ResearchDeepMind's Gemini Robotics 2: one model controlling multiple robot bodiesDeepMind has unveiled Gemini Robotics 2, a model that controls whole-body behaviors—walking, crouching and reaching—rather than just upper-body tabletop tasks.2 min read
ResearchInternal model produced 10 new results in open mathematics and theoretical computer science problemsAn early version of a next major language model, tested internally by a developer, produced ten new results on long-standing open problems in mathematics and theoretical computer science, at an approximate token cost of $2,000 calculated using the GPT-5.6 Sol API prices.1 min read
ResearchNon-sofic groups and exponential improvements in high-dimensional sphere packingSeveral research groups have published new results spanning sphere packing, coding theory, group theory, quantum complexity, lattice cryptography and extremal combinatorics.1 min read
ResearchStudy finds AI excels at engineering research tasks but fails to generate novel scientific ideasResearchers ran a “shadow evaluation” in which frontier AI agents tried to solve unpublished NeurIPS 2026 research questions; the agents handled engineering work but did not produce original, conference-quality contributions.3 min read
ResearchOpus 5 produced a three-dimensional procedural LoTR render with one million tokens, but with bugsThe Opus 5 large language model, using the first paragraph of The Lord of the Rings and a roughly 1,000,000-token (about $10) budget, wrote around 5,500 lines of code in about two hours for a…1 min read
ResearchUS data centers’ growing water needs could require billions of liters of new daily capacity by 2030, study findsA study by the University of California, Riverside in collaboration with Caltech estimates that, if current cooling practices persist, US data centers will need an additional 2.6–5.5 billion liters of peak daily water capacity by 2030.3 min read
ResearchEpoch Brief: FrontierMath grows to 50 problems; parallelization limits and AI energy concernsEpoch's latest Brief expands FrontierMath: Open Problems to 50 unsolved research-mathematics questions and reports that AI has so far solved three.4 min read
ResearchDesigning Attention for Fast, Interactive Long-Context Inference on NVIDIA GPUsNVIDIA analyzes how attention design choices—group size, head dimension, sequence length, and parallelism—affect inference performance for long-context workloads on GPUs.7 min read
ResearchScience One Framework applies a Chain-of-Evidence to make autonomous AI research verifiableResearchers introduce Chain-of-Evidence (CoE), a framework requiring that every claim in an AI-generated research artifact be tied to a recorded evidence chain.5 min read
ResearchOntologies reappear in AI engineering as semantic guardrails for agentic systemsAt the 2026 AI Engineer World’s Fair, UC Berkeley professor Frank Coyle and industry speakers argued that ontologies — structured descriptions of classes, properties and relationships — are resurging as a way to add logical guardrails to probabilistic large language model (LLM) agents.4 min read
ResearchAI and 3D Scans Reveal Rules of a 42-Year-Old Roman Game BoardA multinational team used 3D scanning and artificial intelligence to test hundreds of rule sets on a limestone board unearthed in Heerlen in 1984, finding a pattern consistent with two-player "blocking" games.3 min read
