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.
Research measures how much formal semantic structure explains human label variation in natural language inference datasets, using a rule-based tagger.
Research presents an MLIR-based compilation method to improve LLM deployment on specialized hardware by addressing model import and inference scheduling.
Research studies how diffusion language models (DLMs) implement in-context learning (ICL) via an 'induction' mechanism, comparing them to autoregressive transformers.
Research demonstrates that malicious chain-of-thought (CoT) traces can be transferred between LLMs, inducing harmful behavior in target models.
Research redefines empathy in AI as predictive misalignment tolerance and proposes a co-regulation framework for dialogue repair.
CoWeaver is an arXiv research paper proposing a bidirectional, learnable, and explainable algorithm for matching human scientists and AI agents for collaboration.
Research explored moral biases (Knobe effect) in finetuned LLMs using mechanistic interpretability and Layer-Patching across three open-weight models.
Research finds LLM-as-a-judge scores are unreliable for optimizing closed-loop table recognition, showing weak signals and non-reproducible rankings.
Research uses linear probing and Bloom's Taxonomy to mechanistically interpret cognitive complexity in LLMs via internal neural representations.
Research introduces Mixture-of-Neighbors Induction Memory, enabling semiparametric language models to learn and update non-parametric memory dynamically.
Frontier LLMs struggle with exact string copying due to Transformer positional encodings, indicating limitations in fundamental operations.
ToolSciVer is a new research model that uses visual tool-augmented reinforcement learning for multimodal scientific claim verification.
Researchers demonstrated Latent Fusion Jailbreak (LFJ), a white-box attack manipulating LLM internal representations to bypass safety alignments.
Researchers propose VCG-Bench, a new benchmark and "Diagram-as-Code" paradigm for Vision-Language Models to improve structured diagram generation and editing.
KDFlow is a new knowledge distillation framework claiming improved efficiency for compressing large language models into smaller, more efficient student models.
Research explores 'speculative decoding with a speculative vocabulary' to accelerate large language model inference using a smaller draft model.
PrimeFacts is a new methodology and resource for extracting fine-grained evidence from fact-checking articles for automated verification systems.
Research demonstrates that large language model (LLM) alignment guardrails can be broken with one-shot Group Relative Policy Optimization (GRPO) using a single biased example.
Research analyzes how deep transformers, dominant in language modeling, form hierarchical representations across layers for expressive power.
AgentRedBench is a new benchmark for evaluating indirect prompt injection threats in LLM agents using SaaS integrations, addressing gaps in current methods.
Research paper identifies 'AI-Fiction Paradox': LLMs trained on fiction data struggle to generate compelling long-form fiction.
AuAu is a new research benchmark designed to assess and quantify authoritarian tendencies and alignment in large language models using psychometric and vignette-based tests.
Crayotter is an open-source multimodal multi-agent system for prompt-driven long-form video editing, enabling traceable intermediate states.
Research proposes Contrastive Hypothesis Retrieval to improve RAG for medical Q&A by explicitly ruling out hard negatives semantically close but clinically distinct.
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.
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 SkillCorpus, an effort to consolidate and evaluate fragmented open-source LLM agent skills (SKILL.md files) to assess practical value.
New research introduces the Jacobian lens technique to identify 'verbalizable' representations in LLMs, analogous to conscious accessibility in humans.
Research introduces Bootstrapped Iterative Self-Reasoning Distillation (BISRD) to compress large model reasoning, reducing redundant computation in CoT traces.
Research proposes unifying diverse electronic health record data (text, vitals, labs) into a single multimodal LLM for clinical prediction.
Research introduces adaptive multi-step lookahead decoding for diffusion language models, enhancing parallel text generation efficiency and accuracy.
Kai-Fu Lee's AI startup 01.ai plans to raise funds before a Hong Kong IPO in 2027, delaying its public offering timeline.
Expedia CEO Ariane Gorin defends human touch in trip planning against AI, differentiating from tech start-ups by focusing on customer experience.
Massive investment in memory chip production, driven by AI demand, is raising investor fears of a new boom and bust cycle.
Apple ML Research introduces Length Value Model (LenVM), a token-level framework to model remaining generation length in autoregressive models.
Debate emerges on whether Apple's lawsuit against OpenAI will impact OpenAI's hardware ambitions and potential IPO plans.
Hugging Face reportedly used Chinese LLMs for incident response after proprietary US models were blocked from assisting its blue team.
Christopher Nolan, director of 'Odyssey', labeled AI an "obvious 'Trojan horse'" in a recent statement, implying hidden risks.
An industry analyst suggests the high costs of AI model development and deployment will soon become a significant financial burden.
AWS admits its internal billing error alarms failed to alert engineers, leading to customer escalations a day later.
Netflix CPTO Elizabeth Stone discusses the shift to systems thinking over specialization for AI-era product teams and building organizational excellence.
Alibaba's Qwen3.8 Max model, described as second to Anthropic's Fable 5, drove a 5.4% share increase for the company.
A report indicates that established defense contractors will retain an 80% market share despite growth in drone technology.
Many generative AI pilots fail due to poor data quality, not model limitations, as enterprises mistakenly use RAG to fix bad data.
Author Dave Eggers told OpenAI staff that ChatGPT was 'silencing an entire generation,' raising concerns about AI's impact on creative work.
Chinese company Moonshot AI released a new version of its Kimi model, raising concerns about potential implications for AI development.
Research paper explores methods for controlling the level of reasoning effort LLMs apply to tasks, identifying distinct 'modes'.
Moonshot AI's Kimi Chat demonstrates unexpected long-context capabilities, challenging prior assumptions of China's AI lag behind the US.
Report details privacy issues with period tracking apps and other digital security incidents, including Russian cyber-espionage and an AI music generator data breach.
Delivery apps like DoorDash and Uber Eats are eroding the market share of large corporate pizza chains by leveling the playing field for smaller competitors.
© 2026 OneBench: AI Insights. All rights reserved.
Evidence before opinion