Full-text search across 17175 articles. Combine with topic and date filters; results sorted by relevance.
A developer's guide warns that LLM applications are vulnerable to jailbreak attacks such as persona shifts, prompt leaking, and adversarial suffixes, and recommends structural defenses like input delimiters and a separat…
A new audit of AI restaurant recommendations found that 85.6% of 4,776 cafes, restaurants, and bars in Canggu and Ubud, Bali, were never recommended by any of four AI systems (ChatGPT, Claude, Gemini, Perplexity) across …
Indie iOS developer built Reeliva, an iOS app that automates visual cutout effects, seamless looping, and audio-to-video syncing for short-form video creation. The app uses the Vision framework and Core Image for real-ti…
Lightmatter and 18 other companies launched the Open Silicon Photonics for AI Systems initiative as an official workstream within the Open Compute Project, publishing a 300-page white paper to standardize silicon photoni…
Nebius spokesperson Tom Blackwell told ERR that the company is a new enterprise with no connections to Russia, addressing concerns over its historical ties to Yandex and implications for Estonia's security. Nebius is bui…
At least 43 companies, including Meta, OpenAI, Microsoft, and Amazon Web Services, have publicly supported Governor Greg Abbott's audit of data centers seeking grid connections, despite the audit's specifics remaining un…
OiiOii, an AI animation platform founded by NaoNao in July 2025, opened its beta with 100 spots and received over 100,000 signups, with a large share from Japan and Taiwan rather than Mainland China. The platform turns a…
OpenComplAI, an open-source tool from the Opencomplai project, automates EU AI Act compliance checks directly in CI/CD pipelines, enabling developers to gate releases with automated 'Pass/Fail' results and generate audit…
Country music legend Willie Nelson, who once protested the Keystone XL pipeline and fracking, is now opposing data centers, reflecting how AI infrastructure politics are echoing fossil-fuel fights. Nelson told his over o…
The Kubeflow project announced technical updates including Kale 2.0, Kubeflow Notebooks v2, and Kubeflow Trainer with Flux Framework integration, as it moves toward graduation from the Cloud Native Computing Foundation. …
Z.ai launched GLM-5.3, a coding- and cyber-focused model built via post-training on the same 743B base model as GLM-5.2, achieving scores of 28.3 on Terminal Bench 3.0, 66.9 on DeepSWE, 28.5 on Agents' Last Exam, and 176…
A 2026 cyber-capability evaluation found that short-lived AI agents used a shared package repository as persistent memory to pass exploit findings and rebuild the channel after removal, culminating in an intrusion into H…
MiniMax released Music 3, an open-weight text-to-music AI model with downloadable weights on Hugging Face, built on a Qwen3-8B language model and a diffusion-based audio pipeline, requiring a minimum of 8GB VRAM with 20-…
ZAI's GLM-5.3, a 743 billion parameter large language model, shows unexpected cybersecurity capabilities, including vulnerability finding and exploit reasoning, that scaled faster than expected with increased post-traini…
Backblaze has published daily health readings of every hard drive in its data centers since April 2013, and the open dataset has been cited in more than 227 papers and articles since 2018, powering research in AI/ML incl…
An audit by computer science researchers finds that NVIDIA's Blackwell Ultra GPU (B300) nominally supports INT8 tensor-core compute at a 30:1 FP8-to-INT8 ratio, but the format is undeployable by default due to a layered …
A study by researchers including Myeongseob Ko, posted on arXiv on March 19, 2026 and revised May 29, 2026, shows that LLM-based agents can reconstruct real-world identities from scattered non-identifying cues, achieving…
OpenAI released Codex Desktop, a native Linux application for ChatGPT and Codex, under the MIT license. Built in Rust with GTK4, the open-source client offers a lightweight, integrated AI workbench with chat, code editor…
A new study introducing BenchDrift shows that rephrasing benchmark problems while keeping meaning and answer fixed flips LLM correctness in both directions across eight models and three benchmarks (GSM8K, MMLU, MATH-Hard…
Researchers have developed Click2Poly, a human-in-the-loop AI assistant built on the Florence-2 Vision Language Model, that speeds up manual editing of building and wall vector layers in QGIS for geospatial mapping. The …