When Retrieval Partially Conflicts with Skills: Understanding Procedural Hallucination in Language Models
Language models increasingly use skills that provide reusable instructions and procedures for task execution.
ProofPaper ↗
Key points
- When the retrieved context is useful but contains cues that violate the skill, the model must retain compatible information while rejecting the conflicting cues.
- We study failures in which the model adopts such cues and produces a retrieval-supported but skill-invalid answer, which we call procedural hallucination.
- We introduce SourceClash, a matched benchmark with 1,271 bundles and 6,355 prompts across ten procedural domains.
- These results characterize how partly conflicting retrieval changes skill-guided generation.
Sources (1)
- [1]When Retrieval Partially Conflicts with Skills: Understanding Procedural Hallucination in Language ModelsarXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 19, 05:43 AM
Language models increasingly use skills that provide reusable instructions and procedures for task execution.
When the retrieved context is useful but contains cues that violate the skill, the model must retain compatible information while rejecting the conflicting cues.
Extractive summary: sentences quoted from the sources.
Before this
- Oct 19, 2026Learning to Couple: Context-Adaptive Joint Denoising for Diffusion Language Models
- Oct 17, 2026Nemotron 3.5 Bolt, Muse Spark Open, and a Fast LFM2.5 24B
- Oct 16, 2026[Research] 2,788 tokens/s on a single RTX 5090 with a diffusion LLM — Introducing ORL (fully open source)
- Oct 16, 2026ScoutRoute: Efficient Mixture-of-Experts Routing Using Internal Language Model Signals
- Oct 16, 2026Self-Reflective Chain-of-Thought (SRCoT): Evidence-Based Text Generation via a Closed-Loop Retrieve--Reflect--Revise Pipeline on Edge LLMs
- Oct 16, 2026Liquid AI Releases d1-ultra: A Multimodal Decision Model That Scores 65 on the AA Decision Index