Model releases, architecture work and prompting research on large language models — from frontier-lab announcements to the arXiv papers behind them.
arXiv:2610.02181v1 Announce Type: new Abstract: We present OmniSeek, an agentic framework that transforms an Omni Large Language Model (Omni-LLM) into an active, multi-turn reasoning agent with nativ...
arXiv:2610.02210v1 Announce Type: new Abstract: I don't think I have ever done anything as peculiar in my life. Among other things, it shows a young man looking with interest at a print on the wall o...
arXiv:2610.00680v1 Announce Type: cross Abstract: Curvature is often treated as an intrinsic property of a representation, although its empirical effect also depends on coordinate scale, learned logi...
arXiv:2610.01999v1 Announce Type: new Abstract: Spatial reasoning benchmarks evaluate vision-language models across diverse tasks, but task-level scores do not reveal which underlying capabilities ac...
arXiv:2610.01962v1 Announce Type: cross Abstract: The ability of vision-language models (VLMs) to associate visual identities with biographical information creates a need for selective unlearning of...
arXiv:2610.02196v1 Announce Type: cross Abstract: We study test-time evolution for humanoid loco-manipulation: solving tasks that a controller was never trained for by repurposing its existing skills...
arXiv:2502.14994v2 Announce Type: replace Abstract: The rapid advancement of AI-generated video poses challenges to digital authenticity and security. Current detection methods, often trained on spec...
arXiv:2602.10639v2 Announce Type: replace Abstract: Video Large Language Models (VideoLLMs) have achieved strong performance on video understanding tasks, yet existing benchmarks evaluate only what m...
arXiv:2606.21562v2 Announce Type: replace Abstract: Transformers are AI's workhorse but their computational cost becomes prohibitive when processing long sequences. We target long-horizon streaming v...
arXiv:2607.09086v2 Announce Type: replace Abstract: We present Subtoken Vision Transformer (SubViT), a selective image tokenization method for fine-grained visual recognition. Standard Vision Transfo...
arXiv:2609.32193v2 Announce Type: replace Abstract: Vision Language Action (VLA) models condition actions directly on current visual and language context, without an explicit account of how the scene...
SkillLens introduces a hierarchical skill-evolution framework that organizes skills into a four-layer graph of policies, strategies, procedures, and primitives, allowing retrieval at mixed granularity. The system first retrieves semantically relevant skill seeds, expands them via a degree‑corrected random walk, and uses a verifier to decide whether to accept, decompose, rewrite, or skip each visited unit. This approach enables agents to reuse compatible subskills while locally adapting mismatched components, and theoretical analysis shows sublinear cost under sparse mismatch assumptions, with empirical results on MuLocbench and ALFWorld demonstrating consistent improvements over strong baselines.
The study evaluates Jev 1.13, a non‑generative model that selects from predefined answer options, on four medical benchmarks: MetaMedQA, PubMedQA, DiagnosisArena‑MCQ, and the NEJM Case Challenges. Jev’s top‑1 accuracy matches GPT‑6 Sol with medium reasoning on PubMedQA but falls behind on MetaMedQA, DiagnosisArena‑MCQ, and NEJM cases. While Jev shows strong calibration on MetaMedQA and is fast and inexpensive, its performance on examination and complex diagnostic tasks is substantially lower, indicating the need for task‑specific validation before clinical deployment.
The paper introduces a lightweight deep learning framework that forecasts Brazilian soybean yields using only routine weather data and two simple static inputs (crop year and agro-environmental label). Across 20 seasons, transformer-based models achieved the highest accuracy, outperforming traditional ridge regression and a moving‑average baseline by nearly 48%. Ablation studies show that the static inputs and spatial expansion improve performance without adding complexity, and SHAP analysis highlights the importance of crop year and weather variables in driving yield variations.
The paper introduces a network‑based spatial context retrieval pipeline that uses pedestrian street networks and open data (OpenStreetMap, GHS‑POP) to generate compact spatial briefs for open‑weight large language models. It then builds a faithfulness benchmark that labels each model claim by its source—whether grounded in the brief or drawn from training knowledge—and tests models against planted false premises across multiple cities and model configurations. The study finds that model family and generation influence resistance to false premises more than model size, revealing dimensions of spatial reasoning not captured by traditional correctness metrics.
Lang3DSeg introduces a point‑transformer backbone for open‑vocabulary, annotation‑free 3D LiDAR segmentation, trained from scratch without geometric pre‑training. It tackles noise from 2D‑to‑3D label projections by applying a class‑priority rule and truncating projected instances at depth gaps, thereby correcting depth‑ambiguity errors. The method achieves state‑of‑the‑art results on nuScenes (52.8 % mIoU) and SemanticKITTI (41.4 % mIoU) while operating in real‑time on a single LiDAR sweep.
OneStreamer is a streaming video model that jointly learns to record evidence and respond to tasks through a shared proactive generation process. Its Proactive Hierarchical Caption Memory creates time‑grounded local‑detail captions and event summaries, while Proactive State Transition Learning reduces waiting states by supervising all output anchors. The authors also built a large OneStreamer‑1M dataset and show that a 4B model outperforms baselines on eight streaming video benchmarks, with ablations confirming the benefits of generated captions and PSTL.
The paper introduces PCSR-Bench, a benchmark of 84,373 question‑answer pairs derived from 2,600 omnidirectional images across 26 indoor environments, designed to evaluate perspective‑conditioned spatial reasoning (PCSR) in multimodal large language models (MLLMs). It reports a significant perception–reasoning gap, with accuracy dropping from 57.59% on limited field‑of‑view reasoning to as low as 0.64% on open‑ended compositional directional chains. An RL‑based diagnostic study on a 7B‑scale model shows that reward shaping can improve performance to 60.06% on a controlled task, indicating partial plasticity of PCSR capabilities.
Dyna‑DINO introduces a curriculum for Vision Transformer (ViT) knowledge distillation that uses the teacher’s intermediate feature maps as progressively harder targets, enabling a student to build foundational representations before tackling higher‑level abstractions. The approach accelerates convergence and improves performance across multiple tasks: on ImageNet‑100 the distilled ViT‑S reaches 90.1% accuracy (+12.24% over baseline), while on ImageNet‑1K it yields +3.9% and +6.09% gains on Oxford and Paris retrieval, +1.93% on semantic segmentation, and notable classification improvements. Additionally, the curriculum reduces training FLOPs by 25.1% and training time by 21% on ImageNet‑100 through early‑stopping of teacher inference.
Build2SPARQL is a large-scale benchmark dataset for translating natural-language questions into SPARQL queries over building knowledge graphs. The dataset is generated by a KG‑grounded pipeline that produces 6,136 executable SPARQL queries and 30,680 corresponding natural-language questions across six query-pattern families and five vocabulary registers, covering 201 building KGs. Human validation shows high semantic fidelity, naturalness, and operational plausibility, and retrieval‑augmented evaluation demonstrates significant accuracy gains for open‑weight language models.