When to Think, When to Speak: Learning Disclosure Policies for LLM Reasoning

• URL: http://arxiv.org/abs/2605.03314v2
• Date: Wed, 06 May 2026 01:55:04 GMT
• Title: When to Think, When to Speak: Learning Disclosure Policies for LLM Reasoning
• Abstract summary: In single-stream autoregressive interfaces, the same tokens both update the model state and constitute an irreversible public commitment.<n>We introduce Side-by-Side (SxS) Interleaved Reasoning, which makes disclosure timing a controllable decision within standard autoregressive generation.
• Score: 60.714416943748866
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: In single-stream autoregressive interfaces, the same tokens both update the model state and constitute an irreversible public commitment. This coupling creates a silence tax: additional deliberation postpones the first task-relevant content, while naive early streaming risks premature commitments that bias subsequent generations. We introduce Side-by-Side (SxS) Interleaved Reasoning, which makes disclosure timing a controllable decision within standard autoregressive generation. SxS interleaves partial disclosures with continued private reasoning in the same context, but releases content only when it is supported by the reasoning so far. To learn such pacing without incentivizing filler, we construct entailment-aligned interleaved trajectories by matching answer prefixes to supporting reasoning prefixes, then train with SFT to acquire the dual-action semantics and RL to recover reasoning performance under the new format. Across two Qwen3 architectures/scales (MoE Qwen3-30B-A3B, dense Qwen3-4B) and both in-domain (AIME25) and out-of-domain (GPQA-Diamond) benchmarks, SxS improves accuracy--content-latency Pareto trade-offs under token-level proxies such as inter-update waiting.

Related papers

• Weave of Formal Thought [51.56484100374058]
  We introduce Weave of Formal Thought (WoFT), a paradigm uniting rigorous syntactic validation with learned structural representations.<n>We present a latent-variable fine-tuning method training the language model to interleave non-terminal grammar symbols directly into generation.<n>For Python, fine-tuning StarCoder2-3B with our RWS objective reduces per-token cross-entropy by 14.3% relative to a text-only SFT baseline.
  arXiv  Detail & Related papers  (2026-06-24T15:58:11Z)
• Teaching Diffusion to Speculate Left-to-Right [3.205247598097648]
  Large language models (LLMs) achieve remarkable performance across a wide range of tasks, but their autoregressive decoding process incurs substantial inference costs.<n> Speculative decoding addresses this bottleneck by employing a lightweight draft model to propose multiple future tokens that are subsequently verified in parallel by a larger target model.<n>Recent work has demonstrated that diffusion language models are well suited for this setting, as they can generate entire blocks of draft tokens in parallel.
  arXiv  Detail & Related papers  (2026-06-10T01:21:56Z)
• LASAR: Latent Adaptive Semantic Aligned Reasoning for Generative Recommendation [33.48046116606003]
  Latent reasoning has emerged as an effective paradigm in Large Language Models (LLMs)<n>We propose LASAR (Latent Adaptive Semantic Aligned Reasoning), an SFT-then-RL framework.<n> Experiments on three real-world datasets demonstrate that LASAR outperforms all baselines.
  arXiv  Detail & Related papers  (2026-05-11T08:52:53Z)
• The Extrapolation Cliff in On-Policy Distillation of Near-Deterministic Structured Outputs [52.709361620508595]
  ListOPD brings a student to in-domain with an 8B-SFT baseline at one-fifth the parameters.<n>On Amazon Fashion, three pre-registered tests--a fine-grid cliff interval test, and a small-clip cross-prediction--fall within their locked prediction windows--with the small-clip value matching the closed-form prediction below grid resolution.
  arXiv  Detail & Related papers  (2026-05-09T06:48:00Z)
• TwinGate: Stateful Defense against Decompositional Jailbreaks in Untraceable Traffic via Asymmetric Contrastive Learning [60.68349524623048]
  Decompositional jailbreaks pose a critical threat to large language models.<n>We present TwinGate, a stateful dual-encoder defense framework.<n>We construct a comprehensive dataset of over 3.62 million instructions spanning 8,600 distinct malicious intents.
  arXiv  Detail & Related papers  (2026-04-30T13:44:01Z)
• Progressive Online Video Understanding with Evidence-Aligned Timing and Transparent Decisions [75.23170605943457]
  textbfmodel is a framework that decouples reasoning control from memory integration.<n>emphActive Thinking Decision Maker (ATDM) is a transparent reasoning controller that externalizes its decision process.<n>emphHierarchical Progressive Semantic Integration (HPSI) module acts as an efficient memory system.
  arXiv  Detail & Related papers  (2026-04-20T16:15:33Z)
• Beyond Scattered Acceptance: Fast and Coherent Inference for DLMs via Longest Stable Prefixes [10.877713536966601]
  Longestahead Prefix (LSP) scheduler is a training-free and model-agnostic inference paradigm based on monolithic prefix absorption.<n>LSP evaluates token stability via a single forward pass, dynamically identifies a contiguous left-aligned block of stable predictions.<n>It snaps its boundary to natural linguistic or structural acceptances before an atomic commitment.
  arXiv  Detail & Related papers  (2026-03-05T18:25:26Z)
• Breaking Contextual Inertia: Reinforcement Learning with Single-Turn Anchors for Stable Multi-Turn Interaction [49.03500737694832]
  We introduce textbfReinforcement textbfLearning with textbfTurn textbfRLSTA, a generalizable training approach designed to stabilize multi-turn interaction.<n>Experiments show that RLSTA significantly outperforms standard fine-tuning and abstention-based methods.
  arXiv  Detail & Related papers  (2026-03-05T04:04:59Z)
• CORE: Context-Robust Remasking for Diffusion Language Models [51.59514489363897]
  We propose Context-Robust Remasking (CORE), a training-free framework for inference-time revision.<n>Rather than trusting static token probabilities, CORE identifies context-brittle tokens by probing their sensitivity to targeted masked-context perturbations.<n>On LLaDA-8B-Base, CORE delivers consistent improvements across reasoning and code benchmarks, outperforming compute-matched baselines and improving MBPP by up to 9.2 percentage points.
  arXiv  Detail & Related papers  (2026-02-04T00:12:30Z)
• SoftHateBench: Evaluating Moderation Models Against Reasoning-Driven, Policy-Compliant Hostility [6.293833791034378]
  We introduce textbftextscSoftHateBench, a generative benchmark that produces soft-hate variants while preserving the hostile standpoint.<n>The benchmark spans textbf7 sociocultural domains and textbf28 target groups, comprising textbf4,745 soft-hate instances.<n> Evaluations across encoder-based detectors, general-purpose LLMs, and safety models show a consistent drop from hard to soft tiers.
  arXiv  Detail & Related papers  (2026-01-28T05:04:18Z)
• Double: Breaking the Acceleration Limit via Double Retrieval Speculative Parallelism [19.7914286780195]
  We introduce textscDouble (Double Retrieval Speculative Parallelism)<n>We enable the draft model to execute iterative retrieval speculations to break the theoretical speedup limits.<n>Experiments demonstrate state-of-the-art speedup of $textbf5.3times$ on LLaMA3.3-70B and $textbf2.8times$ on Qwen3-32B.
  arXiv  Detail & Related papers  (2026-01-09T04:35:21Z)
• Exploiting Tree Structure for Credit Assignment in RL Training of LLMs [11.64053639889468]
  Reinforcement learning improves reasoning, yet sparse delayed reward over long sequences makes token-level credit assignment the key bottleneck.<n>We study the verifiable-reward setting, where the final answer is checkable and multiple responses can be drawn per prompt.<n>We propose textbfTEMPO (emphtextbfTree-textbfEstimated textbfMean Prefix Value for textbfPolicy textbfOptimization), a critic-free algorithm that augments the group
  arXiv  Detail & Related papers  (2025-09-22T18:37:24Z)
• Thinking Before You Speak: A Proactive Test-time Scaling Approach [54.8205006555199]
  We implement our idea as a reasoning framework, named emphThinking Before You Speak (TBYS)<n>We design a pipeline for automatically collecting and filtering in-context examples for the generation of emphinsights.<n>Experiments on challenging mathematical datasets verify the effectiveness of TBYS.
  arXiv  Detail & Related papers  (2025-08-26T03:43:32Z)
• What am I missing here?: Evaluating Large Language Models for Masked Sentence Prediction [2.8514881296685113]
  Next Token Prediction (NTP) limits a model's ability to plan ahead or maintain long-range coherence.<n>We evaluate three commercial LLMs on Masked Sentence Prediction (MSP)<n>Our key finding reveals that commercial LLMs, despite their superlative performance in other tasks, are poor at predicting masked sentences in low-structured domains.
  arXiv  Detail & Related papers  (2025-08-11T07:25:50Z)
• TRACEALIGN -- Tracing the Drift: Attributing Alignment Failures to Training-Time Belief Sources in LLMs [7.125400292079228]
  Large Language Models (LLMs) fine-tuned to align with human values often exhibit alignment drift.<n>While prior work has behaviorally characterized alignment failure, little is known about the training-time belief sources underlying these failures.<n>We introduce TraceAlign, a unified framework for tracing unsafe completions back to their root causes in the model's training corpus.
  arXiv  Detail & Related papers  (2025-08-04T05:03:35Z)
• Beyond 'Aha!': Toward Systematic Meta-Abilities Alignment in Large Reasoning Models [86.88657425848547]
  Large reasoning models (LRMs) already possess a latent capacity for long chain-of-thought reasoning.<n>We explicitly align models with three meta-abilities: deduction, induction, and abduction, using automatically generated, self-verifiable tasks.<n>Our three stage-pipeline individual alignment, parameter-space merging, and domain-specific reinforcement learning, boosts performance by over 10% relative to instruction-tuned baselines.
  arXiv  Detail & Related papers  (2025-05-15T17:58:33Z)
• Dual-Level Decoupled Transformer for Video Captioning [15.193977761440404]
  Video captioning aims to understand the video semantic concept of the video and generate descriptive sentences. mathcalD2$ is a dual-level vision decoupled transformer pipeline. emph(i) for video-temporal representation, we decouple the process of it into "first-spatial-then-temporal" paradigm.
  arXiv  Detail & Related papers  (2022-05-06T06:37:07Z)

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.