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 paper introduces Semantic Field Theory (SFT) as a computational model for lexical semantics and stabilized interpretation, refining its mathematical core.
Research finds all frontier LLMs exhibit response drift, producing outputs that deviate from expert-validated references, uncharacterised by human evaluation.
RE-AD framework uses LLMs to proactively validate human data labeling quality in real-time, improving annotation accuracy for model training.
Research fine-tunes small language models (0.6B-20B parameters) to translate natural language into MiniZinc, a domain-specific constraint language.
CAMeR introduces a keyword-gated hybrid activation memory framework for LLM agents to selectively retain relevant information over extended dialogues.
New research proposes THOR, a Theta-Gamma hierarchical oscillatory reasoning framework, to improve multi-hop question answering by addressing attention decay and error accumulation.
Instruct-FD introduces an instruction-conditioned benchmark for evaluating controllable turn-taking in full-duplex spoken dialogue systems.
A preliminary research study explores whether valence (emotional tone) in natural language can reflect morality, impacting AI ethics.
Research tested GPT-4.1's ability to predict opinions using persona simulation, accurately forecasting 2024 election outcomes based on U.S. state personas.
TextGrad, a method for optimizing language model text components from natural language feedback, struggles with agentic systems due to delayed feedback attribution.
Research proposes using reinforcement learning to detect user interface (UI) principle violations, including accessibility and poor visual hierarchy, in LLM-generated front-end code.
Research introduces REGARD to study affective framing differences in LLMs, moving beyond simple sentiment to understand regional biases.
Researchers introduced Rushes, a human preference dataset for pluralistic alignment collected from interactive AI-generated narratives.
REFACT proposes an adaptive method for LLMs to restate facts in their chain-of-thought reasoning, aiming for compact, faithful, and context-aligned outputs.
Researchers propose Constrained Shared-Private Fusion (CSPF), a new method for more reliably evaluating non-verifiable AI tasks by integrating diverse reward models.
LegalCiteTrust benchmark evaluates citation trustworthiness in LLM-generated long-form legal research by identifying misrepresentations.
Tencent introduces WorkBuddy Bench, a multi-domain coding-agent evaluation suite with contamination-resistant task construction methodology.
CultureTalk-ID introduces the first dialogue-based benchmark for Indonesian cultural commonsense across 11 local languages, comprising 4,496 dialogues.
Research explores sample-efficient agent learning by combining in-context learning with context distillation to internalize interaction histories.
Research finds that quantizing LLMs, a common practice for efficiency, measurably increases model bias despite standard safety evaluations missing this effect.
PrefReward is a research proposal for a preference-aware reward model designed to improve personalized text generation by explicitly representing user preferences.
RegretBench, a new multi-turn benchmark for LLMs, evaluates clarification policies by modeling ambiguous user requests as sequential decisions.
Researchers introduced progressive cramming, a method to compress sequences into learned embeddings with near-perfect reconstruction by iteratively growing the token budget.
A Greek book retrieval benchmark (CUP) evaluates sparse, dense, hybrid, and LLM-assisted methods for catalog search.
New research proposes a unified dataset to improve instruction tuning for aligning LLMs to specific human values, addressing current data gaps.
Researchers developed news-crawler-LM, a small long-context model for extracting structured content from diverse and complex news page layouts.
GRADRAG is a new research framework for multi-agent RAG that optimizes the entire pipeline by propagating evaluation feedback to upstream agents.
Researchers trained and evaluated phonetic forced alignment models for Chengdu Mandarin, a low-resource language variety, using a 17-hour corpus.
New arXiv research introduces Capital Markets LLM Reliability Score (CM-LRS) for evaluating LLMs on "bankability" in capital markets workflows.
MemTools is a new research framework designed to improve interoperability and decouple components of memory systems in agent architectures.
© 2026 OneBench: AI Insights. All rights reserved.
Evidence before opinion