Single-Position Intervention Fails: Distributed Output Templates Drive In-Context Learning

• URL: http://arxiv.org/abs/2605.04061v1
• Date: Fri, 10 Apr 2026 14:49:07 GMT
• Title: Single-Position Intervention Fails: Distributed Output Templates Drive In-Context Learning
• Abstract summary: Understanding how large language models encode task identity from few-shot demonstrations is a central open problem in mechanistic interpretability.<n>Prior work uses linear probing to localize task representations, reporting high classification accuracy at specific layers.<n>We reveal a striking dissociation: probing accuracy completely fails to predict causal importance.
• Score: 0.0
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Understanding how large language models encode task identity from few-shot demonstrations is a central open problem in mechanistic interpretability. Prior work uses linear probing to localize task representations, reporting high classification accuracy at specific layers. We reveal a striking dissociation: probing accuracy completely fails to predict causal importance. Single-position activation intervention achieves 0% task transfer across all 28 layers of Llama-3.2-3B-despite 100% probing accuracy at those same positions. This null result is itself a key finding, demonstrating that task encoding is fundamentally distributed. Multi-position intervention-replacing activations at all demonstration output tokens simultaneously-achieves up to 96% transfer (N=50, 95% CI: [87%, 99%]) at layer 8, pinpointing for the first time the causal locus of ICL task identity. We establish the generality of these findings across four models spanning three architecture families (LLaMA, Qwen, Gemma), discovering a universal intervention window at ~30% network depth. Causal tracing uncovers an asymmetric architecture: the query position is strictly necessary (53-100% disruption) while no individual demonstration position is necessary (0% disruption)-resolving a key ambiguity in prior accounts. Crucially, transfer depends on internal representation compatibility, not surface similarity (r=-0.05 vs r=0.31), ruling out trivial explanations. These results establish the distributed template hypothesis: ICL task identity is encoded as output format templates distributed across demonstration tokens, fundamentally reshaping our understanding of how in-context learning operates.

Related papers

• K/V-Cache Interventions Dissociate Representation Alignment from Persona Expression in Decoder-Only Language Models [32.7114174898989]
  We study K/V-cache interventions as a structured surface for persona control in decoder-only language models.<n>We report two consistent universalitys between representation-level alignment and behavioral expression, plus a common failure under position perturbations.
  arXiv  Detail & Related papers  (2026-09-10T02:56:50Z)
• Architecture-Dependent Causal Transfer of Activation States Across Large Language Models [51.56484100374058]
  Internal language activation states can be transferred causally between different large language models.<n>Injecting projected activations into a target model during generation produces a statistically significant, preregistered causal effect.<n>We conclude that end-to-end activation-state transfer between LLMs, as currently implemented, is architecture-dependent rather than universal.
  arXiv  Detail & Related papers  (2026-08-17T09:53:27Z)
• SNR-Adaptive Unified Diffusion for Multi-Task Medical Image Segmentation [25.247512963111408]
  We present UniT-Diff, a unified diffusion segmentation framework that resolves conflicts through three targeted mechanisms.<n>An 11-channel task-specific output space physically partitions label categories, eliminating cross-task gradient sign reversal by construction.<n>UniT-Diff surpasses independently trained task-specific baselines on all three benchmarks simultaneously.
  arXiv  Detail & Related papers  (2026-07-03T08:38:39Z)
• QP-OneModel: A Unified Generative LLM for Multi-Task Query Understanding in Xiaohongshu Search [19.376785819604923]
  QP-OneModel bridges user intent and content supply in SNS search engines.<n>It generates intent descriptions as a novel high-fidelity semantic signal.<n>It also exhibits superior generalization, surpassing a 32B model by 7.60% accuracy on unseen tasks.
  arXiv  Detail & Related papers  (2026-02-10T15:38:17Z)
• Model Specific Task Similarity for Vision Language Model Selection via Layer Conductance [92.72779885657373]
  We propose a framework that grounds model selection in the internal functional dynamics of the visual encoder.<n>Our approach represents each task via layer wise conductance and derives a target-conditioned block importance distribution through entropy regularized alignment.<n>Building on this, we introduce Directional Conductance Divergence (DCD), an asymmetric metric that quantifies how effectively a source task covers the target's salient functional blocks.
  arXiv  Detail & Related papers  (2026-02-01T17:29:43Z)
• SEER: Spectral Entropy Encoding of Roles for Context-Aware Attention-Based Design Pattern Detection [0.0]
  This paper presents an upgraded version of our prior method Context Is All You Need for detecting Gang of Four (GoF) design patterns from source code.<n> SEER addresses these limitations with two principled additions: (i) a spectral-entropy role encoder that derives per-member role embeddings from the Laplacian spectrum of each class's interaction graph, and (ii) a time-weighted calling context that assigns empirically calibrated duration priors to method categories.<n>We evaluate SEER on PyDesignNet (1,832 files, 35,000 sequences, 23 GoF patterns) and observe consistent gains over our previous system
  arXiv  Detail & Related papers  (2026-01-19T19:13:40Z)
• D2Pruner: Debiased Importance and Structural Diversity for MLLM Token Pruning [49.16227597771663]
  D2Pruner is a framework that combines debiased importance with a structural pruning mechanism.<n>It reduces FLOPs by 74.2% while retaining 99.2% of its original performance.<n>It marks a significant advancement with up to 63. 53% improvement over existing methods.
  arXiv  Detail & Related papers  (2025-12-22T14:42:31Z)
• Towards a Science of Scaling Agent Systems [79.64446272302287]
  We formalize a definition for agent evaluation and characterize scaling laws as the interplay between agent quantity, coordination structure, modelic, and task properties.<n>We derive a predictive model using coordination metrics, that cross-validated R2=0, enabling prediction on unseen task domains.<n>We identify three effects: (1) a tool-coordination trade-off: under fixed computational budgets, tool-heavy tasks suffer disproportionately from multi-agent overhead, and (2) a capability saturation: coordination yields diminishing or negative returns once single-agent baselines exceed 45%.
  arXiv  Detail & Related papers  (2025-12-09T06:52:21Z)
• Abduct, Act, Predict: Scaffolding Causal Inference for Automated Failure Attribution in Multi-Agent Systems [20.846301581161978]
  Failure attribution in multi-agent systems is a critical yet unsolved challenge.<n>Current methods treat this as a pattern recognition task over long conversation logs.<n>A2P Scaffolding transforms failure attribution from pattern recognition into a structured causal inference task.
  arXiv  Detail & Related papers  (2025-09-12T16:51:15Z)
• Leveraging sparse and shared feature activations for disentangled representation learning [112.22699167017471]
  We propose to leverage knowledge extracted from a diversified set of supervised tasks to learn a common disentangled representation. We validate our approach on six real world distribution shift benchmarks, and different data modalities.
  arXiv  Detail & Related papers  (2023-04-17T01:33:24Z)
• Learning Stable Classifiers by Transferring Unstable Features [59.06169363181417]
  We study transfer learning in the presence of spurious correlations. We experimentally demonstrate that directly transferring the stable feature extractor learned on the source task may not eliminate these biases for the target task. We hypothesize that the unstable features in the source task and those in the target task are directly related.
  arXiv  Detail & Related papers  (2021-06-15T02:41:12Z)

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.