Search signals, briefings, company results, benchmarks and glossary terms.
Search signals, briefings, company results, benchmarks and glossary terms.
Use this view to inspect the underlying evidence corpus. For ranked developments, decision posture and interpretation, use Signals.
Raw feed or Signals?
Raw feed is chronological evidence. Signals ranks and interprets material change.
Researchers propose Feature-Informed Diffusion Evolution (FIDE), a black-box framework for inverse rendering, bypassing differentiable renderers and gradient descent.
Research proposes Segment-Wise CoT Compression with Answer Alignment (SCA) to reduce LLM inference costs without compromising answer quality.
Research paper details an interpretable machine learning pipeline using XGBoost and TreeSHAP to decompose equity return predictability in Chinese A-shares.
Research explores safe adaptive control, enabling online learning with guaranteed safety across varied initial models while matching a safe oracle's performance.
FAIR-Calib proposes a Post-Training Quantization (PTQ) method to mitigate errors in Diffusion Large Language Models (dLLMs) caused by early, fragile token decisions.
New research explores how data selection and objective design in off-policy distillation influence large language model capabilities and performance.
Research introduces RGMR, an inference-time framework adapting pre-trained foundation models for multi-scale temporal analysis and iterative refinement.
Research introduces ATLAS, a multi-environment factor model to identify invariant and transferable latent factors across heterogeneous data environments.
Researchers developed a Bayesian framework, "diffusion-within-Gibbs sampling," for improved signal component decomposition in noisy data.
Research proposes Precision-Varying Prediction (PVP) to robustify ASR systems against adversarial attacks by randomly sampling precision during inference.
Research introduces a Stochastic Dimension Zeroth-Order Estimator to improve memory and training efficiency for Physics-Informed Neural Networks (PINNs).
A new arXiv survey systematically reviews Graph Neural Network (GNN) architectures for link prediction across diverse graph structures.
Research explores Symmetric Behavior Regularized Policy Optimization (BRPO), finding it can outperform asymmetric BRPO in offline reinforcement learning.
New research proposes a method for sparse-support uncertainty quantification, addressing issues of overconfident conclusions when dictionaries are learned from latent mixtures.
Research introduces Jacobian-Aggregated Group Gradient (AGG) to reduce computational cost for Group Relative Policy Optimization (GRPO) in diffusion models.
Research reveals gradient-free privacy leakage in federated language models (FLMs) through selective weight tampering, exposing sensitive client data.
Research explores optimizing transformer architectures for specific datasets to understand optimal task-specific inductive biases beyond current scaling methods.
Research introduces Time-Aware Prior Fitted Networks for zero-shot time series forecasting, leveraging exogenous variables to improve accuracy.
Research identifies fundamental failures in marginal influence-based attribution methods for explaining global time series models due to computational mismatches.
Research proposes Supervised Moral Rationale Attention (SMRA), a self-explaining hate speech detection framework using moral rationales for attention alignment.
Research addresses model collapse in iterative instruction tuning when training LLMs on synthetic data, proposing methods to ensure performance improvement.
RuBench 1.0 is a new benchmark for evaluating agentic coding models using 25 real-world repository-level tasks with native Russian specifications.
Research decomposes the net gain from test-time collaboration methods (e.g., self-consistency, best-of-N, verifiers) for LLMs into measurable factors.
Research explores using progressive disclosure via 'Agent Skills' for long-document Q&A, allowing agents to dynamically read relevant document parts.
DeLIVeR, a new framework, uses a Planner LLM to decompose complex claims for fact-checking via reinforced knowledge graph exploration.
Research paper proposes "Hallucination-Aware Layered Oversight" (HALO) for enterprise AI, arguing against waiting for hallucination-free models.
Research introduces Scope3Trace, an evidence-based method using LLMs to identify and extract Scope 3 GHG emissions from sustainability reports, improving traceability.
A new benchmark dataset and method for disambiguating culturally entangled Bangla homographs in low-resource LLMs was introduced on arXiv.
New research proposes A-MESS, a defender-centric, Shapley-based evaluation for jailbreak attacks to improve LLM safety alignment through red-teaming.
Research finds open-weight LLMs sometimes commit to an answer and then generate reasoning, even if it contradicts the premise, showing 'pre-commitment'.
Researchers propose a dependency-aware framework for automated code generation, aiming to improve logical completeness and integration.
DocOCR-Eval proposes a framework for selecting OCR tools and MLLMs for document parsing without requiring ground truth data for evaluation.
Researchers propose Trace-Based On-Policy Distillation (TOPD), a new teacher-supervised framework for reasoning-oriented post-training of diffusion LLMs.
ReMIND proposes a four-stage modular LLM framework to generate novel and coherent ideas, balancing originality with consistency.
Research describes a human-in-the-loop AI agent that drafts translational impact summaries, reducing staff time from 15 hours to 1.
Research proposes method to improve answer extraction and generation in context-based question answering systems using LLMs, addressing ambiguity and consistency.
New research identifies 'salience induction' as a third attack vector for Multi-Hop RAG agents, exploiting fact position and framing.
New research introduces LogicIFGen and LogicIFEval to assess LLM performance on complex, logic-rich instruction following, identifying current limitations.
New RL policy gradient method aims to improve long-horizon language agent training by better attributing actions to outcomes, reducing optimization variance.
Research introduces Persistent Sparse Autoencoders (SAEs) that learn feature persistence across language model sequences, improving decomposition.
Research explores methods to safeguard facial identities against unauthorized manipulation by unified multimodal image editing models using 'cross-branch conflict'.
Research uses LLMs to identify social biases against homelessness in online text and city council discourse, highlighting systemic issues.
Researchers introduced ChipChat, a low-latency cascaded conversational agent architecture in MLX, aiming for real-time on-device voice agents.
SWE-Pruner Pro, a new method for LLM coding agents, prunes tool outputs by leveraging the agent's internal relevance representations.
New research proposes Token-Level Off-Policy Learning (TOPL) to improve faithful generation of LLMs by training models to distinguish correct tokens.
Research investigates how shared subword vocabularies in multilingual LLMs handle cross-lingual homographs and false friends, identifying limitations.
Research demonstrates multilingual sentence embeddings can effectively replace translation for Linguistic-Integrated Reliability Auditing across 11 assessment items.
TalTech developed a system using fine-tuned open-weight speech LLMs for direct SOAP note generation from doctor-patient conversations without transcription.
Research finds LLM code generation performance varies significantly based on prompt personas, with model-dependent effects observed.
AEGIS is a new research framework exploring span-guided multilingual detoxification for LLMs across English, Mandarin Chinese, and Korean.
© 2026 OneBench: AI Insights. All rights reserved.
Evidence before opinion