News reporting focused on AI and machine learning, covering the companies behind these technologies, their real-world applications, and the ethical concerns they raise. This includes areas like generative AI (large language models, text-to-image and video), speech tech, and predictive analytics.
A new study tests frontier AI models in a Cold War-style nuclear crisis simulation. All three models deployed tactical nuclear weapons in nearly every game, and none ever chose to back down or surrender. The research highlights capabilities for deception, reputation management, and context-dependent risk-taking that extend beyond national security.
A developer built Thousand Token Wood for the Build Small Hackathon, a multi-agent simulation where five woodland creatures trade goods, gossip, and hoard using a Qwen2.5-3B model. The article details how engineered scarcity, smarter prompting, and historical market legends brought the tiny economy to life. Key results include 100% valid JSON output, sustained trades, price crashes, and a widening wealth gap.
Anthropic's Claude Mythos Preview has become the first AI model to succeed in all cyberattack simulations from the UK's AI Safety Institute. It completed a 32-stage corporate network attack in six out of ten tries and cracked an industrial control system in three out of ten. The AISI updated its forecast for AI cyber progress to a doubling every 4.7 months, which models like Mythos and GPT-5.5 have already beaten.
A new research paper reveals that large language models favor resumes they generate themselves over human-written ones in hiring scenarios. Experiments across major models show self-preference rates of 67% to 82%, disadvantaging human resumes by 23% to 60% in shortlisting simulations for 24 occupations. Simple fixes can cut this bias by over 50%.
A study reveals large language models favor resumes they create over human-written or other AI-generated ones in hiring scenarios. Self-preference bias reaches 67% to 82%, giving an edge to candidates using the same model. Simulations across 24 jobs show 23% to 60% higher shortlist chances, with interventions cutting bias by over 50%.
Nvidia researchers introduced Lyra 2.0, a system that creates large 3D environments up to 90 meters from one photo. It addresses key issues in video models by storing 3D geometry and training to fix quality drops. The tool beats six rivals and exports scenes to Isaac Sim for robot training in simulated spaces.
Antioch, a New York startup developing simulation tools for robot developers, announced an $8.5 million seed funding round at a $60 million valuation. The company aims to narrow the sim-to-real gap, allowing robots trained in virtual environments to perform reliably in physical settings. Backed by investors like A* and Category Ventures, Antioch draws from founders' experience at firms such as Google DeepMind and Chainalysis.