Search signals, briefings, benchmarks and glossary terms.
Search signals, briefings, 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 demonstrates 'LeakyLMs' attacks can infer proprietary model architecture and inference optimizations from production LLMs via token timing analysis.
Research proposes a representation learning technique using joint contrastive and classification learning to improve cross-domain generalization in optical networks.
Research introduces Pipelined Gradient Coding, an improved method for distributed machine learning training that reduces straggler worker impact.
New benchmark, Frontier Financial Judgement, assesses AI agents' ability to replicate expert equity analyst judgments on stock-moving information.
Adaptive Depth Sparse Framework (AdaDSF) proposes converting off-the-shelf pre-trained LLMs into depth-sparse models to reduce inference cost.
LeanFlow is an LLM agent system designed to translate mathematical papers into formal, buildable Lean projects, evaluating runtime mechanisms.
MiniCache is a reusable program caching framework that converts Program-of-Thought (PoT) programs into parameterized cache objects for efficient LLM inference.
Research finds Reinforcement Learning with Verifiable Rewards (RLVR) can improve one-sample accuracy while decreasing distinct problem-solving ability.
ConfidenceBench evaluates 15 frontier LLMs on verbalized confidence calibration using the Brier score, addressing risks of fluent but incorrect answers.
AppWorld-UL is a new benchmark for evaluating agent-user interactions in tool-use agents, addressing diversity and environment scale limitations.
New research introduces DFAH-Bench, a replay benchmark designed to measure observable behavioral instability in financial agent decision-making processes.
Research finds combining regex filters with LLM alignment provides no measurable safety coverage gain against adversarial input, even when designed to bypass regex.
Research introduces an attention-based experience replay framework for continual learning in time series forecasting models, addressing data distribution shifts.
Researchers developed a neural-symbolic pipeline combining compact transformers with a linguistically-informed mediator for gaming toxicity detection.
Research details MuonQ, an optimization method to improve low-bit quantization for the Muon optimizer in large language models by preserving directional fidelity.
Research proposes sparse autoencoders (SAEs) to create interpretable embeddings for analyzing large text corpora, addressing limitations of LLM-based and dense embedding methods.
DC-Leap proposes a training-free framework to accelerate Diffusion Large Language Models (dLLMs) by addressing conservative confidence thresholds.
Research explores the scaling behavior and generalization of deep parameterized quantum circuits (PQCs), a key challenge in quantum machine learning.
SPECTRA is a new probabilistic forecasting architecture designed for complex energy systems, integrating multi-scale decomposition and exogenous variables.
Research proposes Generative Bayesian Filtering, a new method for state estimation in dynamic systems with complex observation models.
SonicSampler presents unified tile-aware kernels for LLM sampling and speculative verification, improving efficiency for dynamic serving workloads.
South Korean companies are investing heavily in US tech acquisitions to enhance AI capabilities and mitigate future tariff risks.
Educators report struggles with AI integration, including concerns about cognitive atrophy and faulty AI cheating detection, prompting calls for 'eval' skills.
The Financial Times discusses frameworks to promote 'deep thinking' with AI in education, counteracting 'cognitive atrophy'.
Universities are discontinuing AI detection tools due to persistent accuracy concerns, shifting assessment focus away from surveillance.
Financial Times op-ed argues universities must equip students with AI model evaluation skills to meet future organizational needs.
Anthropic and OpenAI are offering free and discounted tailored AI solutions to the education sector, signaling direct market engagement.
Universities are developing varied strategies for AI integration, facing challenges related to costs and other policy considerations.
AI-powered schools claim excellence, but long-term effects on learning are unknown, and new inequalities are emerging.
Academic research indicates a significant increase in AI-related research volume, accompanied by a decline in overall quality, prompting concerns.
SK Hynix and Samsung face crucial earnings as memory chip demand driven by AI swings, impacting South Korea's stock market.
OpenAI and Anthropic guardrails are reportedly hindering offensive cybersecurity researchers' ability to identify and exploit vulnerabilities.
Meta secured $12 billion in data center financing, led by BlackRock, facing increased borrowing costs due to investor anxiety over AI exposure.
A top environmental fund is increasing exposure to Japan's tech sector, anticipating its role in addressing AI's growing energy consumption.
Intel's revenue grew 25% in Q2, marking its fastest growth in 15 years, driven by demand for AI data center infrastructure.
Apple ML Research identifies "no-recovery bottleneck" in LLM long-horizon reasoning due to non-uniform error distribution in sub-steps.
AWS details best practices for configuring Bedrock Guardrails to enhance safety and capacity planning for code generation workflows.
CB Insights reports on Q2's major AI venture capital deals, highlighting mega-rounds and new AI unicorns.
Nvidia signed a $1.5 billion agreement with Amkor Technology to boost chip packaging facilities, expanding US semiconductor operations.
China's Xi Jinping is engaging in AI diplomacy to attract Global South countries, aiming to establish an alternative global order.
Singapore's GIC anticipates Chinese AI models will significantly reduce adoption costs, projecting strong growth from Chinese AI firms.
Bonds of major US tech companies declined due to investor concerns over AI-related debt and Middle East conflict.
OpenAI's ChatGPT Business and Enterprise users reportedly lack native chat export options, leading to reliance on third-party tools.
AMD is launching its new Helios AI rack-scale system, directly challenging Nvidia in the AI chip and infrastructure market.
Intel's revenue forecast exceeded analyst estimates, driven by increased data center spending, signaling a market rebound for the chipmaker.
Alphabet's stake in AI startup Anthropic has increased to an estimated $124 billion, reflecting a significant valuation jump.
The FCA's new insurance director published guidance on managing conflicts of interest in vertically integrated insurance business models.
Donald Trump expanded his pledge for power utilities and data center developers to make technology companies pay for AI energy costs.
The FDA implemented an AI platform achieving 85% daily staff adoption by focusing on foundational data infrastructure before specific AI applications.
Anthropic updated Claude's voice mode with more capable models, enabling tasks like meeting rescheduling and email drafting.
© 2026 OneBench: AI Insights. All rights reserved.
Evidence before opinion