AI agents develop physical model from quantum-material observations
Researchers report in an arXiv preprint that their AI Theorist system autonomously developed a physical model explaining previously unpublished experimental observations in α-RuCl₃, a candidate material for realizing a Kitaev quan...
Proofpoint links TA419 phishing campaigns to US AI policy targets
Proofpoint has attributed credential phishing campaigns against U.S. AI experts at think tanks, universities and legal organizations to TA419, a group it describes as China-aligned and motivated by espionage. Its analysis describe...
Strata says it runs a 125-billion-parameter AI model on gaming PCs
The open-source Strata project says it runs Qwen3.8-Flash-Next, a 125-billion-parameter AI model, locally on Windows or Linux PCs with NVIDIA or AMD graphics cards containing at least 12 GB of memory. It supports chat, coding and ...
Researchers use brain signals to distinguish and correct AI errors in simulations
Researchers at KAIST and Microsoft Research Asia demonstrated a system that uses brain signals to distinguish whether an AI has misunderstood a person's goal or chosen an unexpected method. Their Neural Value Alignment technique f...
Extracted Muse instructions reveal hourly profiling of personal relationships
Meta's Muse assistant has instructions to compile pages about people in a user's life through an hourly process, WIRED reports. Security researcher Karan Joshi extracted the instructions through Muse's chat interface and shared hi...
FP4 pretraining study reports faster throughput with training-loss tradeoffs
Researchers report in an arXiv preprint that a custom four-bit floating-point training route reached 37.9K tokens/s/GPU, compared with 18.8K for bfloat16 and 27.6K for Transformer Engine in matched tests on the same accelerator. T...

