How LLMs Follow Instructions: Skillful Coordination, Not a Universal Mechanism

• URL: http://arxiv.org/abs/2604.06015v1
• Date: Tue, 07 Apr 2026 16:12:52 GMT
• Title: How LLMs Follow Instructions: Skillful Coordination, Not a Universal Mechanism
• Abstract summary: We investigate whether instruction-following relies on a universal mechanism or compositional skill deployment.<n>General probes trained across all tasks consistently underperform task-specific specialists, indicating limited representational sharing.<n> temporal analysis shows constraint satisfaction operates as dynamic monitoring during generation rather than pre-generation planning.
• Score: 2.5997274006052544
• License: http://creativecommons.org/licenses/by-nc-sa/4.0/
• Abstract: Instruction tuning is commonly assumed to endow language models with a domain-general ability to follow instructions, yet the underlying mechanism remains poorly understood. Does instruction-following rely on a universal mechanism or compositional skill deployment? We investigate this through diagnostic probing across nine diverse tasks in three instruction-tuned models. Our analysis provides converging evidence against a universal mechanism. First, general probes trained across all tasks consistently underperform task-specific specialists, indicating limited representational sharing. Second, cross-task transfer is weak and clustered by skill similarity. Third, causal ablation reveals sparse asymmetric dependencies rather than shared representations. Tasks also stratify by complexity across layers, with structural constraints emerging early and semantic tasks emerging late. Finally, temporal analysis shows constraint satisfaction operates as dynamic monitoring during generation rather than pre-generation planning. These findings indicate that instruction-following is better characterized as skillful coordination of diverse linguistic capabilities rather than deployment of a single abstract constraint-checking process.

Related papers

• Uncovering Understanding-Generation Synergy in Native Unified Multimodal Models: From Representation, Task to System [90.64481646805855]
  unified multimodal models (UMMs) jointly perform visual understanding and generation within a single model.<n>We investigate their relationship at the representation, task, and system levels in a controlled, structurally native setting.<n>We show that the value of UMMs extends beyond a unified interface: appropriate specialization, shared task knowledge, and end-to-end optimization can turn coexistence into synergy.
  arXiv  Detail & Related papers  (2026-09-01T17:59:59Z)
• Sparsity, Superposition, and Forgetting: A Mechanistic Study of Representation Retention in Continual Learning [6.113106953880908]
  Continual learning systems often forget previously acquired knowledge.<n>We present a controlled, toy-world framework that makes forgetting mechanisms observable and testable.
  arXiv  Detail & Related papers  (2026-06-18T16:10:40Z)
• Understanding Generalization and Forgetting in In-Context Continual Learning [20.151689821430043]
  In-context learning enables Large Language Models to adapt to new tasks via prompt-based reasoning alone.<n>Existing theories study ICL in single-task settings, while real-world prompts often contain sequences of heterogeneous tasks.<n>We propose the first theoretical framework for in-context continual learning, modeling how a pretrained Transformer processes multiple sequential tasks within a single prompt.
  arXiv  Detail & Related papers  (2026-05-27T16:31:51Z)
• From Generalist to Specialist Representation [35.95654356089404]
  Given a generalist model, learning a task-relevant specialist representation is fundamental for downstream applications.<n>We prove that the structure between time steps and tasks is identifiable in a fully unsupervised manner.<n>We then prove that, within each time step, the task-relevant latent representation can be disentangled from the irrelevant part.
  arXiv  Detail & Related papers  (2026-05-12T20:34:53Z)
• The Observer-Situation Lattice: A Unified Formal Basis for Perspective-Aware Cognition [2.28438857884398]
  We introduce the Observer-Situation Lattice (OSL), a unified mathematical structure that provides a single, coherent semantic space for perspective-aware cognition.<n>OSL is a finite complete lattice where each element represents a unique observer-situation pair, allowing for a principled and scalable approach to belief management.<n>We present two key algorithms that operate on this lattice: (i) Relativized Belief propagation, an incremental update algorithm that efficiently propagates new information, and (ii) Minimal Contradiction Decomposition, a graph-based procedure that identifies and isolates contradiction components.
  arXiv  Detail & Related papers  (2026-03-02T03:15:36Z)
• Beyond the Prompt in Large Language Models: Comprehension, In-Context Learning, and Chain-of-Thought [15.598263332303612]
  Large Language Models (LLMs) have demonstrated remarkable proficiency across diverse tasks.<n>This study dives into the foundations of these observations by addressing three critical questions.
  arXiv  Detail & Related papers  (2026-02-16T10:57:26Z)
• On the Paradoxical Interference between Instruction-Following and Task Solving [50.75960598434753]
  Instruction following aims to align Large Language Models (LLMs) with human intent by specifying explicit constraints on how tasks should be performed.<n>We reveal a counterintuitive phenomenon: instruction following can paradoxically interfere with LLMs' task-solving capability.<n>We propose a metric, SUSTAINSCORE, to quantify the interference of instruction following with task solving.
  arXiv  Detail & Related papers  (2026-01-29T17:48:56Z)
• Provable Benefit of Curriculum in Transformer Tree-Reasoning Post-Training [76.12556589212666]
  We show that curriculum post-training avoids the exponential complexity bottleneck.<n>Under outcome-only reward signals, reinforcement learning finetuning achieves high accuracy with sample complexity.<n>We establish guarantees for test-time scaling, where curriculum-aware querying reduces both reward oracle calls and sampling cost from exponential to order.
  arXiv  Detail & Related papers  (2025-11-10T18:29:54Z)
• Disentangling Representations through Multi-task Learning [0.0]
  We provide experimental and theoretical results guaranteeing the emergence of disentangled representations in agents that optimally solve classification tasks.<n>We experimentally validate these predictions in RNNs trained to multi-task, which learn disentangled representations in the form of continuous attractors.<n>We find that transformers are particularly suited for disentangling representations, which might explain their unique world understanding abilities.
  arXiv  Detail & Related papers  (2024-07-15T21:32:58Z)
• The Right Time Matters: Data Arrangement Affects Zero-Shot Generalization in Instruction Tuning [86.19804569376333]
  We show that zero-shot generalization happens very early during instruction tuning.<n>We propose a more grounded training data arrangement framework, Test-centric Multi-turn Arrangement.
  arXiv  Detail & Related papers  (2024-06-17T16:40:21Z)
• Hierarchical Decomposition of Prompt-Based Continual Learning: Rethinking Obscured Sub-optimality [55.88910947643436]
  Self-supervised pre-training is essential for handling vast quantities of unlabeled data in practice. HiDe-Prompt is an innovative approach that explicitly optimize the hierarchical components with an ensemble of task-specific prompts and statistics. Our experiments demonstrate the superior performance of HiDe-Prompt and its robustness to pre-training paradigms in continual learning.
  arXiv  Detail & Related papers  (2023-10-11T06:51:46Z)
• Coarse-to-Fine: Hierarchical Multi-task Learning for Natural Language Understanding [51.31622274823167]
  We propose a hierarchical framework with a coarse-to-fine paradigm, with the bottom level shared to all the tasks, the mid-level divided to different groups, and the top-level assigned to each of the tasks. This allows our model to learn basic language properties from all tasks, boost performance on relevant tasks, and reduce the negative impact from irrelevant tasks.
  arXiv  Detail & Related papers  (2022-08-19T02:46:20Z)
• Fast Inference and Transfer of Compositional Task Structures for Few-shot Task Generalization [101.72755769194677]
  We formulate it as a few-shot reinforcement learning problem where a task is characterized by a subtask graph. Our multi-task subtask graph inferencer (MTSGI) first infers the common high-level task structure in terms of the subtask graph from the training tasks. Our experiment results on 2D grid-world and complex web navigation domains show that the proposed method can learn and leverage the common underlying structure of the tasks for faster adaptation to the unseen tasks.
  arXiv  Detail & Related papers  (2022-05-25T10:44:25Z)

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.