Stateful Reasoning via Insight Replay

• URL: http://arxiv.org/abs/2605.14457v2
• Date: Fri, 15 May 2026 19:02:32 GMT
• Title: Stateful Reasoning via Insight Replay
• Abstract summary: Chain-of-Thought (CoT) reasoning has become a foundation for eliciting multi-step reasoning in large language models.<n>We identify a major cause of this phenomenon: as the CoT grows, the model's attention to critical insights produced earlier in the trace gradually weakens.<n>We propose textbfInsightReplay, a stateful reasoning approach in which the model periodically extracts critical insights from its reasoning trace and replays them near the active generation frontier.
• Score: 51.85629502016196
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Chain-of-Thought (CoT) reasoning has become a foundation for eliciting multi-step reasoning in large language models, but recent studies show that its benefits do not scale monotonically with chain length: while longer CoT generally enables a model to tackle harder problems, on a given problem, accuracy typically increases with CoT length up to a point, after which it declines. We identify a major cause of this phenomenon: as the CoT grows, the model's attention to critical insights produced earlier in the trace gradually weakens, making those insights progressively less accessible when they are most needed. Therefore, we propose \textbf{InsightReplay}, a stateful reasoning approach in which the model periodically extracts critical insights from its reasoning trace and replays them near the active generation frontier, keeping them accessible as the reasoning scales. Extensive experiments on a $\mathbf{2}\!\times\!\mathbf{3}\!\times\!\mathbf{4}$ benchmark grid, covering model scales $\{\text{8B}, \text{30B}\}$, model families $\{\text{Qwen3.5}, \text{DeepSeek-R1-Distill-Qwen}, \text{Gemma-4}\}$, and reasoning benchmarks $\{\text{AIME}, \text{HMMT}, \text{GPQA Diamond}, \text{LiveCodeBench v5}\}$, show that 3-round InsightReplay yields accuracy gains across \textbf{all 24 settings}, with an averaged improvement of $\mathbf{+1.65}$ points over standard CoT, and a largest single-setting gain of $\mathbf{+9.2}$ points on R1-Distill-32B's LiveCodeBench v5 subset. Our results suggest that the effectiveness of test-time scaling depends not only on how much a model reasons, but also on whether critical intermediate insights remain accessible throughout long reasoning trajectories.

Related papers

• ThinkRetrieve: Retrieval-Augmented Reasoning Traces for Test-Time Scaling [60.48279351544583]
  ThinkRetrieve is a test-time scaling framework for large reasoning models.<n>It retrieves relevant exemplars at each intermediate step and injects them directly into the thinking trace.<n>It consistently improves accuracy over standard test-time scaling.
  arXiv  Detail & Related papers  (2026-08-11T13:58:07Z)
• Understanding Diversity Collapse in RLVR via the Lens of Overtraining [78.37408098404312]
  Reinforcement learning with verifiable rewards (RLVR) has become a key approach for enhancing the reasoning abilities of large language models.<n>We formalize this diversity collapse through the lens of emphovertraining<n>We propose emphBayesian Boundary Gating (BBG) which redirects optimization away from overtraining by estimating each problem's marginal contribution to the reasoning boundary.
  arXiv  Detail & Related papers  (2026-06-13T20:13:37Z)
• Why Limit the Residual Stream to Layers and Not Tokens? Persistent Memory for Continuous Latent Reasoning [0.10742675209112622]
  We propose textbfAGCLR (Adaptive Gated Continuous Latent Reasoning), which augments CoCoNuT with a textitGated Concept Stream.<n>AGCLR achieves consistent improvements across all types of datasets.
  arXiv  Detail & Related papers  (2026-06-05T15:45:44Z)
• StoryTR: Narrative-Centric Video Temporal Retrieval with Theory of Mind Reasoning [9.122764358915276]
  Current video moment retrieval excels at action-centric tasks but struggles with narrative content.<n>We introduce textbfStoryTR, the first video moment retrieval benchmark requiring ToM reasoning.<n>Our 7B textbfShorts-Moment model, trained on ToM-guided data, improves +15.1% relative IoU over baselines.
  arXiv  Detail & Related papers  (2026-04-25T08:09:31Z)
• How Much Thinking is Enough? Quantifying and Understanding Redundancy in LLM Reasoning [11.159657169036564]
  Reasoning-capable large language models solve hard problems by emitting long chains of thought.<n>But how much of this deliberation is actually necessary has never been measured at scale or explained from first principles.<n>We formalise reasoning redundancy directly in terms of the reasoning model itself.
  arXiv  Detail & Related papers  (2026-04-21T05:32:02Z)
• Shorter but not Worse: Frugal Reasoning via Easy Samples as Length Regularizers in Math RLVR [21.593023093509302]
  We show that retaining and modestly up-weighting moderately easy problems acts as an implicit length regularizer.<n>Experiments using this approach on textitQwen3-4B-Thinking-2507 achieve baseline pass@1 AIME25 accuracy while generating solutions that are, on average, nearly twice as short.
  arXiv  Detail & Related papers  (2025-11-02T17:29:16Z)
• Simple Convergence Proof of Adam From a Sign-like Descent Perspective [58.89890024903816]
  We show that Adam achieves the optimal rate of $cal O(frac1Ts14)$ rather than the previous $cal O(fracln TTs14)$.<n>Our theoretical analysis provides new insights into the role of momentum as a key factor ensuring convergence.
  arXiv  Detail & Related papers  (2025-07-08T13:19:26Z)
• Fractured Chain-of-Thought Reasoning [61.647243580650446]
  We introduce Fractured Sampling, a unified inference-time strategy that interpolates between full CoT and solution-only sampling.<n>We show that Fractured Sampling consistently achieves superior accuracy-cost trade-offs, yielding steep log-linear scaling gains in Pass@k versus token budget.
  arXiv  Detail & Related papers  (2025-05-19T11:30:41Z)
• Reasoning with Latent Thoughts: On the Power of Looped Transformers [52.84192961524481]
  We show that for many synthetic reasoning problems, a $k$-layer transformer looped $L$ times nearly matches the performance of a $kL$-layer non-looped model.<n>Our empirical analysis reveals an intriguing phenomenon: looped and non-looped models exhibit scaling behavior that depends on their effective depth.
  arXiv  Detail & Related papers  (2025-02-24T18:49:05Z)
• Improved Bound for Robust Causal Bandits with Linear Models [16.60875994745622]
  This paper investigates the robustness of causal bandits in the face of temporal model fluctuations. The proposed algorithm achieves nearly optimal $tildemathcalO(sqrtT)$ regret when $C$ is $o(sqrtT)$.
  arXiv  Detail & Related papers  (2024-05-13T14:41:28Z)

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.