Full-text search across 70367 articles. Combine with topic and date filters; results sorted by relevance.
Meta launched Muse Spark Open on Thursday, an open-weight AI model that Meta's chief of AI Alexandr Wang called "the most capable US open-weight model to date" and that Meta claims rivals or outperforms leading Chinese o…
A developer built ScopeKit, an MIT-licensed local AI workbench that uses Gemma 3 1B via Ollama to turn informal client briefs into reviewable scope suggestions, deliverables, hours, and Markdown proposals for freelancers…
Snowflake announced its intent to collaborate with Orange Business as an operating partner in France, pairing Snowflake's AI Data Cloud with Orange Business's managed services on sovereign infrastructure such as Bleu to …
The Institute of Foundation Models at Mohamed bin Zayed University of Artificial Intelligence (MBZUAI) released Jais 3 on October 15th, a six-model Arabic AI family spanning 0.9 billion to 375 billion parameters paired w…
A developer at Munim, Inc. built forecast-answer-formats, the first public cross-tool benchmark comparing which valid LLM answer formats two deployed forecasting-bot extractors read versus silently drop. Testing mv2-fore…
A wave of Model Context Protocol vulnerabilities disclosed in late September and October 2026 exposed remote code execution, server-side request forgery, unauthenticated debugger access and credential theft across both t…
Researchers released GhostFaceMPC, described as the first end-to-end privacy-preserving face recognition system to evaluate a deep convolutional neural network extractor and matching under secure multi-party computation,…
Samsung is demonstrating a CXL memory pool validated with AMD at the OCP Global Summit in San Jose, claiming up to 38 percent lower time to first token and up to 3.3x higher goodput than static allocation for agentic AI …
A benchmark of 14 decision models — hosted Jev and OpenAI's GPT-6 Luna plus twelve local models including the untuned baseline Laya — found that on the classic Trolley Problem switch at five versus one, 11 of the 13 trai…
NVIDIA released Nemotron 3.5 Super VL on October 15th, an open multimodal model for AI agents with 120 billion total parameters and 12 billion active per token that processes text, images, documents and video. NVIDIA sai…
Google announced a preview integration that connects its Gemini AI to the borderless Lakehouse, letting business users query data across clouds with natural language and letting AI agents ground reasoning and generate co…
US authorities charged OpenAI employee Abdulrahman Alfozan, a Saudi national employed at OpenAI since July 2025, with planting a secret audio recording device in an executive meeting room at the company's San Francisco o…
Google said Thursday that Universal Cart, a tool that lets users collect and monitor products found across Search and in the Gemini app, is now available to U.S. users, following its preview at Google I/O in May. The car…
Anthropic co-founder Benjamin Mann said at the company's Singapore office launch on Wednesday that Anthropic will not relinquish oversight of how customers use its most powerful models, citing the technology's "dual use"…
Snowflake announced its intent to partner with Bleu, the French sovereign cloud operated by Orange Business, to bring Snowflake's AI Data Cloud to France's regulated markets on infrastructure that is French-owned, French…
Make0 AI launched a plug-and-play memory layer that connects to any AI agent via MCP, letting agents read and write a shared graph-based "Brain" so context persists across tools like Codex and Claude Code. The developer …
JetBrains' "The State of CI/CD in 2026" report found that 83.4% of respondents who use DevOps practices run CI/CD pipelines, making it the most widely adopted practice in that group, ahead of containerisation and orchest…
A post-training method called Titan-Llama retrofits pre-trained attention-based language models with hierarchical neural memory modules that update parameters at test time, according to a paper posted to arXiv (2610.1921…
A controlled benchmark of nine representative class-incremental learning (CIL) methods across five frozen protein encoders and two enzyme function datasets found that no method transfers reliably across encoders, accordi…
A study of eight prediction methods across three Perturb-seq datasets found that specialized neural predictors do not consistently outperform a simple additive baseline that sums the two observed single-gene effects when…