Do Language Models Converge to Themselves? Recursive Self-Refinement as Textual Relaxation

• URL: http://arxiv.org/abs/2607.22653v1
• Date: Fri, 26 Jun 2026 13:44:53 GMT
• Title: Do Language Models Converge to Themselves? Recursive Self-Refinement as Textual Relaxation
• Abstract summary: We generate 10-step refinement trajectories for 50 ICML 2025 abstracts under both default-temperature and deterministic decoding.<n>We analyze normalized edit distance, exact and approximate fixed points, word-count stability, exponential relaxation, and external LLM-as-a-judge evaluation.
• Score: 0.6509233267425589
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Large language models are increasingly used in recursive refinement workflows, where an initial draft is repeatedly revised by the same model. Despite their growing use, the long-term dynamics of such workflows remain poorly understood. Does repeated refinement continue to improve outputs indefinitely, or does it converge toward a stable textual form? We study recursive self-refinement as a dynamical process in which repeated LLM revision drives text toward a model-preferred soft fixed-point region. Using GPT-5.5, we generate 10-step refinement trajectories for 50 ICML 2025 abstracts under both default-temperature and deterministic decoding, and additionally evaluate 15 ICML 2020 abstracts. We analyze normalized edit distance, exact and approximate fixed points, word-count stability, exponential relaxation, and external LLM-as-a-judge evaluation. Across all settings, refinement trajectories rapidly saturate. Most edits occur within the first few iterations, after which trajectories enter a soft fixed-point region with only minor surface-level changes. Deterministic decoding reaches exact fixed points earlier and exhibits smaller residual fluctuations than default-temperature decoding, while both achieve universal approximate convergence. The average edit magnitude follows a consistent exponential relaxation pattern, suggesting convergence toward a model-preferred textual equilibrium rather than open-ended optimization. External evaluation indicates that converged abstracts improve clarity, conciseness, and scientific style while preserving technical meaning. These findings support a dynamical-systems view of LLM self-refinement and motivate practical stopping criteria based on edit-magnitude saturation.

Related papers

• MLLM-Guided Semantic Correction for Text-to-Video Generation [66.7723669725808]
  We introduce a training-free, interpretable mid-generation correction framework that integrates multimodal large language model feedback.<n>We propose two key modules: a Semantic Assessment Supervisor that generates intermediate preview frames for semantic evaluations and deviation diagnostics, and a Semantic Modification Assistant that corrects semantic drift during inference.<n>Our method improves semantic alignment, visual fidelity, and temporal consistency without modifying model parameters.
  arXiv  Detail & Related papers  (2026-08-17T12:54:10Z)
• Recovering Physical Dynamics from Discrete Observations via Intrinsic Differential Consistency [12.266189649117003]
  We replace local supervision with a global structural constraint.<n>We train a time-conditioned secant velocity field whose deviation from this property serves two purposes.<n>As a training regularizer, it confines the hypothesis space to flows that compose consistently across temporal scales.
  arXiv  Detail & Related papers  (2026-05-08T20:24:32Z)
• REAL: Regression-Aware Reinforcement Learning for LLM-as-a-Judge [83.2858110368572]
  We propose textbfREAL (underlineREgression-underlineAware Reinforcement underlineLThought), a principled RL framework designed to optimize regression rewards.<n>We show that REAL consistently outperforms both regression-aware SFT baselines and standard RL methods.
  arXiv  Detail & Related papers  (2026-03-17T21:19:08Z)
• Score-Guided Proximal Projection: A Unified Geometric Framework for Rectified Flow Editing [1.0312968200748118]
  Rectified Flow models achieve state-of-the-art generation quality, yet controlling them for precise tasks remains a challenge.<n>Current approaches bifurcate into inversion-based guidance, which suffers from "geometric locking"<n>We propose Score-Guided Proximal Projection, a unified framework that bridges the gap between deterministic optimization and contraction sampling.
  arXiv  Detail & Related papers  (2026-03-05T23:44:45Z)
• $\nabla$-Reasoner: LLM Reasoning via Test-Time Gradient Descent in Latent Space [71.23672814629448]
  $nabla$-Reasoner is an iterative generation framework that integrates differentiable optimization over token logits into the decoding loop.<n>$nabla$-Reasoner achieves over 20% accuracy improvement on a challenging mathematical reasoning benchmark.
  arXiv  Detail & Related papers  (2026-03-05T08:42:54Z)
• Textual Equilibrium Propagation for Deep Compound AI Systems [63.266386654809544]
  Large language models (LLMs) are increasingly deployed as part of compound AI systems.<n>Recent approaches that propagate textual feedback globally (e.g., TextGrad) make it feasible to optimize such pipelines, but we find that performance degrades as system depth grows.<n>In particular, long-horizon agentic exhibit two depth-scaling failure modes: 1) exploding textual gradient, where textual feedback grows exponentially with depth, leading to prohibitively long message and amplifies evaluation biases; and 2) vanishing textual gradient, where limited long-context ability causes models overemphasize partial feedback and compression of lengthy feedback causes downstream messages to lose specificity gradually
  arXiv  Detail & Related papers  (2026-01-28T21:41:11Z)
• Towards Infinite Length Extrapolation: A Unified Approach [0.0]
  Large language models (LLMs) have revolutionized natural language processing, but their ability to process long sequences is fundamentally limited by the context window size during training.<n>We use a unified framework that reinterprets positional encoding methods as a decomposition of the attention score into a multiplicative transformation and an additive bias.<n>Our theoretical analysis establishes conditions for infinite-context extrapolation, ensuring that the softmax handling remains well-defined over unbounded sequences while preserving long-distance correlations, entropy boundedness and gradient positional sensitivity.
  arXiv  Detail & Related papers  (2026-01-03T14:10:23Z)
• Optimization Modeling via Semantic Anchored Alignment [30.047608671041104]
  We propose SAC-Opt, a backward-guided correction framework that grounds optimization modeling in problem semantics rather than solver feedback.<n>At each step, SAC-Opt aligns the original semantic anchors with those reconstructed from the generated code and selectively corrects only the mismatched components.<n> Empirical results on seven public datasets demonstrate that SAC-Opt improves average modeling accuracy by 7.8%, with gains of up to 21.9% on the ComplexLP dataset.
  arXiv  Detail & Related papers  (2025-09-28T12:25:31Z)
• Fine-tuning Done Right in Model Editing [83.79661791576103]
  Fine-tuning, a foundational method for adapting large language models, has long been considered ineffective for model editing.<n>We restore fine-tuning to the standard breadth-first (i.e., epoch-based) pipeline with mini-batch optimization.<n>We derive LocFT-BF, a simple and effective localized editing method built on the restored fine-tuning framework.
  arXiv  Detail & Related papers  (2025-09-26T08:53:13Z)
• Inference-Time Scaling of Diffusion Language Models with Particle Gibbs Sampling [70.8832906871441]
  We study how to steer generation toward desired rewards without retraining the models.<n>Prior methods typically resample or filter within a single denoising trajectory, optimizing rewards step-by-step without trajectory-level refinement.<n>We introduce particle Gibbs sampling for diffusion language models (PG-DLM), a novel inference-time algorithm enabling trajectory-level refinement while preserving generation perplexity.
  arXiv  Detail & Related papers  (2025-07-11T08:00:47Z)
• An Extra RMSNorm is All You Need for Fine Tuning to 1.58 Bits [0.0]
  Post-training quantization reduces memory and computation but often degrades accuracy.<n>Pushing quantization to the ternary (2-bit) regime yields even larger savings but is notoriously unstable.<n>Our approach matches or surpasses more elaborate knowledge-distillation pipelines on standard language-modeling benchmarks.
  arXiv  Detail & Related papers  (2025-05-12T21:14:29Z)
• LLMRefine: Pinpointing and Refining Large Language Models via Fine-Grained Actionable Feedback [65.84061725174269]
  Recent large language models (LLM) are leveraging human feedback to improve their generation quality. We propose LLMRefine, an inference time optimization method to refine LLM's output. We conduct experiments on three text generation tasks, including machine translation, long-form question answering (QA), and topical summarization. LLMRefine consistently outperforms all baseline approaches, achieving improvements up to 1.7 MetricX points on translation tasks, 8.1 ROUGE-L on ASQA, 2.2 ROUGE-L on topical summarization.
  arXiv  Detail & Related papers  (2023-11-15T19:52:11Z)

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.