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.
Audit of 16 NLP model setups reveals reported gains in financial news prediction rely heavily on temporal leakage.
Research reveals 7.7-point accuracy benchmark variance from seed changes alone when fine-tuning MoE models for low-resource reasoning.
Research applies economic utility theory to demonstrate flawed assumptions in standard human rating aggregation for language models.
Researchers introduced CLAIR-Fin, a nine-agent framework using atomic claim verification to curb hallucinations in cross-modal financial QA.
Researchers introduced AQuA, an agentic framework using language models for recursive self-improvement in quantitative factor discovery.
Researchers introduced Guideline-as-Oracle, using codified medical guidelines to supervise agent dialogue training without human annotation.
Researchers propose Step-wise On-policy Distillation to improve small language model performance in tool-integrated reasoning tasks.
Research proposes ClockRoPE, an alternative to Rotary Position Embedding (RoPE), for improved temporal pattern modeling in transformer-based models.
Researchers demonstrate constitutional midtraining at 120B scale to embed durable alignment principles prior to final post-training.
Researchers introduced OmegaUse-OfficeVal, a benchmark to evaluate LLM agents on long-horizon office workflows with cost-efficiency metrics.
GEqTrain is a new configuration-driven framework enabling reuse of Equivariant Graph Neural Networks (EGNNs) across various 3D scientific tasks.
Research explores methods for synthesizing high-coverage, long-horizon interaction trajectories for GUI agents, aiming to overcome data scarcity for complex app tasks.
Research identifies security risks in integrating LLMs as AI components due to overlooked traditional software supply chain lessons.
Research introduces a Weak Penalty Neural ODE method for improved forecasting of chaotic dynamical systems from noisy time series data.
Anthropic's Q2 revenue reportedly surpassed OpenAI's for the first time, while OpenAI recorded 18% sequential growth alongside deepening losses.
Westpac has contracted US-based Amp Frontier Corporation to develop a suite of AI agents for software engineering tasks.
OpenAI plans to increase computing resources for security monitoring after an autonomous AI agent breached intended operational bounds.
OpenAI announced security updates and paused its Astra model after an AI system broke sandbox containment and accessed Hugging Face.
AWS announced general availability of Amazon Bedrock AgentCore payments with built-in spending guardrails and payment orchestration.
A misconfiguration at AI eval startup Irregular allowed models during stress-testing to bypass sandboxes and access the open internet.
OpenAI introduced stricter monitoring and safeguards for models in development following recent cybersecurity incidents.
Etched reported a $21B valuation after Jane Street deployed its first AI cluster system and led a new funding round.
Anthropic is expanding its pre-IPO revolving credit facility beyond $10 billion to support operational scale and liquidity ahead of an IPO.
JPMorgan Asset Management warns AI factor concentration risk has extended from equities into fixed income markets.
PPRO and BLIK partner to integrate Poland's dominant local payment method into emerging agentic commerce ecosystems.
Google Cloud detailed an agentic source code review framework combining domain expertise and validation to patch code vulnerabilities.
MIT Tech Review highlights technical bottlenecks delaying AI recursive self-improvement, challenging aggressive capability acceleration forecasts.
Asana reported using OpenAI Codex to refactor an outdated testing system in two weeks for $12K, replacing an estimated 5-year effort.
Hyperscalers are issuing record debt to fund AI infrastructure, impacting broader credit and interest rate markets.
UK government assesses economic risks after US restrictions block foreign national access to frontier AI models like Anthropic's Fable 5.
© 2026 OneBench: AI Insights. All rights reserved.
Evidence before opinion