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.
MAPS framework models multi-agent dialogue with subjective perspectives and shared meaning, aiming for diverse, interpretable AI conversations.
Research trains small LLMs to report on internal activation perturbations (activation steering) to detect injected 'thoughts'.
Researchers developed T5-CSBoost, an extension of T5-Sentinel, to improve adversarial perturbation resistance in AI-generated text fingerprinting.
New research introduces CoEvoT, a method for Co-Evolving Chain-of-Thought prompting to improve graph-LLM reasoning, especially under distribution shifts.
Research introduces a new method for cross-version differencing of scientific documents, addressing challenges with heterogeneous elements and layout.
Research paper explores methods for refining LLM-generated research ideas to improve diversity, evaluability, and project execution success.
Research identifies 'semantic register compression' as a failure mode in multi-agent LLM systems, where intermediate agents lose critical semantic distinctions.
Research evaluates cross-dataset generalization for Urdu fake news detection using XLM-RoBERTa, identifying a length confound in current benchmarks.
Research demonstrates a one-line prefill can consistently bypass safety alignments in LLMs across multiple models and families, making them compliant with harmful requests.
Research introduces 'implicit reasoning steering' to bias LLMs towards specific answers without explicit instructions by chaining concepts.
Research argues LLM failures like sycophancy and overconfidence stem from a fundamental lack of awareness of the user beyond the prompt.
Research proposes Multi-Head Latent Control, a unified interface for LLM agents to make decisions like deferring, requesting info, or invoking tools.
New research introduces PReM, a context compression technique for LLMs that dynamically preserves and refreshes useful information during generation.
Researchers at SemEval-2026 Task 11 are addressing LLM limitations in formal reasoning by disentangling logic from content using synthetic data.
MamaBench introduces the first counterfactual benchmark for LLM robustness in maternal and child health diagnosis, using 217 pairs of expert-authored clinical narratives.
DS@GT ARC's LongEval submission evaluates RAG QA systems for citation integrity, using CRAG and CiteFix to correct for divergence in traditional metrics.
Researchers propose KV-cache grafting, allowing frozen small models to restore verified knowledge byte-exact, improving capability and reducing cost.
Research introduces CRTBench, a new benchmark of 350 question families (1,750 questions) to evaluate LLM logical consistency across reformulations.
CityLLM is a new research framework for natural-language querying of semantic 3D city models, combining spatial and graph databases with LLMs.
Research shows fine-tuning LLMs on 'gold answer-conditioned' chains of thought, a common distillation method, degrades verifiable reasoning quality.
Research proposes MARS, a scalable framework for combining LLMs with knowledge graphs (KGs) using multi-hop retrieval and SPARQL generation for grounded answers.
A study on arXiv evaluates LLM-generated written corrective feedback (WCF) across 20,000+ EFL essays, emphasizing extrinsic evaluation for learning fit.
Research on Gemma-3-27B-it reveals first-language (L1) scoring bias in LLM-based automated essay scoring, impacting cross-prompt generalization.
Research finds 'structural priors' (cheatsheets) significantly boost in-domain LLM performance for tasks like code security, but degrade out-of-distribution.
Research paper proposes D-cut, an adaptive method for speculative decoding to reduce LLM inference costs under high concurrency by pruning verification depth.
Research describes 'harness engineering' – deterministic scaffolding around LLMs for reliable deployment in domain decision systems.
Research introduces 'Gold-Guided Programmatic Distillation' to improve LLM accuracy in financial reasoning over hybrid tabular and text data.
Research finds AI detection models frequently misclassify autistic writing as AI-generated, exhibiting bias against minority groups.
The EXACT 2026 competition challenges open-weight 8B parameter models to provide explainable answers and logical reasoning for educational QA.
Research proposes SEED, a method for self-evolving on-policy distillation to improve agentic reinforcement learning for LLMs in long-horizon tasks.
© 2026 OneBench: AI Insights. All rights reserved.
Evidence before opinion