EnigmaForge presents a novel benchmark where models are given only a stack of old documents—letters, receipts, and logbook margins—without an explicit question. Within these texts lies a unique logic puzzle, whose solution is guaranteed by a SAT solver at generation time and verified with an ablation certificate that every clue is essential. The benchmark continually renews itself through generated instances, and evaluation focuses on models’ intuition in solving the puzzle from the story alone, with secondary metrics on world reconstruction.
By Daniel Eisner
ExplorationBench is a new benchmark designed to evaluate AI systems’ ability to conduct scientific exploration in verifiable alien worlds. It comprises two sandbox environments—AlienCode and AlienLogic—each containing discovery targets, tasks, flawed manuals, and tool‑call schemas that force systems to formulate hypotheses, design experiments, and iterate on results. Ten AI systems were tested, revealing that while the best performers can learn and apply unfamiliar rules, their progress varies across exploration trajectories and can even regress with continued exploration.
By Ming Zhang, Zhenghao Xiang, Peizhong Gao, Yujiong Shen, Yuhui Wang, Zhonghan Yue, Shihan Dou, Zhangyue Yin, Junjie Ye, Shichun Liu, Weihuang Zheng, Jiahao Chen, Jiayi Chen, Hongzhang Liu, Jiaqi Shao, Tao Gui, Qi Zhang, Xuanjing Huang, Suncong Zheng, Maxm Pan
The paper introduces SAGE, a framework designed to reduce long‑horizon reasoning biases in large language models. It identifies two key biases—exploration bias and compounding bias—arising from complex reasoning spaces and sparse rewards, and proposes Symbolic Closure Analysis (SCA) to understand these effects. SAGE applies algebraic sparsification and hyperbolic structural guidance to suppress spurious branching and provide dense depth‑wise signals, achieving up to an eight‑fold improvement on the Andrews‑Curtis problem across multiple benchmarks and model families.
By Xinyue Zeng, Jiawei Zhang, Yujun Yan, Dawei Zhou
The paper introduces AgentX-Model, a dual‑agent framework that links proposal development with model experimentation in industrial recommender systems. The Research Agent drafts proposals from literature and prior findings, while the Model Agent runs multi‑round experiments, returning code, metrics, and open questions. The framework iteratively selects starting implementations and formulates new research questions, organizing work into Reproduce, Follow‑up, Composition, and Diagnose actions. Across production evaluations, most experiments exceeded business baselines, with recent A/B tests showing significant gains in acquisition efficiency, advertising spend, and watch time while reducing computational cost.
By Shuang Yang, Zijie Zhuang, Changxin Lao, Pengbo Xu, Hanwen Xu, Yusheng Huang, Han Gao, Guanchen Wang, Tianbao Ma, Linxun Chen, Peilin Song, Xuming Wang, Chen Li, Fan Wu, Tao Wang, Zibo Zhao, Xiangyu Wu, An Liu, Fei Pan, Peng Jiang, Chen Yang, Zhaojie Liu, Wenwu Ou
UO-FIE is a parameter‑efficient system for the Factivity Inference Evaluation 2026 (FIE2026) that predicts a distribution over nine ordered factivity intervals. It combines hard‑label supervision, utility‑based soft targets, scheduled class weights, and an ordinal loss to balance exact predictions with proximity to the correct interval. Using Qwen3.5‑9B with LoRA, UO‑FIE achieved the highest macro utility (0.8316) in the fine‑tuning track, while a prompt‑based ensemble ranked third in the non‑fine‑tuning track (macro utility 0.8450).
By Xinchen Xiao
The paper presents an explainable hate‑speech detection framework that combines DistilBERT embeddings, a Bi‑LSTM network, and an attention mechanism to capture contextual and sequential information. It uses LIME to highlight influential text features, providing transparency in predictions. Evaluated on two benchmark datasets for both binary and multi‑class tasks, the model achieves F1‑scores of 96.78%–99.53% for binary classification and 94.99%–97.00% for multi‑class classification, outperforming existing baselines.
By Rameesha Zia, Muhammad Shahid Iqbal Malik
The study evaluates various uncertainty estimation methods—probability-based, sampling-based, self-verification, evidential, and contrastive—across four open-weight audio-language models and five audio QA benchmarks. In multiple-choice settings, first-token probability measures outperform others, achieving a mean AUROC of .740, while open-ended evaluation shows lower accuracy but still predictive uncertainty. Ablation experiments reveal that removing audio evidence significantly degrades error-detection performance, indicating that uncertainty relies more on audio than on question text.
By Aaron Isidore Grace, Weiran Wang
The paper introduces the Agent Cross‑Layer Evidence (ACE) corpus, pairing application‑level telemetry with kernel‑level syscall traces to study agent security. It shows that kernel evidence alone is discriminative and that combining it with application‑level data outperforms either layer alone, revealing complementary signals. The study also demonstrates that this cross‑layer approach generalizes to unseen attack families and works across different agent runtimes.
By Spencer King, Zhilu Zhang, Mikhail Kuznetsov, Kay Liu, Baris Coskun, Wei Ding
The paper introduces an LLM chaining architecture for General Purpose Service Robots that splits instruction classification and action generation into two stages, cutting prompt length by about 45% and boosting planning consistency. Evaluation on 100 synthetic GPSR commands across three language models shows consistent improvements over single-prompt methods, with up to +37 percentage points gain on local models. Real‑robot trials on the Toyota HSR confirm that while planning success improves, execution-layer failures remain the main obstacle to full task completion.
By Lucas Da Mota Bruno, Jiahao Sim, Yoshinobu Hagiwara
EIB-Net is an Entropy‑Guided Information Bottleneck Network designed to detect AI‑generated images across diverse generative models. It introduces an Image Entropy metric to automatically select the most informative, low‑entropy patch and applies a Variational Information Bottleneck to learn compact, generalizable features. Experiments on DIFF, DiffusionForensics, and GenImage benchmarks show state‑of‑the‑art performance, achieving 85.7% accuracy with only 2% of training data and maintaining robust cross‑generator generalization.
By Zhida Zhang, Xinlei Ma, Jie Cao
DAWN (Denoising and Alignment in World models for Noise-robustness) is a perception framework that builds noise robustness directly into a world model for vision-based legged locomotion. It achieves this by feeding noisy depth to the encoder while reconstructing clean depth, and by using contrastive learning to align latent states of noisy and clean depth. The method requires no manual filter tuning, incurs no extra inference cost, and enables zero‑shot quadruped parkour on a Unitree Go1, successfully traversing stairs, gaps, and steps from raw depth observations.
By Yohan Choi, Min-Jun Kim, Jin-Sung Kim, Yong-Jae Kim, Youn-Hee Han
Med-AR introduces two autoregressive vision‑language models, Med‑AR‑8B and Med‑AR‑2B, pretrained on structured radiology reports, abnormality‑focused text, and region annotations to address long‑tailed chest X‑ray classification. The models outperform existing contrastive, self‑supervised, and supervised encoders—including Med‑CLIP, CheXFound, EVA‑Base, ARK, and BioViL‑T—across PadChest, MIMIC‑CXR, and CheXpert, achieving higher mean AUROC and AUPRC for head, medium, and tail findings and lower excess area under the risk‑coverage curve. Med‑AR also demonstrates improved selective‑prediction performance, with Med‑AR‑8B raising tail‑label mean AUPRC on MIMIC‑CXR from 0.1033 to 0.1441 and Med‑AR‑2B delivering the strongest discrimination on PadChest.
By Janhavi Prabhu, Sahil, Akshay V, Shivam Shukla, Manoj Tadepalli, Preetham Putha
The paper introduces NormPaST‑Risk, a novel framework that detects Alzheimer’s disease from online handwriting by focusing on local, segment‑level risk rather than whole‑trajectory features. It employs a multi‑scale temporal encoder, a Paper‑Air state‑space model to separate on‑paper motor execution from in‑air planning, and a healthy‑normative branch to learn normal handwriting dynamics. A weakly supervised segment‑risk module identifies high‑risk handwriting segments, achieving superior AD/HC classification on the DARWIN benchmark and offering interpretable evidence linked to disease‑related handwriting changes.
By Changqing Gong, Huafeng Qin, Moun\^im A. El-Yacoubi
AgriCountDINO is a parameter‑efficient, exemplar‑guided framework that jointly counts and localizes plants and their organs by conditioning frozen multiscale DINOv3 features on exemplar appearance and size, then decoding them into target points. It introduces missed‑object recovery and exemplar‑adaptive point NMS to improve detection accuracy. With only 8.4 M trainable parameters, it achieves a three‑shot MAE of 11.92 on the TPC‑268 benchmark and a zero‑shot MAE of 14.25 on unseen generic object categories in FSC‑147, outperforming previous methods without target‑domain training.
By Shengjie Guo, Xin Li, Borjana Arsova, Hanno Scharr, Silvio Salvi
The paper "Less is More: Encoder-only Audio-Visual Segmentation" introduces EASE, an encoder-only model for Audio‑Visual Semantic Segmentation (AVSS). EASE achieves state‑of‑the‑art accuracy while running at up to 365 FPS—about three times faster than previous Transformer‑based AVSS models—and trains in under 11 GPU‑hours. The authors demonstrate that simpler, faster architectures can match or exceed the performance of more complex models across various backbones and resolutions.
By Ilpo Viertola, Vladimir Iashin, Sophie T\"otterstr\"om, Esa Rahtu
GeoRefer-Bench is a new benchmark for verifiable geospatial referring segmentation that evaluates whether models correctly resolve spatial relations in overhead imagery. Each query is expressed as an executable logical form over a metric scene graph, and predictions are scored with Exact Query Success (EQS), requiring an exact match to the query’s referent set. The dataset contains 700 UAV scenes, 26,217 instances, 142,796 spatial relations, 20,916 executable queries across five reasoning levels, and additional paraphrases, unanswerable queries, counterfactual pairs, and leakage‑controlled splits.
By Shuaishuai Cao, Min Huang, Meng Tang, Xuan Liu, Youjin Wang, Hui Lin
The paper reports a single‑seed ablation study of the TALH language model, which combines a Multi‑head Latent Attention (MLA) branch with a custom recurrent state‑space model (SSM). Five variants ranging from 117 M to 217 M active parameters per token were trained on a FineWeb sample, and the results show that removing the SSM branch causes the largest drop in validation perplexity (315) compared to removing MLA (239). A dense‑FFN hybrid achieved a perplexity of 231, outperforming the tested top‑2 ternary‑MoE hybrid (240) while using 3.87 GB less peak training memory, and MLA‑only exhibited the flattest time‑to‑first‑token curve on an Apple M3, though the dense Transformer was faster overall.
By Christos Koutsiaris
PPTBench is a new benchmark that tests coding agents’ ability to reconstruct scientific flow diagrams from arXiv papers into editable PowerPoint slides. The dataset contains 500 tasks, each requiring agents to produce a single PPTX page with native, editable objects, and a four‑stage Agentic Judge evaluates validity, semantic correctness, rendering quality, and fine‑grained visual quality. Across 31 model configurations, the best score is 67.80, with a median of 19.47, showing that while agents can generate valid PPTX files, they still struggle with semantic and visual accuracy, especially text details.
By Xiaoqiu Wang, Yizhe Chi, Wenyi Li, Deyao Hong, Zhihan Shan, Mingju Gao, Kaisen Yang, Youjie Zheng, Calvin Xiao, Qinhuai Na
The paper introduces Selective Probability Mass Concentration (sPMC), a training framework that strengthens implicit visual grounding in multimodal large language models by selectively regularizing attention heads most responsive to visual evidence. sPMC treats attention over visual tokens as a spatial probability distribution and encourages mass to concentrate on semantically relevant regions using segmentation-derived priors, while leaving other heads unconstrained. Across six multimodal benchmarks, sPMC yields an average zero‑shot improvement of 3% and gains up to 11.3% for various models by regularizing only 3%–15% of their attention heads.
By Jiaqi Deng, Zonghan Wu, Zhan Heng, Xiaoshui Huang, Huan Huo, Guandong Xu
The paper introduces Evo-Rec, a three‑stage framework that improves generative recommendation by learning better reasoning traces for Semantic ID (SID) generation. It first aligns SIDs with textual and behavioral contexts, then selects candidate reasoning traces that improve ground‑truth item prediction, and finally refines the reasoning policy via reinforcement learning with catalog‑constrained generation and ranking‑aware feedback. Experiments on Amazon Review datasets show Evo‑Rec consistently outperforms existing discriminative, generative, and reasoning‑enhanced recommenders across all metrics.
By Mengdan Zhu, Yufan Zhao, Sophie Di, Yao Zhao, Tao Di, Yulan Yan, Sridhar Iyer, Liang Zhao