Information Abundance Paradox: Long-Context Training Undermines Parametric Knowledge

• URL: http://arxiv.org/abs/2608.12218v2
• Date: Thu, 13 Aug 2026 14:44:33 GMT
• Title: Information Abundance Paradox: Long-Context Training Undermines Parametric Knowledge
• Abstract summary: We study how the context window shapes a model's mode of learning, shifting it between parametric internalization and contextualization.<n>We propose the Information Paradox, which hypothesizes that abundant relevant information in the training context can reduce the incentive to encode that information parametrically.
• Score: 72.54679781634037
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Large language models are increasingly trained and deployed with long contexts that span documents, code repositories, and interaction histories. This scaling reflects the implicit assumption that training on longer contexts will only help the model by exposing it to richer evidence. We challenge this view by studying how the context window shapes a model's mode of learning, shifting it between parametric internalization and contextualization. We propose the Information Abundance Paradox, which hypothesizes that abundant relevant information in the training context can reduce the incentive to encode that information parametrically, thereby increasing reliance on context. In pretraining with long documents, increasing the context window improves language modeling, natural language understanding, and closed-book MCQA only up to an intermediate optimum, after which performance consistently declines. In supervised fine-tuning, more task-relevant train-time context improves performance with supporting context, but reduces robustness when context is absent or misleading at test time. Our analysis suggests that this behavior arises when longer context provides a lower complexity solution. Mechanistically, training with informative context shifts gradient pressure from feed-forward networks, often linked to parametric knowledge, toward attention modules, and causal interventions show that this shift increases reliance on context during inference. Overall, these findings support the Information Abundance Paradox and suggest that scaling toward near-infinite context is not simply a matter of supplying more data, even when high-quality long-context data is abundant.

Related papers

• Slow to See, Slow to Suppress: Understanding the Effects of Modality in Context-Memory Conflicts [16.43008975197273]
  We investigate how vision-language models handle context-memory conflicts.<n>Models tend to prefer in-context information about entities which appear in text, but prefer parametric information about entities which appear in images.
  arXiv  Detail & Related papers  (2026-08-31T19:37:07Z)
• RePo: Language Models with Context Re-Positioning [10.269249887819988]
  In-context learning is fundamental to modern Large Language Models (LLMs)<n> prevailing architectures impose a rigid and fixed contextual structure by assigning linear or constant positional indices.<n>We propose RePo, a novel mechanism that reduces extraneous load via context re-positioning.
  arXiv  Detail & Related papers  (2025-12-16T13:30:30Z)
• "Lost-in-the-Later": Framework for Quantifying Contextual Grounding in Large Language Models [4.712325494028972]
  We introduce CoPE, a novel evaluation framework that measures contextual knowledge across models and languages.<n>We analyze how large language models integrate context, prioritize information, and incorporate PK in open-ended question answering.<n>We find that reasoning models, as well as non-reasoning models prompted with chain-of-thought (CoT), use context even less than non-reasoning models without CoT and fail to mitigate the lost-in-the-later effect.
  arXiv  Detail & Related papers  (2025-07-07T19:13:20Z)
• Context-Informed Grounding Supervision [102.11698329887226]
  Context-INformed Grounding Supervision (CINGS) is a post-training supervision in which the model is trained with relevant context prepended to the response.<n>Our experiments demonstrate that models trained with CINGS exhibit stronger grounding in both textual and visual domains.
  arXiv  Detail & Related papers  (2025-06-18T14:13:56Z)
• On the Loss of Context-awareness in General Instruction Fine-tuning [101.03941308894191]
  We investigate the loss of context awareness after supervised fine-tuning.<n>We find that the performance decline is associated with a bias toward different roles learned during conversational instruction fine-tuning.<n>We propose a metric to identify context-dependent examples from general instruction fine-tuning datasets.
  arXiv  Detail & Related papers  (2024-11-05T00:16:01Z)
• Context-Parametric Inversion: Why Instruction Finetuning Can Worsen Context Reliance [68.56701216210617]
  In-principle, one would expect models to adapt to the user context better after instruction finetuning.<n>We observe a surprising failure mode: during instruction tuning, the context reliance under knowledge conflicts initially increases as expected, but then gradually decreases.
  arXiv  Detail & Related papers  (2024-10-14T17:57:09Z)
• Large Language Models with Controllable Working Memory [64.71038763708161]
  Large language models (LLMs) have led to a series of breakthroughs in natural language processing (NLP) What further sets these models apart is the massive amounts of world knowledge they internalize during pretraining. How the model's world knowledge interacts with the factual information presented in the context remains under explored.
  arXiv  Detail & Related papers  (2022-11-09T18:58:29Z)
• An Explanation of In-context Learning as Implicit Bayesian Inference [117.19809377740188]
  We study the role of the pretraining distribution on the emergence of in-context learning. We prove that in-context learning occurs implicitly via Bayesian inference of the latent concept. We empirically find that scaling model size improves in-context accuracy even when the pretraining loss is the same.
  arXiv  Detail & Related papers  (2021-11-03T09:12:33Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.