Leaderboards, eval harnesses and ablations — the contested business of deciding which model is actually better.
Verifiable instruction-following benchmarks emphasize output-constraint satisfaction but can overlook whether responses meaningfully answer the underlying request, creating an evaluation blind spot. W...
Post-training often improves the performance of large audio-language models on audio understanding and reasoning benchmarks. However, these improvements may reflect adaptation to specific question for...
Chest X-Ray (CXR) is a primary modality for diagnosing pulmonary diseases, yet the overlap of bony structures often obscures critical radiographic findings, increasing the risk of diagnostic uncertain...
Reliable recognition of accessibility-critical objects (e.g., audible pedestrian signals, door-activation buttons, and handrails) is essential for assistive navigation technologies that support safe,...
Recent advancements in topic modeling have split into two main paths: efficient embedding-based methods that offer scalability but are often outperformed on standard clustering metrics, and Large Lang...
In so-called "Industry 4.0," integrated sensors, analytics tools, and other digital technologies supplement industrial workflows, potentiating enhanced visibility and automation. Prognostics and healt...
The article announces the release of llm‑gemini 0.35, adding support for the new Gemini 4 Argon model available through the Gemini API and for customers on the $100/month subscription plan. It highlights Argon’s pricing—$2/million input and $10/million output at launch, with a planned price increase on December 1st—and compares its cost-effectiveness to other models, noting it remains cheaper than many competitors even after the increase. The post also showcases example SVG outputs and creative illustrations generated by Argon, such as a pelican riding a bicycle and an armadillo in fishnet tights on Mars.
We present LidarVerse, a real-time unified LiDAR world model for sensor simulation across tasks and sensors. Existing LiDAR simulators largely follow one of two complementary paradigms: reconstruction...
Evaluation of large language models (LLMs) remains largely pre-scientific: it involves many design decisions, with little guidance on how to make them. These choices can substantially affect measured...
Aerial and remote sensing video is increasingly evaluated like web video, by posing multiple-choice questions to multimodal large language models (MLLMs) over sampled frames and reporting accuracy. Th...
Inference of phylogenetic trees from sequences plays a central role in understanding evolutionary relationships, with applications ranging from tracking pathogen spread to reconstructing the history o...
The paper addresses the computational challenge of sampling equilibrium field configurations in lattice field theory by proposing a data‑free training method for diffusion models. Instead of relying on target distribution samples, the authors train the models using their own generated samples and the target score identity, leveraging importance‑weighted and self‑consistent fixed‑point formulations. Experiments on two‑dimensional φ⁴ theory demonstrate that this approach achieves performance comparable to training with HMC reference data across various phases.
The paper demonstrates that Earth observation (EO) models can be automatically constructed and refined by a general-purpose coding agent when provided with a harness (defining permissible tools) and a verifier (defining success criteria). By casting any verifiable EO task—using labeled benchmarks, independent sensors, physical models, or repeated observations—as such a harness-verifier pair, the agent recursively improves its own outputs. On four EO benchmarks covering optical, SAR, and multispectral time-series imagery, the agent-built models outperform state‑of‑the‑art methods while retaining most accuracy with only about 1 % of the training labels.
Complex image editing requires faithful task interpretation and accurate execution, but a failed edit does not reveal which responsibility failed. We present DuplexBridge, which uses a unified model's...
Reinforcement learning (RL) has emerged as the primary paradigm for optimizing Large Language Model (LLM) agents, yet efficient critic-free algorithms (e.g., GRPO) suffer from a fundamental uniform cr...
MultiAlloc is a profiling‑guided framework that allocates limited quantization resources between layer‑wise precision enhancement and token‑wise preservation for KV cache quantization in large language models. By using lightweight offline profiling to assess layer sensitivity and token importance, it automatically derives an effective precision allocation strategy that satisfies a fixed memory budget without online search. Experiments on reasoning and long‑context benchmarks show that MultiAlloc consistently outperforms existing layer‑wise, token‑wise, and multi‑granularity quantization methods across models such as Mistral, Llama, DeepSeek‑Qwen, and QwQ.
Jev-ToM evaluates Theory of Mind (ToM) in typed decision models, comparing five such models and seven generative baselines on 11,117 English text questions from nine ToM benchmarks across six capability groups. The TypeSafe Jev model achieves 73.2% source‑balanced accuracy, ranking first or second in all six capabilities, including leading beliefs, goals/desires/intentions, and emotions. Other typed models show distinct strengths, and resource comparisons suggest Jev could reduce cost and waiting time for repeated ToM selection, while confidence‑based selection and candidate expansion reveal task‑dependent challenges and opportunities for improvement.
Beyond Answer Voting: Fine-Grained Correction of LLM Reasoning through Graph Alignment across Sampled Traces proposes DDRG, a framework that uses graph alignment to detect claim-level disagreements in reasoning traces and selectively corrects them with a verifier. DDRG builds source‑grounded claim graphs, sends localized conflicts to an answer‑blind verifier, and applies a focused repair only when the verifier supports the majority answer’s path. Experiments on five benchmarks show that zero‑shot DDRG improves macro‑average accuracy by 8.65–15.83 percentage points over direct answering and outperforms CoT, GoT, GoV, and SC on this metric.
Language models are increasingly expected to support users in stressful or uncertain situations, where warmth and empathy are desirable. While recent work associates warmer models with reduced reliabi...
ProTok introduces a unified protein tokenizer that maps protein sequence, structure, and functional annotations into a single modular discrete vocabulary using product quantization with per-modality codebook groups and modality dropout. The tokenizer supports flexible inference from any subset of modalities without retraining and achieves top performance on seven StructTokenBench benchmarks, with high reconstruction accuracy (0.939 lDDT on CATH 4.2) and superior zero‑shot functional retrieval. Additionally, a compact language model trained on ProTok tokens can jointly generate backbone and sequence, yielding a higher proportion of designable backbones compared to a GeoBPE baseline.