The paper introduces T1, a 122‑billion‑parameter Mixture‑of‑Experts model trained with reinforcement learning to perform long‑horizon terminal tasks such as coding and scientific discovery. T1 operates a real shell in a cloud sandbox, making over 300 tool‑call turns per task and receiving rewards from task‑specific verifiers. The authors detail a training recipe that includes aggressive warm‑starting, TITO construction with drift repair, and rollout‑routing replay, achieving significant performance gains on Terminal‑Bench 2.1 and surpassing GPT‑5.4 and GLM‑5.1 on the Long‑Horizon Terminal Bench.
By Junyao Yang, Yucheng Shi, Zhongzhi Li, Ruhan Wang, Zongxia Li, Haitao Mi, Leowei Liang
GRIPNet is a new CT‑based pulmonary nodule detector that incorporates a Gaussian radial intensity prior, reflecting the regular pattern of intensity peaks at nodule centers and Gaussian decay outward. By replacing generic square convolutions with pinwheel, dual‑frequency, dilated masked attention, and an adaptive loss, the network aligns each module with measurable intensity properties. The method achieves state‑of‑the‑art mAP@0.5 scores of 95.3%, 91.6%, and 97.9% on KanserSet, LUNA16, and Lung‑PET‑CT‑Dx, respectively, while maintaining real‑time speed and improving high‑IoU localization.
By Haojie Yang, Ran Su
E-CONAN introduces Arabic textual entailment and natural inference benchmarks comprising two datasets: E-CONAN-2 (2-way RTE) and E-CONAN-3 (3-way NLI). The datasets are built from automatically-translated pairs, human-validated machine translations, hand-crafted pairs from Arabic teaching books, and rumor-containing news headlines. The authors evaluated nine multilingual pretrained models and five large language models on these benchmarks, demonstrating that E-CONAN offers a more diverse and robust assessment than existing datasets like XNLI and ArNLI.
By Khloud AL Jallad, Nada Ghneim, Ghaida Rebdawi
VikingRAG is a directory‑aware semantic data management system that reduces token usage in retrieval‑augmented generation by tightly integrating semantic and structural access. It employs multi‑round retrieval traces as reusable experience edges and an adaptive escalation strategy to avoid unnecessary multi‑round exploration. Experiments show that VikingRAG achieves comparable accuracy to state‑of‑the‑art methods while using only 11.6%–51.9% of their tokens, and further reductions to 5.1%–32.5% with trace reuse and escalation.
By Peiyuan Gao, Gaoyuan Zhang, Haojie Qin, Yahui Sun, Qianyi Zhang, Yunhao Zhang, Zeyu Wang, Wei Lu
The paper introduces a multi‑signal pipeline for detecting hallucinations in large language models, combining fine‑tuned DeBERTa‑v3 classification, Monte Carlo Dropout uncertainty, and temperature‑scaled calibration. On the HaluEval benchmark it achieves high performance (F1 = 0.915, AUROC = 0.977) across QA, summarization, and dialogue, and shows that 25 % of training data yields 77 % of full‑data performance. The authors also demonstrate that applying Direct Preference Optimization to a Qwen2.5‑0.5B generator cuts hallucination rates from 85.5 % to 37.7 %, and that domain‑specific fine‑tuning (PubMedBERT on SciFact) outperforms general‑domain models for biomedical text.
By Varun Teja Chundru, Debasmita Biswas
The paper introduces IRS (Incongruity-Resolution Supervision), a framework that breaks humor understanding into three parts: identifying mismatches in a visual scene, creating coherent reinterpretations of those mismatches, and aligning these interpretations with human preferences. IRS uses structured reasoning traces to guide models from visual perception to humorous interpretation, and it is evaluated on the New Yorker Cartoon Caption Contest. Experiments on 7B, 32B, and 72B models show that IRS improves caption matching and ranking, with the 72B model achieving 76.10% ranking accuracy—outperforming non-expert humans and all other multimodal baselines—and demonstrates transferable reasoning patterns in zero‑shot settings.
By Hatice Merve Vural, Doga Kukul, Ege Erdem Ozlu, Demir Ekin Arikan, Bob Mankoff, Erkut Erdem, Aykut Erdem
The paper introduces a framework for out-of-distribution (OOD) detection that addresses the trade‑off between detection performance and classification accuracy caused by fine‑tuning with auxiliary outlier data. It optimizes three factors—model reminder, data sampling, and representation learning—by proposing Self‑Knowledge Distillation to preserve accuracy, Semi‑hard Outlier Sampling to enhance detection with minimal data, and Outlier‑aware Supervised Contrastive Learning to improve ID‑OOD separability. The combined approach yields cumulative gains, outperforming existing methods on diverse benchmarks, especially in long‑tailed scenarios, and offers a robust baseline for real‑world OOD detection.
By Hyunjun Choi, JaeHo Chung, Hawook Jeong
The paper introduces AmazonSWE, a dataset covering over 19,000 river sections in the Amazon basin for 10 years (2016‑2026) that integrates satellite altimetry, including SWOT, to enable large‑scale spatiotemporal graph imputation. The dataset is extremely sparse—fewer than 1% of sections are observed daily—and features a directed acyclic river topology that is larger and structurally distinct from existing benchmarks. The authors demonstrate that conventional imputation methods struggle with this topology, scale, and sparsity, and propose a bidirectional selective state‑space model that outperforms prior approaches, reducing RMSE against in‑situ gauges by 18‑39% and providing predictions for every river section.
whyItMatters":"AmazonSWE offers a novel, real‑world use case that could improve flood forecasting and water resource management by enabling more accurate and comprehensive water surface elevation estimates across a vast, sparsely monitored river network."
By Ruben Cartuyvels, Karim Douch, Gabriele Bertoli, Mounia El Baz, Artemis Vrettou, S\'ebastien Lef\`evre, Diego Fernandez Prieto
Recursive Code World Models (RCWM) present a framework that reconstructs complex 3D worlds from a single reference image by combining a Recursive Scene Program (RSP) representation with a construction solver that recursively calls itself. The RSP encodes the world as compositional scene code, while each solver invocation follows a global‑local‑global recursion: it first establishes the whole scene, then recursively reconstructs unresolved parts, and finally revisits the whole to refine composition. This approach preserves scene‑wide geometry and relationships, propagates a shared camera projection across levels, and uses a vision‑language coding agent to compare reference images with scene renders, enabling fine‑scale structure refinement and improved reconstruction over prior methods.
By Zhiqi Li, Yuxuan Liao, Bo Zhu
The paper introduces LLM-Anchored Paralinguistic Enrichment (LAPE), a method that enhances language model representations with speech-based paralinguistic cues such as pauses and word elongations. LAPE incorporates three innovations: prosodic event textualization, lexico-prosodic unitization and chunking, and text-anchored paralinguistic fusion using NormGate. Evaluations on the ADReSS and ADReSSo datasets show that LAPE achieves state‑of‑the‑art performance in detecting Alzheimer’s disease from speech.
By Xiao Wei, Yuqin Lin, Yaru Cao, Jinyu Li, Bin Wen, Kai Li, Yueying Chen, Longbiao Wang, Jianwu Dang
The paper introduces an experiment where a small masked language model (DeBERTa) is initialized with visual embeddings for tokens that correspond to image regions, following St. Augustine’s ostensive definition of word learning. The visual initialization leaves a measurable imprint that persists through training, yet it does not improve performance on most BabyLM benchmarks that test abstract grammatical knowledge. However, the seeded models show a consistent advantage in zero‑shot object‑property tasks and in a custom Visual‑Property Swap benchmark that probes color, material, size, and shape knowledge, with the advantage confined to the seeded words and transferable to newly seeded words.
By Lisa Bylinina
The paper introduces a Database Normalization Benchmark (DNBENCH) with 3,275 samples to evaluate how well Large Language Models (LLMs) can perform database normalization from 1NF to BCNF, assessing semantic equivalence, structural accuracy, and logical validity. It identifies common failures in dependency inference, schema decomposition, and inter-table constraint reconstruction across various complexity levels. The authors also propose a Multi-Agent Reasoning for Schemas (MARS) framework that separates evidence extraction, violation diagnosis, and decomposition planning from schema generation, achieving an 82.0% improvement in DNB-SCORE over a single-prompt baseline.
By Dong-Jae Koh, Huisu Kim, SeongHwan Yoon, Lasse M. Jantsch, Chun-Hee Lee, Seonghyeon Lee, Young-Kyoon Suh
The paper introduces StructMemEval, a benchmark designed to assess how well large language model (LLM) agents can organize their long‑term memory rather than merely recall facts. It compiles tasks that humans typically solve by structuring knowledge—such as transaction ledgers, to‑do lists, and trees—and evaluates agents on these. Experiments show that simple retrieval‑augmented LLMs struggle with such organization tasks, while memory‑augmented agents perform better when explicitly prompted to structure their memory, yet many modern LLMs still fail to recognize memory structures without prompting.
By Alina Shutova, Alexandra Olenina, Ivan Vinogradov, Anton Sinitsin
The paper revisits the continuous diffusion language model Plaid and introduces RePlaid, aligning its architecture with modern discrete diffusion models. RePlaid achieves a compute gap of only 20× compared to autoregressive models, surpasses Duo with fewer parameters, and outperforms MDLM in over‑trained settings. On OpenWebText, RePlaid sets a new state‑of‑the‑art continuous diffusion perplexity of 22.1 and demonstrates superior generation quality, while theoretical analysis links likelihood‑based training to linear cross‑entropy over time and structured embedding geometries.
By Zhihan Yang, Wei Guo, Shuibai Zhang, Subham Sekhar Sahoo, Yongxin Chen, Arash Vahdat, Morteza Mardani, John Thickstun
CausalArena is a unified, evolvable benchmark designed to evaluate causal discovery methods across diverse structural causal models (SCMs). It incorporates synthetic SCMs for controlled structural variation, semantic operational SCMs for human-auditable environments, and formula-grounded SCMs to test discovery under explicit scientific mechanisms, along with real-world datasets for external validity. Experiments show that performance rankings vary significantly across SCM families and protocols, indicating that strong results on one benchmark do not generalize to others, especially in the context of causal discovery foundation models.
The paper introduces a multi‑signal pipeline for detecting hallucinations in large language model outputs, combining fine‑tuned DeBERTa‑v3 classification, Monte Carlo Dropout uncertainty, and temperature‑scaled calibration. On the HaluEval benchmark it achieves strong performance (F1 = 0.915, AUROC = 0.977) and further improves accuracy to 93.2% with MC Dropout. The authors also demonstrate that applying Direct Preference Optimization to a Qwen2.5‑0.5B generator reduces hallucination rates from 85.5% to 37.7%, and show that domain‑specific fine‑tuning (PubMedBERT on SciFact) yields better results than general‑domain training.
RAG-Safety-Bench is a benchmark designed to evaluate how retrieval-augmented generation (RAG) affects the safety of large language models (LLMs). It isolates safety impacts by testing four conditions: non-RAG, RAG with an oracle document, RAG with related but non-answer documents, and RAG with random safe documents. Results on five open-source LLMs reveal an inverse relationship between benign and unsafe capabilities, show that standard safety guardrails do not guarantee safety in RAG, and confirm that even benign documents can trigger unsafe outputs.
The paper introduces a quantum method for extracting spectral features from the density of states (DOS) of a problem-dependent Hamiltonian, applied to signed graphs represented as Ising models. It demonstrates that standardized moments of the Ising DOS count signed closed walks, are switching‑invariant, and size‑free, enabling accurate learning of the frustration index with a mean error of 0.4 on 140,000 labeled graphs. The authors propose DOS‑QPE, a phase estimation technique that requires far fewer shots than traditional methods, and highlight potential applications in social network analysis, spin‑glass studies, correlation clustering, and protein‑interaction networks.
The paper introduces AmazonSWE, a dataset covering over 19,000 river sections in the Amazon basin for 10 years, combining satellite altimetry and in‑situ gauge data to enable large‑scale spatiotemporal graph imputation. The authors highlight the extreme sparsity of observations—less than 1% of sections per day—and the directed acyclic topology of river networks, which challenge existing imputation methods. They propose a bidirectional selective state‑space model that samples connected subgraphs and uses topology‑aware positional encodings, achieving 18–39% lower RMSE than the current state‑of‑the‑art SWOT‑based approach while providing predictions for all river sections.
Boundary representation (B‑rep) is the standard format for parametric 3D models in CAD systems, yet the same solid can be encoded by different B‑reps due to varying operations, kernel rebuilds, or export settings. Existing B‑rep encoders fail to handle these variations, collapsing on standard benchmarks and real‑world perturbations. The authors introduce the canonical region graph, an input representation derived directly from the solid, which offers theoretical invariance to repartitioning and rigid motions and performs as well as the best baseline while remaining stable across all tested perturbations.