❌

Normal view

Received today — 7 October 2026 ⏭

Standardized pre-consultation by a large language model agent vs ophthalmology residents: a randomized clinical trial

✇npj Digital Medicine
By: Ming-Jie Luo · Yunxi Lai · Wei Wang · Shaowei Bi · Ling Jin · Jingjing Chen · Xiaohang Wu · Zhenzhe Lin · Jingrui Wang · Jianyu Pang · Meimei Dongye · Yue Wu · Zhenzhen Liu · Haotian Lin
5 October 2026 at 08:00

npj Digital Medicine, Published online: 05 October 2026; doi:10.1038/s41746-026-03232-x

Standardized pre-consultation by a large language model agent vs ophthalmology residents: a randomized clinical trial
Received — 15 September 2026 ⏭

VRL-Bench: Benchmarking agents on computer control tasks under finite trial budgets

✇cs.AI, q-bio.NC updates on arXiv.org
By: Yu Bai · Yukai Miao · Dawei Wang · Li Chen · Yanyu Ren · Yuqian Shi · Dan Li · Ying Xiong · Chengqiu Tan · Run Zhou · Li Li
14 September 2026 at 12:00
arXiv:2609.12404v1 Announce Type: new Abstract: Learning from trial and error is a promising way to improve language agents on complex tasks such as computer control. Reflexion introduced verbal reinforcement learning, which turns failed trials into text that guides later attempts without updating model parameters. We introduce VRL-Bench, a harness for fair evaluation of trial-and-error learning under finite trial budgets. Across three models on MiniWoB and WebShop, we evaluate updates from several prominent verbal-memory methods spanning Reflexion and later work: each improves observed success over memory-free retry in some settings but reduces it in others. Replay experiments show that using reflection can reduce success rates, revealing a trade-off between exploiting experience and continued exploration. We propose VEX$^2$, a verbal exploration--exploitation scheduler that uses a language model to jointly select policies and allocate the remaining trial budget. VEX$^2$ is the only evaluated update to achieve positive observed success-rate gains over retry in all six settings.

Confidence-Gated Transductive Test Generation for Code Reranking

✇cs.AI, q-bio.NC updates on arXiv.org
By: Sungjae Lee · Youngsik Yoon · Seockbean Song · Siwei Wang · Wei Chen · Jungseul Ok
14 September 2026 at 12:00
arXiv:2609.12489v1 Announce Type: new Abstract: Test case synthesis is crucial for evaluating and ranking programs generated by large language models (LLMs). However, constructing high-quality test cases remains challenging because reliable expected outputs are often difficult to obtain. We propose Confidence-Gated Transductive Test Generation (CoTT), which first uses an efficient inductive procedure and invokes transductive generation only when inductive confidence is low. This adaptive design improves output reliability while allocating extra computation only when needed. On code reranking benchmarks, CoTT outperforms prior baselines across the reported metrics while reducing cost relative to applying transductive generation to every input. These results show that confidence-based allocation of test-time computation provides a favorable efficiency-effectiveness trade-off with a single efficient LLM.

Hepatic Usp2 orchestrates de novo lipogenesis through G3bp2 stabilization and β-catenin activation

✇Cell Death Discovery nature.com science feeds
By: Wei Wang · Hui Wang · Yihan Liu · Siyuan Cui · Chuanxi Tang · Mengxue Wang · Chunyan Mu · Jingyu Li · Xianlong Ye · Xinmiao Liang · Shenglong Zhu
14 September 2026 at 08:00

Cell Death Discovery, Published online: 14 September 2026; doi:10.1038/s41420-026-03333-2

Hepatic Usp2 orchestrates de novo lipogenesis through G3bp2 stabilization and β-catenin activation
Received — 10 September 2026 ⏭

Gene Therapy for Hereditary Hematological Disorders: From Clinical Breakthroughs to Future Horizons

✇Molecular Therapy Nucleic Acids
By: Zhenghe Chen · Wei Wang · Youlan Wu · Yawen Qiang · Weisheng Cheng · Fang Liu
27 August 2026 at 08:00
Gene therapy is transforming hereditary hematological disorders. This review summarizes approved gene addition, editing, and silencing strategies for sickle cell disease, thalassemia, and hemophilia, highlights curative potential, and discusses remaining challenges such as immune responses, cost, and accessibility

Synchronized latency reversal and immune clearance by a multifunctional fusion protein enables HIV-1 reservoir reduction

✇Molecular Therapy
By: Fei Luo · Yinghui Cao · Na Liu · Xiaowei Wang · Wenjia Hu · Yajun Yan · Xiaoqing Xie · Dongmei Jiang · Yangyang Xu · Jun Wang · Hongyu Wang · Zhenyang Yu · Peng Qian · Shicheng Wan · Zhida Liu · Yan Gong · Tielong Chen · Shihui Song · Yong Xiong · Xiaoping Tang · Linghua Li · Li Wang · Haisheng Yu · Liang Cheng
29 July 2026 at 08:00
Latent HIV reservoirs evade both antiviral therapy and immune surveillance. Luo and colleagues develop a multifunctional fusion protein that couples reservoir reactivation with targeted immune engagement and clearance, offering a coordinated strategy to expose and eliminate persistent HIV-infected cells.

Beyond One-Size-Fits-All: Sample-Adaptive Strategy Routing for Vision Token Pruning in MLLMs

✇cs.AI, q-bio.NC updates on arXiv.org
By: Haiji Liang · Pengfei Zhou · Zhenglin Wan · Wei Wang · Yang You · Wangbo Zhao
10 September 2026 at 12:00
arXiv:2609.10346v1 Announce Type: cross Abstract: Multimodal large language models (MLLMs) process hundreds or thousands of visual tokens per image, incurring prohibitive inference costs. While existing vision token pruning methods mitigate this overhead, they implicitly assume that a single fixed pruning strategy can be applied uniformly across all inputs. Our analysis further reveals that ranking pruning methods by average benchmark accuracy conceals substantial sample-wise complementarity: although the average-best strategy excels overall, alternative strategies prove superior on a significant fraction of individual samples. To harness this diversity, we propose VIP-Router, a lightweight VIsion Pruning Router that adaptively selects the pruning strategy predicted to be best suited to each input at a specified pruning level. Conditioned on low-cost visual and textual features, VIP-Router identifies the most suitable candidate strategy while retaining full-token inference as an option when pruning is predicted to be unfavorable. Evaluated on a curated suite of pruning-sensitive visual perception benchmarks, VTC-Bench Group A, VIP-Router consistently outperforms the best fixed strategy baseline across all reduction ratios, achieving a 26.9% relative improvement in average accuracy, and a 22.0% relative increase in average utility after accounting for realized token cost. Crucially, VIP-Router operates in a plug-and-play manner without modifying underlying pruning algorithms or model weights, introducing trainable parameters equivalent to merely 0.017\% of the backbone. Furthermore, VIP-Router proves effective across various MLLM backbones and yields consistent gains on unseen benchmarks, highlighting the potential of sample adaptive routing for visual token pruning.

Bringing Value Models Back: Generative Critics for Value Modeling in LLM Reinforcement Learning

✇cs.AI, q-bio.NC updates on arXiv.org
By: Zikang Shan · Han Zhong · Liwei Wang · Li Zhao
10 September 2026 at 12:00
arXiv:2604.10701v2 Announce Type: replace-cross Abstract: Credit assignment is a central challenge in reinforcement learning (RL). Classical actor-critic methods address this challenge through fine-grained advantage estimation based on a learned value function. However, learned value models are often avoided in modern large language model (LLM) RL because conventional discriminative critics are difficult to train reliably. We revisit value modeling and argue that this difficulty is partly due to limited expressiveness. In particular, representation complexity theory suggests that value functions can be hard to approximate under the one-shot prediction paradigm used by existing value models, and our scaling experiments show that such critics do not improve reliably with scale. Motivated by this observation, we propose Generative Actor-Critic (GenAC), which replaces one-shot scalar value prediction with a generative critic that performs chain-of-thought reasoning before producing a value estimate. We further introduce In-Context Conditioning, which helps the critic remain calibrated to the current actor throughout training. GenAC improves value approximation, ranking reliability, and out-of-distribution generalization, and these gains translate into stronger downstream RL performance than both value-based and value-free baselines. Overall, our results suggest that stronger value modeling is a promising direction for improving credit assignment in LLM reinforcement learning.

CAFs shape the immunosuppressive microenvironment of pancreatic cancer through the Lin28b-STING Axis

✇(Multiomics OR Omics) AND (Pancreatic)
By: Minghe Fan · Ziyang Zhang · Wei Xu · Yuwei Liu · Ruoxuan Wang · Wenqin Hao · Sihan Huo · Yuyao Guo · Lina Wang · Ang Lv · Xiangyu Liu · Wei Wang · Chuanhui Han · Ying Zhao
3 September 2026 at 18:00

Nat Commun. 2026 Aug 7;17(1):9491. doi: 10.1038/s41467-026-76495-3.

ABSTRACT

Cancer-associated fibroblasts comprise diverse functionally distinct cellular subsets, with certain subpopulations exerting pivotal influence in shaping the pancreatic cancer immune microenvironment. Here we show that Lin28b+ cancer-associated fibroblasts contribute to establishing an immunologically cold tumor microenvironment in pancreatic ductal adenocarcinoma. Mechanistically, Lin28b directly binds to STING mRNA and promotes its degradation, thereby suppressing STING expression and downstream type I interferon signaling. Loss of Lin28b in cancer-associated fibroblasts activates the cGAS-STING-interferon signaling cascade, enhancing dendritic cell antigen presentation and CD8+ T cell cytotoxic function. Importantly, genetic inhibition of Lin28b in cancer-associated fibroblasts enhances sensitivity to anti-PD-L1 immune checkpoint blockade therapy. These findings reveal that targeting the Lin28b-STING axis represents a promising therapeutic strategy for overcoming the intrinsic resistance of pancreatic ductal adenocarcinoma to immunotherapy.

PMID:42693143 | PMC:PMC13542369 | DOI:10.1038/s41467-026-76495-3

Artificial intelligence-assisted early screening of lung cancer and accurate diagnosis of pulmonary nodules: research progress and clinical prospects from radiomics to multi-omics integration: a narrative review

✇Pulmonary nodule
By: Junwei Wang · Guixue Yang · Zhuoxin Dai · Miao Yang · Peng Dai · Mengyang Chang · Quan-Xing Liu · Ji-Gang Dai
17 June 2026 at 18:00

J Thorac Dis. 2026 May 31;18(5):537. doi: 10.21037/jtd-2026-1-0315. Epub 2026 Apr 30.

ABSTRACT

BACKGROUND AND OBJECTIVE: Lung cancer remains one of the leading causes of cancer-related death worldwide. Although low-dose computed tomography (LDCT) has improved early detection, false-positive results, overdiagnosis, and interobserver variability continue to limit screening efficiency and downstream management of pulmonary nodules. This narrative review summarizes recent progress in artificial intelligence (AI)-assisted screening, radiomics-based nodule characterization, and multi-omics integration for the precision diagnosis of lung cancer.

METHODS: A narrative review with thematic analysis was conducted using representative literature on AI-assisted lung cancer screening, quantitative imaging analysis of pulmonary nodules, radiogenomic and multi-omics integration, and clinical translation challenges. Studies were synthesized to highlight technical advances, diagnostic performance, strengths, limitations, and barriers to implementation.

KEY CONTENT AND FINDINGS: AI improves nodule detection, second-reader support, workflow efficiency, and malignancy-risk estimation in LDCT screening. Radiomics converts CT images into quantitative features that can improve discrimination between benign and malignant nodules, especially when combined with clinical variables or deep-learning models. Beyond imaging alone, radiogenomic and other multi-omics approaches link imaging phenotypes with molecular alterations, treatment response, and prognosis, thereby supporting more individualized management. However, current evidence remains limited by dataset heterogeneity, retrospective design, limited interpretability, and insufficient multicenter prospective validation.

CONCLUSIONS: AI-based imaging and multi-omics integration offer a promising pathway toward earlier detection and more precise diagnosis of lung cancer. Broader clinical adoption will depend on standardized data acquisition, robust external validation, interpretable models, and careful governance of privacy, ethics, and workflow integration.

PMID:42306713 | PMC:PMC13266817 | DOI:10.21037/jtd-2026-1-0315

Received — 27 May 2026 ⏭

DRIVE: Modeling Skills at the Reasoning and Interaction Levels for Web Agents under Continual Learning

✇cs.AI, q-bio.NC updates on arXiv.org
By: Xirui Liu · Sihang Zhou · Yanning Hou · Rong Zhou · Haoyuan Chen · Maolin He · Siwei Wang · Hao Chen · Jian Huang
26 May 2026 at 12:00
arXiv:2605.23939v1 Announce Type: new Abstract: Web agents require both high-level reasoning (for task decomposition) and low-level interactions (for page elements manipulation) to conduct different tasks. However, these knowledge types differ fundamentally: reasoning knowledge (e.g., booking a flight requires first searching for routes) is abstract and transferable across websites, while interaction knowledge (e.g., clicking the Search button at a specific coordinate on Site A) depends heavily on page-specific contexts. Existing methods store experiences uniformly. This creates a dilemma: abstract representations lose executability on concrete pages, while concrete representations fail to generalize across domains. This entanglement limits capability accumulation: on new websites, agents either fail to recognize reusable task logic due to surface-level differences or attempt infeasible actions from outdated page structures. To disentangle them, we propose DRIVE, a dual-level skill modeling framework separating historical experience into natural language reasoning skills, which capture transferable task logic, and programmatic interaction skills, grounding abstract actions to executable operations. A scene-aware coordination mechanism adaptively retrieves and invokes these dual-level skills based on task semantics. DRIVE also uses skill-level reflection to identify hierarchy-specific failure modes, enabling targeted skill library expansion and refinement. Experiments across five WebArena domains show DRIVE attains an average task success rate of 52.8%, exceeding the skill-free baseline by 7.3 percentage points. Further ablations show reasoning and interaction skills provide distinct, complementary benefits, supporting separation of transferable task logic from executable page-level operations.

A Sober Look at Agentic Misalignment in Automated Workflows

✇cs.AI, q-bio.NC updates on arXiv.org
By: Wenqian Ye · Bo Yuan · Zhichao Xu · Yijun Tian · Yawei Wang · Henry Kautz · Aidong Zhang
26 May 2026 at 12:00
arXiv:2605.24197v1 Announce Type: new Abstract: We study a class of emergent misalignment in multi-agent systems (MAS), with a focus on automated workflows, which we refer to agentic misalignment. Although these systems can solve complex tasks, they often fail because agents act according to implicit proxy utilities that do not align with the intended human goals. We formally define these behaviors and analyze them within a Bayesian framework, showing that generic utilities naturally lead to posterior collapse of agents in automated workflows. To address this issue, we propose Agentic Evidence Attribution (AEA), a novel alignment paradigm that improves agent posteriors using context-specific evidence. AEA reasons over agent actions and provides structured evidence to correct misaligned behavior during collaboration. To better understand the role of evidence, we study two instantiations of AEA: self-reflection (internal evidence from the model) and weak-to-strong generalization (external evidence on the agentic trajectory). We show that a small evidence model effectively aligns the MAS by providing orthogonal failure attribution. Our results clarify the sources of agentic misalignment in automated workflows and show that evidence-based alignment can effectively improve agent collaboration and leads to reliable multi-agent systems built on automated workflows.

Divide-and-Conquer Inference for Large-Scale Visual Recognition with Multimodal Large Language Models

✇cs.AI, q-bio.NC updates on arXiv.org
By: Zhipeng Ye · Jiaqi Huang · Feng Jiang · Qiufeng Wang · Yikang Duan · Dawei Wang · Xihang Zhou · Qian Qiao
26 May 2026 at 12:00
arXiv:2605.24799v1 Announce Type: cross Abstract: Multimodal Large Language Models (MLLMs) have demonstrated strong capabilities across a wide range of vision language tasks. However, when applied to large scale image classification, their performance degrades significantly as the label space expands a phenomenon we define as Performance Collapse in Long Sequence Recognition. Through an information theoretic analysis, we reveal that this collapse stems from a fundamental conflict between the escalating information entropy and the prominent attention dilution and decay within attention mechanisms, which impairs the model's ability to maintain a sufficient signal-to-noise ratio when processing extremely long prompts. To mitigate this, we propose Divide-and-Conquer Inference (DCI), a novel test-time scaling strategy for visual recognition with MLLMs. DCI recursively decomposes complex global classification tasks into multiple simpler, localized subproblems and employs a dynamic pruning mechanism to compress the search space. This method effectively improves the local signal to noise ratio and model accuracy by mitigating the inherent weight dilution issues in long-sequence inference. Moreover, while traditional self-attention incurs a prohibitive quadratic computational complexity, DCI achieves more favorable scaling behavior and substantially accelerates inference in large scale classification scenarios. Extensive experiments on benchmarks such as ImageNet-1K and ImageNet-21K demonstrate that DCI consistently improves classification accuracy. This enables lightweight open-source models to rival or even surpass frontier closed-source giants without any additional training or fine-tuning. As a model-agnostic, plug-and-play paradigm, DCI offers an efficient approach for scaling the inferential precision of MLLMs in large-scale scenarios.

SEP-Attack: A Simple and Effective Paradigm for Transfer-Based Textual Adversarial Attack

✇cs.AI, q-bio.NC updates on arXiv.org
By: Han Liu · Zhi Xu · Xiaotong Zhang · Feng Zhang · Xiaoming Xu · Wei Wang · Fenglong Ma · Hong Yu
26 May 2026 at 12:00
arXiv:2605.24958v1 Announce Type: cross Abstract: Despite the strong performance of deep neural networks in modern Web and language applications, they remain vulnerable to adversarial attacks, especially transferable attacks that generate adversarial examples using surrogate models without accessing the victim model. Transferable attacks in the text domain are still under-explored, with only a few studies addressing this challenging issue, often with suboptimal results due to equal treatment of submodels or inaccurate estimation of importance scores. To address these challenges, we propose a simple yet effective paradigm for transfer-based textual adversarial attack, named SEP-Attack. Specifically, we employ the Determinantal Point Process (DPP) to generate diverse surrogate ensemble weights, representing the transferability of submodels. Using these weights, we introduce a new metric to evaluate prediction confidence scores, which in turn are used to calculate word importance scores and generate adversarial candidates. Finally, we quantify the transferability score for each candidate and select the top ones as the final transferable adversarial examples. Experiments conducted on four datasets and two real-world APIs validate the efficacy of SEP-Attack, significantly outperforming state-of-the-art baselines.

Anatomy-Anchored Self-Supervision: Distilling Vision Foundation Models for Invariant Ultrasound Representation

✇cs.AI, q-bio.NC updates on arXiv.org
By: Chunzheng Zhu · Yijun Wang · Jianxin Lin · Feng Wang · Hongwei Wang · Lei Zhao · Shengli Li · Kenli Li
26 May 2026 at 12:00
arXiv:2605.25402v1 Announce Type: cross Abstract: Self-supervised pre-training paradigm has gained increasing prominence for learning transferable representations in medical imaging, yet existing methods for ultrasound (US) images operate at the image or frame level, overlooking the anatomical context for clinical-aligned representation learning. In this work, we propose an anatomy-anchored ultrasound self-supervision framework ANAUS that shifts representation learning from generic visual regions to clinically meaningful anatomical structures. Utilizing a learnable latent prompt engine alongside a one-time domain adaptation on existing public image--mask pairs, we empower the LP-SAM module to achieve annotation-free anatomy delineation at scale. Building upon this anatomical grounding, we propose a dual-policy self-supervised learning paradigm consisting of inter-view semantics-aware anatomy-separating alignment and contextual core-region prediction to enhance representation learning. Specifically, the former enforces feature invariance within identical anatomical regions while promoting discriminability across distinct structures; the latter compels the model to reconstruct corrupted regions, thereby capturing fine-grained structural details. Extensive evaluations on six public datasets demonstrate that \ours{} consistently outstrips current state-of-the-art methods while maintaining the computational efficiency essential for clinical deployment. Code is available at https://github.com/zhcz328/ANAUS.

SeqRoute: Global Budget-Aware Sequential LLM Routing via Offline Reinforcement Learning

✇cs.AI, q-bio.NC updates on arXiv.org
By: Zhongling Xu · Shunan Zheng · Wei Wang
26 May 2026 at 12:00
arXiv:2605.25424v1 Announce Type: cross Abstract: Existing LLM routing frameworks treat queries as independent events, neglecting the sequential nature of real-world user sessions constrained by global computational budgets. This mismatch inevitably leads to budget bankruptcy: myopic routing policies exhaust resources on early interactions, forcing subsequent and often more complex queries onto inadequate models. We introduce SeqRoute, a framework that formulates multi-turn routing as a finite-horizon Markov Decision Process and solves it via offline reinforcement learning. By incorporating the remaining budget into the state space and training with Conservative Q-Learning (CQL), SeqRoute learns delayed gratification to strategically preserve resources for high-stakes turns later in the session. To overcome data starvation, we propose Hindsight Budget Relabeling (HBR). This technique retrospectively simulates historical trajectories under diverse hypothetical budgets, expanding 10,000 raw sessions into 2.38 million transitions enriched with critical bankruptcy signals. At deployment, a dynamic $\lambda$-sweep mechanism enables zero-shot navigation of the cost-quality Pareto frontier without retraining. Extensive evaluations demonstrate that SeqRoute reduces operational costs by 6.0-73.5% while maintaining or improving quality, and suppresses bankruptcy rates to under 1%, strictly dominating behavior cloning, budget-aware heuristics, and static baselines across the entire Pareto frontier.

MoBiQuant: Mixture-of-Bits Quantization for Token-Adaptive Any-Precision LLM

✇cs.AI, q-bio.NC updates on arXiv.org
By: Dongwei Wang · Jinhee Kim · Seokho Han · Denis Gudovskiy · Yohei Nakata · Tomoyuki Okuno · KhayTze Peong · Kang Eun Jeon · Jong Hwan Ko · Yiran Chen · Huanrui Yang
26 May 2026 at 12:00
arXiv:2602.20191v2 Announce Type: replace-cross Abstract: Dynamic runtime latency and memory constraints necessitate flexible large language model (LLM) deployment, where an LLM can be inferred with various quantization precisions based on available computational resources. Recent work on such any-precision quantization either relies on hardware-inefficient vector quantization or induces additional scaling factors when switching between bit-widths. Meanwhile, existing post-training quantization (PTQ) methods calibrated for a fixed low precision show poor generalizability under runtime precision change. In this work, we attribute the source of poor generalization across bit-widths to a precision-dependent \textit{outlier migration} phenomenon where the distribution of PTQ-sensitive tokens changes across precisions. Motivated by this observation, we propose \texttt{MoBiQuant}, a novel any-precision Mixture-of-Bits quantization framework that adjusts weight precision for flexible LLM inference based on token sensitivity. Specifically, we propose a many-in-one recursive residual quantization that can iteratively reconstruct higher-precision weights at runtime and mitigates \textit{outlier migration} with a token-aware router to dynamically select the optimal inference precision of each token.Extensive experiments show that \texttt{MoBiQuant} matches or surpasses frontier single-precision PTQ while exhibiting strong elasticity, achieving significant memory savings and throughput gains of up to $1.34\times$ over state-of-the-art any-precision methods.

Design Conditions for Intra-Group Learning of Sequence-Level Rewards: Token Gradient Cancellation

✇cs.AI, q-bio.NC updates on arXiv.org
By: Fei Ding · Yongkang Zhang · youwei wang · Zijian Zeng
26 May 2026 at 12:00
arXiv:2604.13088v2 Announce Type: replace-cross Abstract: Reinforcement learning for multi-step reasoning with large language models (LLMs) typically relies on sparse terminal rewards, which creates a poorly conditioned credit-assignment problem: the final feedback is propagated uniformly across all intermediate decisions. This leads to high gradient variance, unstable training, and many ineffective updates, ultimately limiting sustained model improvement. We propose a counterfactual-comparison framework for credit assignment. For each input, the framework samples multiple reasoning trajectories and treats their differences as implicit approximations to alternative decisions. This yields an implicit process-level advantage estimator that converts sparse terminal rewards into step-sensitive learning signals. Building on this framework, we introduce Implicit Behavior Policy Optimization (IBPO), which substantially improves training stability and the performance ceiling on mathematical and code-reasoning benchmarks. Our results point to a promising direction for unlocking the reasoning potential of LLMs.

Reducing Credit Assignment Variance via Counterfactual Reasoning Paths

✇cs.AI, q-bio.NC updates on arXiv.org
By: Fei Ding · Yongkang Zhang · Youwei Wang · Zijian Zeng
26 May 2026 at 12:00
arXiv:2605.16302v2 Announce Type: replace-cross Abstract: Reinforcement learning for multi-step reasoning with large language models (LLMs) typically relies on sparse terminal rewards, which creates a poorly conditioned credit-assignment problem: the final feedback is propagated uniformly across all intermediate decisions. This leads to high gradient variance, unstable training, and many ineffective updates, ultimately limiting sustained model improvement. We propose a counterfactual-comparison framework for credit assignment. For each input, the framework samples multiple reasoning trajectories and treats their differences as implicit approximations to alternative decisions. This yields an implicit process-level advantage estimator that converts sparse terminal rewards into step-sensitive learning signals. Building on this framework, we introduce Implicit Behavior Policy Optimization (IBPO), which substantially improves training stability and the performance ceiling on mathematical and code-reasoning benchmarks. Our results point to a promising direction for unlocking the reasoning potential of LLMs.
❌