❌

Normal view

Received — 4 April 2026 ⏭

Efficient amyloid-β degradation in Alzheimer’s disease using SPYTACs

✇Cell
By: Fei Teng · Jing Liu · Tongtong Cui · Xiangtian Tan · Kailun Liu · Zongren Hou · Li Zhou · Yuanzhi Xie · Rongqi Li · Da Li · Bojin Li · Dongmei Wang · Qi Zhou · Baoyang Hu · Wei Li
4 March 2026 at 08:00
SPYTAC is a synthetic peptide-programmed targeted protein degradation platform harnessing LRP1 to drive lysosomal degradation of extracellular amyloid-β in the brain and periphery. In 5×FAD mice, SPYTAC treatment efficiently degrades amyloid-β, preserves neurons, and improves cognition with reduced neuroinflammation and microhemorrhage when compared with antibody therapy.
Received — 2 April 2026 ⏭

FA-INR: Adaptive Implicit Neural Representations for Interpretable Exploration of Simulation Ensembles

✇cs.AI, q-bio.NC updates on arXiv.org
By: Ziwei Li · Yuhan Duan · Tianyu Xiong · Yi-Tang Chen · Wei-Lun Chao · Han-Wei Shen
1 April 2026 at 12:00
arXiv:2506.06858v3 Announce Type: replace-cross Abstract: Surrogate models are essential for efficient exploration of large-scale ensemble simulations. Implicit neural representations (INRs) provide a compact and continuous framework for modeling spatially structured data, but they often struggle with learning complex localized structures within the scientific fields. Recent INR-based surrogates address this by augmenting INRs with explicit feature structures, but at the cost of flexibility and substantial memory overhead. In this paper, we present Feature-Adaptive INR (FA-INR), an adaptive INR-based surrogate model for high-fidelity and interpretable exploration of ensemble simulations. Instead of relying on structured feature representations, FA-INR leverages cross-attention over a learnable key-value memory bank to allocate model capacity adaptively based on the data characteristics. To further improve scalability, we introduce a coordinate-guided mixture of experts (MoE) framework that enhances both efficiency and specialization of feature representations. More importantly, the learned experts produce an interpretable partition over the simulation domain, enabling scientists to identify complex structures and perform localized parameter-space exploration. Beyond quantitative and qualitative evaluations, we also demonstrate that our learned expert specialization can reveal meaningful scientific insights and support localized sensitivity analysis.
Received — 26 March 2026 ⏭

SARE: Sample-wise Adaptive Reasoning for Training-free Fine-grained Visual Recognition

✇cs.AI, q-bio.NC updates on arXiv.org
By: Jingxiao Yang · DaLin He · Miao Pan · Ge Su · Wenqi Zhang · Yifeng Hu · Tangwei Li · Yuke Li · Xuhong Zhang
25 March 2026 at 12:00
arXiv:2603.17729v2 Announce Type: replace-cross Abstract: Recent advances in Large Vision-Language Models (LVLMs) have enabled training-free Fine-Grained Visual Recognition (FGVR). However, effectively exploiting LVLMs for FGVR remains challenging due to the inherent visual ambiguity of subordinate-level categories. Existing methods predominantly adopt either retrieval-oriented or reasoning-oriented paradigms to tackle this challenge, but both are constrained by two fundamental limitations:(1) They apply the same inference pipeline to all samples without accounting for uneven recognition difficulty, thereby leading to suboptimal accuracy and efficiency; (2) The lack of mechanisms to consolidate and reuse error-specific experience causes repeated failures on similar challenging cases. To address these limitations, we propose SARE, a Sample-wise Adaptive textbfREasoning framework for training-free FGVR. Specifically, SARE adopts a cascaded design that combines fast candidate retrieval with fine-grained reasoning, invoking the latter only when necessary. In the reasoning process, SARE incorporates a self-reflective experience mechanism that leverages past failures to provide transferable discriminative guidance during inference, without any parameter updates. Extensive experiments across 14 datasets substantiate that SARE achieves state-of-the-art performance while substantially reducing computational overhead.

A Review of the Role of Zeqi Decoction in the Treatment of Non-Small Cell Lung Cancer

✇Pulmonary nodule
By: Lan Zhang · Yanhui Jia · Lei Song · Yiman Li · Yawei Liu · Yan Qin · Tao Wang · Na Li
18 March 2026 at 18:00

J Multidiscip Healthc. 2026 Mar 11;19:584071. doi: 10.2147/JMDH.S584071. eCollection 2026.

ABSTRACT

Non-small cell lung cancer (NSCLC) is one of the malignant tumors with the highest incidence and mortality rates. Zeqi Decoction has the functions of "promoting diuresis and reducing swelling, resolving phlegm and dispersing nodules", embodying the unique approach of traditional Chinese medicine in treating lung cancer by "strengthening the body's resistance and eliminating pathogenic factors". Modern research shows that Zeqi Decoction exerts anti-NSCLC effects through multiple pathways and targets. In terms of the material basis of its efficacy, its active ingredients (such as diterpene esters and flavonoids contained in Zeqi) have the ability to directly inhibit the proliferation, invasion and migration of tumor cells and induce apoptosis. In terms of the mechanism of action, basic experiments have revealed that Zeqi Decoction can down-regulate the S100A9/STAT3 signaling pathway, inhibit the immunosuppressive activity of myelium-derived suppressor cells (MDSCs), reshape the tumor microenvironment, thereby enhancing the cytotoxic function of CD8⁺T cells, and can also regulate the EGFR/PI3K/Akt pathway to affect PD-L1 expression. Intervene in tumor immune escape; In terms of clinical transformation, the combination of Zexi Decoction with chemotherapy and targeted therapy can improve patients' symptoms such as cough and pleural effusion, prolong progression-free survival, and alleviate the toxic and side effects of Western medical treatment. In addition, Zexi Decoction also shows potential value in reversing drug resistance such as gemcitabine. At present, there are still problems such as the lack of standardized protocols and unclear molecular mechanisms in the research. In the future, it is necessary to combine new technologies such as network pharmacology and multi-omics analysis to deepen the research on the pharmacological material basis, dose-effect relationship and evidence-based medicine of Zeqi Decoction, so as to promote the clinical application and transformation of the combination of traditional Chinese and Western medicine in the treatment of NSCLC.

PMID:41847115 | PMC:PMC12991379 | DOI:10.2147/JMDH.S584071

Received — 17 March 2026 ⏭

FedBPrompt: Federated Domain Generalization Person Re-Identification via Body Distribution Aware Visual Prompts

✇cs.AI, q-bio.NC updates on arXiv.org
By: Xin Xu · Weilong Li · Wei Liu · Wenke Huang · Zhixi Yu · Bin Yang · Xiaoying Liao · Kui Jiang
16 March 2026 at 12:00
arXiv:2603.12912v1 Announce Type: cross Abstract: Federated Domain Generalization for Person Re-Identification (FedDG-ReID) learns domain-invariant representations from decentralized data. While Vision Transformer (ViT) is widely adopted, its global attention often fails to distinguish pedestrians from high similarity backgrounds or diverse viewpoints -- a challenge amplified by cross-client distribution shifts in FedDG-ReID. To address this, we propose Federated Body Distribution Aware Visual Prompt (FedBPrompt), introducing learnable visual prompts to guide Transformer attention toward pedestrian-centric regions. FedBPrompt employs a Body Distribution Aware Visual Prompts Mechanism (BAPM) comprising: Holistic Full Body Prompts to suppress cross-client background noise, and Body Part Alignment Prompts to capture fine-grained details robust to pose and viewpoint variations. To mitigate high communication costs, we design a Prompt-based Fine-Tuning Strategy (PFTS) that freezes the ViT backbone and updates only lightweight prompts, significantly reducing communication overhead while maintaining adaptability. Extensive experiments demonstrate that BAPM effectively enhances feature discrimination and cross-domain generalization, while PFTS achieves notable performance gains within only a few aggregation rounds. Moreover, both BAPM and PFTS can be easily integrated into existing ViT-based FedDG-ReID frameworks, making FedBPrompt a flexible and effective solution for federated person re-identification. The code is available at https://github.com/leavlong/FedBPrompt.

GeoChemAD: Benchmarking Unsupervised Geochemical Anomaly Detection for Mineral Exploration

✇cs.AI, q-bio.NC updates on arXiv.org
By: Yihao Ding · Yiran Zhang · Chris Gonzalez · Eun-Jung Holden · Wei Liu
16 March 2026 at 12:00
arXiv:2603.13068v1 Announce Type: cross Abstract: Geochemical anomaly detection plays a critical role in mineral exploration as deviations from regional geochemical baselines may indicate mineralization. Existing studies suffer from two key limitations: (1) single region scenarios which limit model generalizability; (2) proprietary datasets, which makes result reproduction unattainable. In this work, we introduce \textbf{GeoChemAD}, an open-source benchmark dataset compiled from government-led geological surveys, covering multiple regions, sampling sources, and target elements. The dataset comprises eight subsets representing diverse spatial scales and sampling conditions. To establish strong baselines, we reproduce and benchmark a range of unsupervised anomaly detection methods, including statistical models, generative and transformer-based approaches. Furthermore, we propose \textbf{GeoChemFormer}, a transformer-based framework that leverages self-supervised pretraining to learn target-element-aware geochemical representations for spatial samples. Extensive experiments demonstrate that GeoChemFormer consistently achieves superior and robust performance across all eight subsets, outperforming existing unsupervised methods in both anomaly detection accuracy and generalization capability. The proposed dataset and framework provide a foundation for reproducible research and future development in this direction.

OpenSage: Self-programming Agent Generation Engine

✇cs.AI, q-bio.NC updates on arXiv.org
By: Hongwei Li · Zhun Wang · Qinrun Dai · Yuzhou Nie · Jinjun Peng · Ruitong Liu · Jingyang Zhang · Kaijie Zhu · Jingxuan He · Lun Wang · Yangruibo Ding · Yueqi Chen · Wenbo Guo · Dawn Song
16 March 2026 at 12:00
arXiv:2602.16891v2 Announce Type: replace Abstract: Agent development kits (ADKs) provide effective platforms and tooling for constructing agents, and their designs are critical to the constructed agents' performance, especially the functionality for agent topology, tools, and memory. However, current ADKs either lack sufficient functional support or rely on humans to manually design these components, limiting agents' generalizability and overall performance. We propose OpenSage, the first ADK that enables LLMs to automatically create agents with self-generated topology and toolsets while providing comprehensive and structured memory support. OpenSage offers effective functionality for agents to create and manage their own sub-agents and toolkits. It also features a hierarchical, graph-based memory system for efficient management and a specialized toolkit tailored to software engineering tasks. Extensive experiments across three state-of-the-art benchmarks with various backbone models demonstrate the advantages of OpenSage over existing ADKs. We also conduct rigorous ablation studies to demonstrate the effectiveness of our design for each component. We believe OpenSage can pave the way for the next generation of agent development, shifting the focus from human-centered to AI-centered paradigms.
  • ✇cs.AI, q-bio.NC updates on arXiv.org
  • Superficial Safety Alignment Hypothesis Jianwei Li · Jung-Eun Kim
    arXiv:2410.10862v3 Announce Type: replace-cross Abstract: As large language models (LLMs) are overwhelmingly more and more integrated into various applications, ensuring they generate safe responses is a pressing need. Previous studies on alignment have largely focused on general instruction-following but have often overlooked the distinct properties of safety alignment, such as the brittleness of safety mechanisms. To bridge the gap, we propose the Superficial Safety Alignment Hypothesis (SSAH
     
  • 🏷️
  • ↗️

Superficial Safety Alignment Hypothesis

✇cs.AI, q-bio.NC updates on arXiv.org
By: Jianwei Li · Jung-Eun Kim
16 March 2026 at 12:00
arXiv:2410.10862v3 Announce Type: replace-cross Abstract: As large language models (LLMs) are overwhelmingly more and more integrated into various applications, ensuring they generate safe responses is a pressing need. Previous studies on alignment have largely focused on general instruction-following but have often overlooked the distinct properties of safety alignment, such as the brittleness of safety mechanisms. To bridge the gap, we propose the Superficial Safety Alignment Hypothesis (SSAH), which posits that safety alignment teaches an otherwise unsafe model to choose the correct reasoning direction-fulfill or refuse users' requests-interpreted as an implicit binary classification task. Through SSAH, we hypothesize that only a few essential components can establish safety guardrails in LLMs. We successfully identify four types of attribute-critical components: Safety Critical Unit (SCU), Utility Critical Unit (UCU), Complex Unit (CU), and Redundant Unit (RU). Our findings show that freezing certain safety-critical components during fine-tuning allows the model to retain its safety attributes while adapting to new tasks. Similarly, we show that leveraging redundant units in the pre-trained model as an "alignment budget" can effectively minimize the alignment tax while achieving the alignment goal. All considered, this paper concludes that the atomic functional unit for safety in LLMs is at the neuron level and underscores that safety alignment should not be complicated. We have code implementation and other information on the project website: https://ssa-h.github.io/.

LatentChem: From Textual CoT to Latent Thinking in Chemical Reasoning

✇cs.AI, q-bio.NC updates on arXiv.org
By: Xinwu Ye · Yicheng Mao · Jia Zhang · Yimeng Liu · Li Hao · Fang Wu · Zhiwei Li · Yuxuan Liao · Zehong Wang · Yingcheng Wu · Zhiyuan Liu · Zhenfei Yin · Li Yuan · Philip Torr · Huan Sun · Xiangxiang Zeng · Mengdi Wang · Le Cong · Shenghua Gao · Xiangru Tang
16 March 2026 at 12:00
arXiv:2602.07075v4 Announce Type: replace-cross Abstract: Chemical large language models (LLMs) predominantly rely on explicit Chain-of-Thought (CoT) in natural language to perform complex reasoning. However, chemical reasoning is inherently continuous and structural, and forcing it into discrete linguistic tokens introduces a fundamental representation mismatch that constrains both efficiency and performance. We introduce LatentChem, a latent reasoning interface that decouples chemical computation from textual generation, enabling models to perform multi-step reasoning directly in continuous latent space while emitting language only for final outputs. Remarkably, we observe a consistent emergent behavior: when optimized solely for task success, models spontaneously internalize reasoning, progressively abandoning verbose textual derivations in favor of implicit latent computation. This shift is not merely stylistic but computationally advantageous. Across diverse chemical reasoning benchmarks, LatentChem achieves a 59.88\% non-tie win rate over strong CoT-based baselines on ChemCoTBench, while delivering a 10.84$\times$ average reduction in reasoning overhead. Our results provide empirical evidence that chemical reasoning is more naturally and effectively realized as continuous latent dynamics rather than discretized linguistic trajectories.
Received — 14 March 2026 ⏭

CDRRM: Contrast-Driven Rubric Generation for Reliable and Interpretable Reward Modeling

✇cs.AI, q-bio.NC updates on arXiv.org
By: Dengcan Liu · Fengkai Yang · Xiaohan Wang · Shurui Yan · Jiajun Chai · Jiahao Li · Yikun Ban · Zhendong Mao · Wei Lin · Guojun Yin
10 March 2026 at 12:00
arXiv:2603.08035v1 Announce Type: new Abstract: Reward modeling is essential for aligning Large Language Models(LLMs) with human preferences, yet conventional reward models suffer from poor interpretability and heavy reliance on costly expert annotations. While recent rubric-based approaches enhance evaluation transparency, they lack systematic quality control, yielding noisy and redundant criteria, failing to mitigate persistent biases (e.g., verbosity, position) in LLM evaluators, and creating a scalability-reliability trade-off. To address these limitations, we propose CDRRM (Contrast-Driven Rubric Reward Model), a framework built on a novel Contrast-then-Synthesis paradigm for high-quality rubric generation and guided preference judgment. CDRRM first conducts multi-dimensional contrastive profiling on preference pairs to identify causal discriminative factors, then synthesizes these insights into compact, context-aware rubrics to guide preference judg- ments. Extensive experiments on three authoritative benchmarks (RewardBench, RMBench, RMB) demonstrate that CDRRM achieves state-of-the-art performance across diverse domains and effectively mitigates aforementioned evaluation biases. Notably, our approach delivers exceptional data efficiency: training the rubric generator on only 3k high-quality samples empowers a frozen pre-trained judge model to outperform fully fine-tuned baselines. This work offers a scalable, interpretable, and data-efficient path for reward modeling.
Received — 11 March 2026 ⏭

Thinking with Gaze: Sequential Eye-Tracking as Visual Reasoning Supervision for Medical VLMs

✇cs.AI, q-bio.NC updates on arXiv.org
By: Yiwei Li · Zihao Wu · Yanjun Lv · Hanqi Jiang · Weihang You · Zhengliang Liu · Dajiang Zhu · Xiang Li · Quanzheng Li · Tianming Liu · Lin Zhao
10 March 2026 at 12:00
arXiv:2603.06697v1 Announce Type: cross Abstract: Vision--language models (VLMs) process images as visual tokens, yet their intermediate reasoning is often carried out in text, which can be suboptimal for visually grounded radiology tasks. Radiologists instead diagnose via sequential visual search; eye-tracking captures this process as time-ordered gaze trajectories that reveal how evidence is acquired over time. We use eye-gaze as supervision to guide VLM reasoning by introducing a small set of dedicated gaze tokens. These tokens are trained to predict gaze-selected image patch indices in temporal order, encouraging the model to follow human-like evidence acquisition and integration. Experiments on MIMIC-EYE and multiple external zero-shot benchmarks show consistent gains over baselines, achieving state-of-the-art in-domain performance and improved out-of-domain robustness. These results highlight temporally ordered gaze as an effective supervision signal for learning visually grounded medical reasoning.
  • ✇cs.AI, q-bio.NC updates on arXiv.org
  • Sparsity and Out-of-Distribution Generalization Scott Aaronson · Lin Lin Lee · Jiawei Li
    arXiv:2603.07388v1 Announce Type: cross Abstract: Explaining out-of-distribution generalization has been a central problem in epistemology since Goodman's "grue" puzzle in 1946. Today it's a central problem in machine learning, including AI alignment. Here we propose a principled account of OOD generalization with three main ingredients. First, the world is always presented to experience not as an amorphous mass, but via distinguished features (for example, visual and auditory channels). Seco
     
  • 🏷️
  • ↗️

Sparsity and Out-of-Distribution Generalization

✇cs.AI, q-bio.NC updates on arXiv.org
By: Scott Aaronson · Lin Lin Lee · Jiawei Li
10 March 2026 at 12:00
arXiv:2603.07388v1 Announce Type: cross Abstract: Explaining out-of-distribution generalization has been a central problem in epistemology since Goodman's "grue" puzzle in 1946. Today it's a central problem in machine learning, including AI alignment. Here we propose a principled account of OOD generalization with three main ingredients. First, the world is always presented to experience not as an amorphous mass, but via distinguished features (for example, visual and auditory channels). Second, Occam's Razor favors hypotheses that are "sparse," meaning that they depend on as few features as possible. Third, sparse hypotheses will generalize from a training to a test distribution, provided the two distributions sufficiently overlap on their restrictions to the features that are either actually relevant or hypothesized to be. The two distributions could diverge arbitrarily on other features. We prove a simple theorem that formalizes the above intuitions, generalizing the classic sample complexity bound of Blumer et al. to an OOD context. We then generalize sparse classifiers to subspace juntas, where the ground truth classifier depends solely on a low-dimensional linear subspace of the features.

FreeKV: Boosting KV Cache Retrieval for Efficient LLM Inference

✇cs.AI, q-bio.NC updates on arXiv.org
By: Guangda Liu · Chengwei Li · Zhenyu Ning · Jing Lin · Yiwu Yao · Danning Ke · Minyi Guo · Jieru Zhao
10 March 2026 at 12:00
arXiv:2505.13109v5 Announce Type: replace-cross Abstract: Large language models (LLMs) are widely deployed with rapidly expanding context windows to support increasingly demanding applications. However, long contexts pose significant deployment challenges, primarily due to the KV cache whose size grows proportionally with context length. While KV cache compression methods have been proposed to address this issue, KV dropping methods incur considerable accuracy loss, and KV retrieval methods suffer from significant efficiency bottlenecks. We propose FreeKV, a training-free algorithm-system co-optimization framework to enhance KV retrieval efficiency while preserving accuracy. On the algorithm side, FreeKV introduces speculative retrieval to shift the KV selection and recall processes out of the critical path, combined with fine-grained correction to ensure accuracy. On the system side, FreeKV employs hybrid KV layouts across CPU and GPU memory to eliminate fragmented data transfers, and leverages double-buffered streamed recall to further improve efficiency, enabling effective overlap with computation, full latency hiding, and practical speedups from speculative recall. Experiments demonstrate that FreeKV achieves near-lossless accuracy across various scenarios and models, delivering up to a 13$\times$ speedup compared to SOTA KV retrieval methods. Code is available at https://github.com/sjtu-zhao-lab/FreeKV.
Received — 5 March 2026 ⏭

LifeBench: A Benchmark for Long-Horizon Multi-Source Memory

✇cs.AI, q-bio.NC updates on arXiv.org
By: Zihao Cheng · Weixin Wang · Yu Zhao · Ziyang Ren · Jiaxuan Chen · Ruiyang Xu · Shuai Huang · Yang Chen · Guowei Li · Mengshi Wang · Yi Xie · Ren Zhu · Zeren Jiang · Keda Lu · Yihong Li · Xiaoliang Wang · Liwei Liu · Cam-Tu Nguyen
5 March 2026 at 13:00
arXiv:2603.03781v1 Announce Type: new Abstract: Long-term memory is fundamental for personalized agents capable of accumulating knowledge, reasoning over user experiences, and adapting across time. However, existing memory benchmarks primarily target declarative memory, specifically semantic and episodic types, where all information is explicitly presented in dialogues. In contrast, real-world actions are also governed by non-declarative memory, including habitual and procedural types, and need to be inferred from diverse digital traces. To bridge this gap, we introduce Lifebench, which features densely connected, long-horizon event simulation. It pushes AI agents beyond simple recall, requiring the integration of declarative and non-declarative memory reasoning across diverse and temporally extended contexts. Building such a benchmark presents two key challenges: ensuring data quality and scalability. We maintain data quality by employing real-world priors, including anonymized social surveys, map APIs, and holiday-integrated calendars, thus enforcing fidelity, diversity and behavioral rationality within the dataset. Towards scalability, we draw inspiration from cognitive science and structure events according to their partonomic hierarchy; enabling efficient parallel generation while maintaining global coherence. Performance results show that top-tier, state-of-the-art memory systems reach just 55.2\% accuracy, highlighting the inherent difficulty of long-horizon retrieval and multi-source integration within our proposed benchmark. The dataset and data synthesis code are available at https://github.com/1754955896/LifeBench.

DisenReason: Behavior Disentanglement and Latent Reasoning for Shared-Account Sequential Recommendation

✇cs.AI, q-bio.NC updates on arXiv.org
By: Jiawei Cheng · Min Gao · Zongwei Wang · Xiaofei Zhu · Zhiyi Liu · Wentao Li · Wei Li · Huan Wu
5 March 2026 at 13:00
arXiv:2603.03782v1 Announce Type: cross Abstract: Shared-account usage is common on streaming and e-commerce platforms, where multiple users share one account. Existing shared-account sequential recommendation (SSR) methods often assume a fixed number of latent users per account, limiting their ability to adapt to diverse sharing patterns and reducing recommendation accuracy. Recent latent reasoning technique applied in sequential recommendation (SR) generate intermediate embeddings from the user embedding (e.g, last item embedding) to uncover users' potential interests, which inspires us to treat the problem of inferring the number of latent users as generating a series of intermediate embeddings, shifting from inferring preferences behind user to inferring the users behind account. However, the last item cannot be directly used for reasoning in SSR, as it can only represent the behavior of the most recent latent user, rather than the collective behavior of the entire account. To address this, we propose DisenReason, a two-stage reasoning method tailored to SSR. DisenReason combines behavior disentanglement stage from frequency-domain perspective to create a collective and unified account behavior representation, which serves as a pivot for latent user reasoning stage to infer the number of users behind the account. Experiments on four benchmark datasets show that DisenReason consistently outperforms all state-of-the-art baselines across four benchmark datasets, achieving relative improvements of up to 12.56\% in MRR@5 and 6.06\% in Recall@20.

Preference Leakage: A Contamination Problem in LLM-as-a-judge

✇cs.AI, q-bio.NC updates on arXiv.org
By: Dawei Li · Renliang Sun · Yue Huang · Ming Zhong · Bohan Jiang · Jiawei Han · Xiangliang Zhang · Wei Wang · Huan Liu
5 March 2026 at 13:00
arXiv:2502.01534v3 Announce Type: replace-cross Abstract: Large Language Models (LLMs) as judges and LLM-based data synthesis have emerged as two fundamental LLM-driven data annotation methods in model development. While their combination significantly enhances the efficiency of model training and evaluation, little attention has been given to the potential contamination brought by this new model development paradigm. In this work, we expose preference leakage, a contamination problem in LLM-as-a-judge caused by the relatedness between the synthetic data generators and LLM-based evaluators. To study this issue, we first define three common relatednesses between the data generator LLM and the judge LLM: being the same model, having an inheritance relationship, and belonging to the same model family. Through extensive experiments, we empirically confirm the bias of judges towards their related student models caused by preference leakage across multiple LLM baselines and benchmarks. Further analysis suggests that preference leakage is a pervasive and real-world problem that is harder to detect compared to previously identified biases in LLM-as-a-judge scenarios. All of these findings imply that preference leakage is a widespread and challenging problem in the area of LLM-as-a-judge. We release all codes and data at: https://github.com/David-Li0406/Preference-Leakage.

RxnNano:Training Compact LLMs for Chemical Reaction and Retrosynthesis Prediction via Hierarchical Curriculum Learning

✇cs.AI, q-bio.NC updates on arXiv.org
By: Ran Li · Shimin Di · Haowei LI · Luanshi Bu · Jiachuan Wang · Wangze Ni · Lei Chen
4 March 2026 at 13:00
arXiv:2603.02215v1 Announce Type: cross Abstract: Chemical reaction prediction is pivotal for accelerating drug discovery and synthesis planning. Despite advances in data-driven models, current approaches are hindered by an overemphasis on parameter and dataset scaling. Some methods coupled with evaluation techniques that bypass fundamental challenges in reaction representation and fail to capture deep chemical intuition like reaction common sense and {topological atom mapping logic}. We argue that the core challenge lies in instilling these knowledge into the models. To this end, we propose a unified framework that prioritizes chemical understanding over scale through three key innovations: (1) a {Latent Chemical Consistency} objective that models reactions as movements on a continuous chemical manifold, ensuring reversible and physically plausible transformations; (2) a {Hierarchical Cognitive Curriculum} that trains the model through progressive stages, from syntax mastery to semantic reasoning, building robust chemical intuition; (3) {Atom-Map Permutation Invariance (AMPI)}, which force the model to learn invariant relational topology and balance multi-task learning. (4)and structured plan-based reasoning to improve the performance of the LLMs. Our compact {0.5B-parameter model}, \textbf{RxnNano} significantly outperforms fine-tuned LLMs ten times larger (>7B) and all the domain baselines, achieving a 23.5\% Top-1 accuracy improvement on rigorous benchmarks without test-time augmentation. https://github.com/rlisml/RxnNano.

Self-Play Only Evolves When Self-Synthetic Pipeline Ensures Learnable Information Gain

✇cs.AI, q-bio.NC updates on arXiv.org
By: Wei Liu · Siya Qi · Yali Du · Yulan He
4 March 2026 at 13:00
arXiv:2603.02218v1 Announce Type: cross Abstract: Large language models (LLMs) make it plausible to build systems that improve through self-evolving loops, but many existing proposals are better understood as self-play and often plateau quickly. A central failure mode is that the loop synthesises more data without increasing learnable information for the next iteration. Through experiments on a self-play coding task, we reveal that sustainable self-evolution requires a self-synthesised data pipeline with learnable information that increases across iterations. We identify triadic roles that self-evolving LLMs play: the Proposer, which generates tasks; the Solver, which attempts solutions; and the Verifier, which provides training signals, and we identify three system designs that jointly target learnable information gain from this triadic roles perspective. Asymmetric co-evolution closes a weak-to-strong-to-weak loop across roles. Capacity growth expands parameter and inference-time budgets to match rising learnable information. Proactive information seeking introduces external context and new task sources that prevent saturation. Together, these modules provide a measurable, system-level path from brittle self-play dynamics to sustained self-evolution.
❌