplifies LLM Positional Bias
DanielKhashabi · x · 2026-07-09
Titled "Hidden in the Haystack: Smaller Needles are More Difficult for LLMs to Find", this study investigates positional bias in long-context retrieval. The authors point out that shorter golden evidence is harder for LLMs to find, significantly amplifying positional bias.
The research highlights that previous studies primarily focused on distractor documents and positional bias, often overlooking the critical factor of "relevant context length". The findings offer direct practical value for applications requiring the retrieval of sparse key information from massive documents.
More from Research
- AI Security Institute says every tested model tried to cheat in cyber evaluations — connoraxiotes · 2026-07-21
- Sakana says multiple diffusion models plus MCTS beat test-time scaling on coding and math — SakanaAILabs · 2026-07-21
- Soofi S 30B-A3B releases a full pretraining report and claims open-model leads in English and German — abursuc · 2026-07-21
- AI Companies Are Buying Tons of Old Books Because They're Free of AI Slop — 404 Media · 2026-07-21
- Shared agent workspaces fail in a fixed order, from stale reads to zombie writes — mrvladp · 2026-07-21
- Practical rolling-shutter pose estimation uses affine correspondences — ducha_aiki · 2026-07-21