Topic
Explore the latest artificial intelligence news and research.
San Mateo County has moved forward a permit plan for autonomous robots. We compare its three worker-protection options—reassignment, severance, or a fund contribution—and explain where the ordinance stands ahead of final adoption.
Internal remarks by a Microsoft executive have surfaced in the NYT copyright lawsuit against OpenAI and Microsoft. This article examines the click-through decline cited by plaintiffs and unpacks the tension between news revenue and AI's reliance on that same content.
AWS has streamlined sign-up and initial setup for new users, making it easier to build with AI and collaborate. But spend limits have a minimum tied to projected usage, and projects stop when the limit is reached, so users need to judge whether that fits their use.
A fact-check of OpenAI's Noam Brown and his remarks on "thermal communication in air-gapped environments." We contrast the physical limits shown in the 2015 BitWhisper paper—just 1 to 8 bits per hour—with the actual vulnerability exploited in the July 2026 Hugging Face breach, separating theoretical risk from real-world perimeter defense.
A researcher using Claude combined an image-processing flaw with a shared authentication vulnerability to reach OpenAI's internal repository. Here's what the incident means, based on the fix rollout, the scope of linked-account permissions, and the disclosed time and cost figures.
DRAM maker CXMT is reportedly planning a NAND research and development line in Beijing and talking with potential customers, according to Reuters. Here is what the report confirms, and what remains unknown before any commercial supply.
OpenAI has announced Astra for Law, which pairs legal search with specialized instructions. Its pass rate on US legal research questions rose to 54.0%. Here are the comparison conditions, the initial US-firm rollout, and how data handling differs between the API and ChatGPT Enterprise.
Google's Android Bench 2.0 introduces long-horizon tasks of up to 8,200 lines and near-real-device verification. Even the top performer's 28.0% pass rate versus 82.2% weighted completion rate reveals the wall AI coding hits when it comes to fully finishing the work.
NVIDIA, Google and Emerald AI have formed the AI Energy Management Alliance (AEMA). We weigh small-scale pilots, a 155,410-GPU operating study and FERC's reforms to ask what it takes to turn AI data center load reduction into a contractable power product.
A close look at NVIDIA's CUDA-Q Logical based on its papers and official documentation. Examining what the P0–P4 stages, the four-tier resource estimates, QUOPS, and the 150,000-qubit figure mean—and where their limits lie.
