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.
NVIDIA Nemotron 3.5 ASR 0.6B was adapted for Kikuyu, Dholuo, and Kalenjin languages, demonstrating data-centric methods for low-resource ASR.
Research explores small language models (SLMs) for creative plot generation, focusing on global coherence, character, pacing, and emotional progression.
Research proposes RLAES, an LLM framework using reinforcement learning with rubric rewards for automated essay scoring and feedback generation.
Research paper discusses agentic LLM systems moving from prototypes to production across software, science, and finance, highlighting deployment challenges.
Research explores using depthwise convolutions within Transformer blocks to improve local inductive bias in LLMs, comparing placements across Qwen3.
New research proposes Causal Alignment and Structural Enforcement (CASE) to improve chain-of-thought faithfulness in LLMs, ensuring reasoning supports answers.
Research finds reasoning fine-tuning induces persistent, global internal changes in LLMs for multi-step reasoning, not just local token competence.
AFIR, a Romanian government agency, deployed RAGAL, a fully local RAG-based assistant for technical support, adhering to zero data egress.
Research identifies and quantifies "post-hoc rationalization" in Reverse Chain-of-Thought (RCG) LLM generation, where models justify a pre-committed answer.
ScienceMeter is a new framework introduced by arXiv for evaluating how well LLMs update their scientific knowledge over time.
Research explores machine unlearning for vision-language models (VLMs), noting that language backbone unlearning doesn't guarantee VLM unlearning due to visual data influence.
A new survey on arXiv details the evolution of AI for mathematical reasoning, from rule-based solvers to contemporary LLM-driven approaches.
Research explores using translated data to augment scarce expert-annotated corpora for text difficulty assessment, particularly for lower-resource languages.
Research introduces FiT, a diagnostic method to select small LLMs for fine-tuning in critical QA tasks like cybersecurity, before costly adaptation.
Researchers propose HindsightBench, a black-box audit protocol to detect parametric hindsight (leakage of future knowledge) in LLMs used for time-indexed decision tasks.
Research paper proposes a new method for benchmarking LLMs on culturally specific affective responses, moving beyond declarative knowledge.
Research introduces P-StaT, a framework to evaluate LLM belief stability under semantic perturbations, finding familiarity impacts stability.
AutoJourn is a research system demonstrating multi-perspective summarization, bias detection, and bias neutralization for LLM-generated news.
TReB is a new benchmark to evaluate large language models' (LLMs) table reasoning capabilities, addressing gaps in current evaluations.
Research explores enhancing legal machine translation using reasoning-capable language models to improve precision and address linguistic complexity.
AdaFlash introduces adaptive speculative decoding using diffusion drafters to accelerate large language model inference by generating drafts in a single pass.
Research proposes a new method using decoding-constrained beam search to quantify near-verbatim data extraction risk in LLMs, beyond simple memorization.
MUX proposes distilling discrete LLM reasoning steps into continuous multiplexed tokens for more efficient, high-bandwidth computation.
Research introduces MaLoRA, a dynamic low-rank adaptation method for LLMs, enabling token and instance-level state adaptation during inference.
Research explores self-evolution of LLM dialogue skills using future-feedback prediction to address unstable validation signals in open-ended conversations.
DAIS (Dependency-Aware Intermediate QA Supervision) is a new training framework converting teacher rationales into stage-level QA records to improve complex reasoning.
Research on multi-step, tool-augmented LLM agents identifies 'binding drift' where agents silently misattribute entities across sequential actions.
Research explores cost-quality tradeoffs in Reinforcement Learning with Verifiable Rewards (RLVR) for Neural Machine Translation, particularly for legal documents.
Research evaluates multimodal LLMs' theory-of-mind (ToM) reasoning in multi-party meetings, identifying current limitations beyond overt signals.
New benchmark, GAMUT, focuses on evaluating the factual completeness of open-ended LLM generations, addressing the gap beyond factual precision.
Research investigates mitigating cross-lingual factual inconsistency in LLMs at inference time, addressing biases toward high-resource languages.
MIRA-Ev is a new clinical NLP benchmark for granular evidence detection and relational reasoning, using Spanish medical licensing exams.
New research proposes PINT, an invariant speech tokenization method that disentangles semantic content from non-linguistic speech variations.
Research explores using 'semantic primes' as non-circular explanations for emotion mechanisms in large language models, addressing current limitations.
Research proposes a new taxonomy for curriculum learning in NLP, disentangling difficulty evaluation from training scheduling to improve analysis.
Research identifies a 'hidden auto-regressive risk regime' where large language models' answers degrade faster, compounding mistakes.
Research identifies 'Safety Drift' and 'Operational Hallucination' in multi-turn LLM agent interactions, degrading initial alignment over time.
Research explores using LLMs for multilingual privacy policy analysis, extending beyond English-centric systems to improve transparency audits.
Research proposes "token inoculation" to safely retain dual-use knowledge in LLMs, gating hazardous content with control tokens instead of erasing it.
Researchers introduced MedDDC-Eval, a diagnosis-decoupled evaluation framework for multi-turn medical consultation agents to isolate policy elicitation from diagnosis generation.
Research indicates transformers without feed-forward networks (attention-only) can match or exceed standard transformers in performance when parameters are controlled.
AgentDebugX is an open-source toolkit designed to improve failure observability, attribution, and recovery for LLM agents by organizing debugging as a closed loop.
LatentMT explores latent-reasoning LoopLMs for machine translation, adapting a small 2.6B-parameter model with recurrent computation.
AutoIndex is a framework for learning representation programs that transform raw documents for retrieval systems, optimizing document preprocessing.
Research proposes Multi-Task On-Policy Distillation (MTOPD) using soft prompts for LLM self-distillation, avoiding weight drift and input post-hoc rationalization.
Australia's Northern Territory approved a major data center project projected to attract A$40 billion ($28 billion) in private investment.
Expert commentary notes AI cybersecurity is a growing concern due to recent cyber headlines, indicating an emerging trend.
Rumors of aggressive acquisition sprees by Anthropic and OpenAI circulated on social media over the weekend, hinting at M&A in the AI sector.
Sony and Mitsubishi Electric are forming a joint venture to provide factory automation services using AI and image sensors.
China's AI models are rapidly gaining global competitiveness, challenging US containment strategies and raising alarms in Washington.
© 2026 OneBench: AI Insights. All rights reserved.
Evidence before opinion