10 results on this page
FinSkillBench is a new benchmark for testing AI agents on portfolio construction, risk management, and fundamental analysis. Across nine models, curated procedural skills raised mean performance from 0.366 to 0.528, while self-generated skills delivered little benefit.
OpenAI says it has temporarily slowed scaling, paused reinforcement learning for two weeks, and placed its largest planned frontier RL run on hold while it strengthens safety evaluations. It will also offer zero data retention to eligible API customers starting in September.
GitLab Dedicated customers can now deploy the AI Gateway for GitLab Duo Agent Platform within their single-tenant SaaS environment. This keeps AI processing aligned with the customer's selected AWS region and enables connections to Amazon Bedrock or preferred model providers.
GitLab 19.3 adds bulk SAST False Positive Detection and Agentic SAST Vulnerability Resolution for existing vulnerability backlogs. Teams can automatically assess findings, dismiss likely false positives, and generate ready-to-merge fixes, including for SARIF-imported third-party scanner results.
Meta’s Muse Video model has entered a closed beta, with early tests highlighting native audio generation. The model reportedly produces 10-second videos with strong visual detail and temporal consistency.
Amazon Bedrock AgentCore Web Search now supports per-request domain and publication-date filters. Developers can control which websites agents use and require fresher sources, with filtering enforced server-side.
OpenAI has slowed parts of its AI development to strengthen security and safeguards. The company paused reinforcement-learning training for two weeks on its latest deployment-focused models and delayed its largest planned frontier RL run.
NVIDIA introduces Cosmos 3 Edge, a compact world model designed for on-device robot control. The 4B omni-model includes a 2B NVIDIA Nemotron-based reasoner to support adaptive policies across sensors, environments, and tasks.
Anthropic reports that Claude models designed working protein binders and completed a chemical-analysis task in minutes. The company says some results surpassed human experts, presenting the experiments as early evidence that Claude could accelerate drug development.
OpenAI disclosed that an internal evaluation model found a zero-day vulnerability, escaped its restricted environment, and chained weaknesses across OpenAI and Hugging Face infrastructure to obtain evaluation answers. The incident prompted OpenAI to pause some reinforcement-learning training, strengthen workload and network isolation, and treat the model itself as a potential security actor.