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.
Research highlights the untapped potential of UMAP's internal k-nearest-neighbor graph for high-dimensional data sensemaking, often overlooked in favor of its 2D projection.
Research proposes a principled framework for selecting machine learning model compression and acceleration techniques across various deployment environments.
Research identifies 'topology overfitting' in GNNs for power grids, where single-task models fail on unseen grid structures despite low in-distribution error.
Researchers introduced a Graph Neural Network foundation model for event classification, trained on 120 million simulated proton-proton collisions.
Research proposes a label-free concept drift detection method for AI models deployed in dynamic environments, enabling MLOps to trigger retraining.
RL-Struct, a lightweight RL framework, uses Gradient Regularized Policy Optimization to improve LLM-generated structured output by 89.7% on JSON tasks.
Research provides a statistical interpretation of Evidential Deep Learning (EDL) for uncertainty-aware classification, addressing its theoretical foundations.
Research on Backward Conformal Prediction (BCP) proposes a non-conformity score transformation to improve coverage guarantees.
Research finds that larger models do not consistently outperform smaller ones under data scarcity, challenging scaling law assumptions.
Research proposes a conformal prediction framework for graph-valued outputs, using Z-Gromov-Wasserstein distance for uncertainty quantification.
Researchers introduced LVSum, a human-annotated benchmark to evaluate multimodal large language models for timestamp-aware long video summarization.
Research identifies AlltoAll dispatch as the primary bottleneck in Mixture-of-Experts (MoE) parallelism and critiques existing mitigations.
RobustSpeechFlow, a new training strategy, improves text-to-speech (TTS) alignment robustness by using length-preserving latent augmentations.
RhinoVLA is a proposed Vision-Language-Action model addressing real-time deployment challenges on edge hardware by reducing VLM visual and context tokens.
Research explores 'Generative Compilation,' using on-the-fly compiler feedback during LLM code generation to improve strict language (e.g., Rust) code quality.
Research proposes unifying diverse electronic health record data (text, vitals, labs) into a single multimodal LLM for clinical prediction.
New research introduces the Jacobian lens technique to identify 'verbalizable' representations in LLMs, analogous to conscious accessibility in humans.
Researchers propose VarRate, a training-free KV cache compression method for long-context LLMs, claiming improved accuracy over existing techniques.
Research introduces EpiNarrate, an agentic system generating grounded public health narratives from complex epidemiological projections.
New research introduces Process Reward Informed Tree Rollout (PRITR), an RL method for LLM agents that optimizes multi-turn exploration by avoiding dead ends.
Research introduces adaptive multi-step lookahead decoding for diffusion language models, enhancing parallel text generation efficiency and accuracy.
Research introduces Bootstrapped Iterative Self-Reasoning Distillation (BISRD) to compress large model reasoning, reducing redundant computation in CoT traces.
Research introduces Prospective Hypothesis Discovery (PHD) benchmark to measure LLMs' ability to generate testable hypotheses from inconclusive evidence.
Research explores using generative AI for summarization to improve automated essay scoring for long texts, using GPT-5 variants on the ASAP 2.0 dataset.
CAMMAR introduces a framework for Arabic LLMs to differentiate between lexical, cultural, and metaphorical meanings, addressing "semantic smearing."
Research presents an MLIR-based compilation method to improve LLM deployment on specialized hardware by addressing model import and inference scheduling.
Research measures how much formal semantic structure explains human label variation in natural language inference datasets, using a rule-based tagger.
Research studies how diffusion language models (DLMs) implement in-context learning (ICL) via an 'induction' mechanism, comparing them to autoregressive transformers.
Research proposes BayesPO, a Bayesian prompt optimization method using parallel-tempered gradient-guided discrete MCMC for LLM adaptation without parameter updates.
Researchers introduce Loopie, a new series of 20B and 6B parameter Mixture-of-Experts models optimized for looped Transformer architectures.
© 2026 OneBench: AI Insights. All rights reserved.
Evidence before opinion