Google DeepMind unveils SIMA 2, a groundbreaking AI gaming companion powered by Gemini that evolves from following simple instructions to reasoning, conversing, and self-improving through trial-and-error in 3D virtual worlds, demonstrating unprecedented adaptability across unseen games while understanding sketches, emojis, and multiple languages.
Groundbreaking research reveals that AI language models wastefully activate their entire networks for social reasoning tasks while actually using only a tiny specialized subset of parameters, creating massive inefficiency compared to human brains and pointing toward revolutionary energy-saving AI designs.
Scientists develop world's first microwave-powered computer chip that processes over 20 billion operations per second using analog microwaves instead of digital circuits, consuming less than 200 milliwatts while achieving 88% accuracy in pattern recognition tasks for AI and wearable applications.
New research testing 24 advanced AI language models reveals they systematically fail to distinguish between belief and knowledge, with accuracy plummeting from over 90% to as low as 14.4% when processing first-person false beliefs, exposing critical flaws in their reasoning capabilities.
MIT researchers unveil SEAL, a groundbreaking technique that enables AI models to permanently learn new information by creating their own study sheets and updating internal weights, boosting accuracy by 15% on question-answering and over 50% on skill-learning tasks.
Google Research unveils Nested Learning, a revolutionary AI paradigm that solves catastrophic forgetting by treating models as interconnected, multi-level systems where components update at different frequencies, enabling machines to learn new tasks without losing previous knowledge.
Generalist AI releases GEN-θ, breakthrough robot foundation models trained on 270,000 hours of real-world manipulation data, discovering a critical 7B parameter intelligence threshold where robots dramatically improve at complex tasks and need only thousands of training steps to master new skills.
MIT researchers unveil a revolutionary coding framework that transforms AI-generated software by organizing code into modular 'concepts' and explicit 'synchronizations,' dramatically improving safety and reliability while enabling AI assistants to propose new features without dangerous hidden side effects.
Oxford researchers discover nearly half of 445 AI benchmarks lack scientific rigor and oversell performance, with many failing to define what they measure or conduct proper statistical analysis, prompting calls for stricter evaluation standards.
Google dramatically expands NotebookLM AI assistant with massive 1 million token context window, 6x longer conversational memory, 50% better response quality, and customizable personas that can act as specialized roles like PhD advisor or marketing strategist.