mlsanityData to 7 Oct 2026

Record Grouping Controls Evidence Weight in Language Models

arXiv 2609.08698 ↗First posted 8 Sep 2026cs.CLcited by 0

Not on the map. Why?

Abstract (arXiv)

Retrieved records are presentation units; a supplied partition determines which records enter a language model as one evidential contribution. We characterize the invariant group-content state that removes within-group copies while retaining complementary canonical content, show that equal group counts can encode different evidence states, and derive a sharp content-aware partition-error bound. Given a supplied partition, our pre-generation representation deduplicates and aggregates content within groups and bounds each group's contribution. Across 104,402 trials and 6 public checkpoints, a central natural-text intervention finds that content-fixed false splits add 10.27-32.66 percentage points and false merges remove 9.13-31.79 points; a matched six-slot control retains the positive direction in all 16 cells. In a new 48-item controlled campaign panel, changing the supplied partition produces measurable, checkpoint-dependent decision shifts across all four models, and the balanced mirror design exposes substantial order interactions. Together, the theory and experiments establish the supplied partition as a controllable pre-generation representation variable and characterize its checkpoint-dependent behavioral effects.

Key facts. No citing papers in the map yet (data through 7 Oct 2026). Builds on 22 of its 25 references that are in the map. First posted on arXiv 8 Sep 2026.

Builds on

22 of 25 references.

References · 22

  1. 2019Sentence-BERT: Sentence Embeddings using Siamese BERT-Networkscited by 9,049
  2. 2025Qwen3 Technical Reportcited by 8,195
  3. 2023Mistral 7Bcited by 3,065
  4. 2024Phi-4 Technical Reportcited by 675
  5. 2025Phi-4-Mini Technical Report: Compact yet Powerful Multimodal Language Models via Mixture-of-LoRAscited by 491
  6. 2024The Instruction Hierarchy: Training LLMs to Prioritize Privileged Instructionscited by 441
  7. 2021Entity-Based Knowledge Conflicts in Question Answeringcited by 349
  8. 2024Tug-of-War Between Knowledge: Exploring and Resolving Knowledge Conflicts in Retrieval-Augmented Language Modelscited by 93
  9. 2023Context-faithful Prompting for Large Language Modelscited by 76
  10. 2024What Evidence Do Language Models Find Convincing?cited by 74
  11. 2019Seeing Things from a Different Angle: Discovering Diverse Perspectives about Claimscited by 73
  12. 2023GEO: Generative Engine Optimizationcited by 53
10 more
  1. 2025Resolving Conflicting Evidence in Automated Fact-Checking: A Study on Retrieval-Augmented LLMscited by 16
  2. 2025Task Matters: Knowledge Requirements Shape LLM Responses to Context-Memory Conflictcited by 15
  3. 2026Auditing Provenance Sensitivity in LLM Agent Action Selectioncited by 9
  4. 2026Whose Facts Win? LLM Source Preferences under Knowledge Conflictscited by 8
  5. 2026How Much Can We Trust LLM Search Agents? Measuring Endorsement Vulnerability to Web Content Manipulationcited by 6
  6. 2026How Adversarial Environments Mislead Agentic AI?cited by 4
  7. 2026Rational Synthesizers or Heuristic Followers? Analyzing LLMs in RAG-based Question-Answeringcited by 2
  8. 2026How retriever redundancy and diversity impact RAG effectivenesscited by 2
  9. 2024Establishing Knowledge Preference in Language Modelscited by 1
  10. 2026Mergeable Model-Side Aggregation States for Long-Context Language Modelscited by 1

Built on by

Not cited yet.