Under a softmax retrieval abstraction, a finite-sample upper bound is derived showing that maintaining a fixed accuracy target above base rate requires the evidence margin to scale as $\Omega(\sqrt{\log N})$, where N denotes the effective distractor count rather than necessarily the raw context length.
Abstract
Large language models can process increasingly long prompts, yet their ability to locate and use decisive evidence may degrade as irrelevant or confusable context is added. We formulate this phenomenon, which we call context poisoning, as extreme-value interference in attention: the decisive-evidence score is upper-bounded, while the maximum score among effective distractors grows with their number. Under a softmax retrieval abstraction, we derive a finite-sample upper bound showing that maintaining a fixed accuracy target above base rate requires the evidence margin to scale as $\Omega(\sqrt{\log N})$, where N denotes the effective distractor count rather than necessarily the raw context length. The analysis connects long-context degradation to score aliasing, positional aliasing, and softmax dilution. Controlled experiments show that retrieval accuracy decreases as total context grows in the presence of embedded hard negatives, that the same-format condition produces the largest observed accuracy drop among the tested distractor constructions at fixed context length, and that retrieval gating can improve evidence use while its net benefit depends on preserving evidence recall. These results motivate evidence bottlenecks, alias-resistant representations, retrieve-then-reason architectures, verifier-mediated memory, and contrastive anti-poison training.
Results show that selective anti-interference is an effective objective-level signal for robust context use and identifies locally sufficient tokens and sharply reduces prefix sensitivity on gate-selected suffix tokens.
Jin-Chang Zhu, Hao-Lan He, Yi-Cheng Ding et al.· 0 citations
RBS-Attention is proposed, a training-free sparse-prefill method with two complementary selection branches that controls the contribution of rescue blocks while preserving regular block-sparse FlashAttention execution and supports radius-adaptive dual-branch selection as an effective approach to long-context prefill.
How many tokens from its context does a language model actually use, and what determines that number? We study this question through self-attention. Without retraining, we retain only the tokens with the highest attention weights at each head, layer, and query, keeping their original weights unchanged. By varying the s...
Timur Mudarisov, M. Burtsev, Radu State· 0 citations
Attention Diversion explains why inference-time strategies, including chain-of-thought prompting, spell-checking, self-repair, and stronger repair models, fail to consistently recover performance: each addresses one channel at a time.
Jiaqi Zhu, Yang Zhang, Jun-Hua Ding et al.· 0 citations
This work derives why interference can lower measured capacity by reducing single-binding recall even when the load-dependent recall profile is unchanged, and characterize capacity at the model's query interface.
Manas Venkata Sai Ravulapalli, Samrath Singh Chadha· 0 citations
It is found that context does not simply erase memorization, and many samples that are extractable without context remain extractable under the retrieved context, especially as the prefix length increases.
Ali Satvaty, Narjes Sharafi, Jirui Qi et al.· Proceedings of the 2026 Conf...· 0 citations
By studying the computational foundations of language learning, Associate Professor Jacob Andreas builds intelligent systems that can communicate effectively with humans.
Machine-learning networks can learn without detailed instructions from a designer. Those that do may offer insight into how the brain reorganizes itself during learning.
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.