ResearchAI agents, not just big datasets, may drive the next wave of scientific breakthroughsAlphaFold showed that AI trained on massive, high-quality datasets can solve specific, long-standing scientific problems.5 min read
ResearchStartup uses AI agent swarms to hunt for cooler semiconductor materialsDiscovered Materials, a startup founded by Advaith Sridhar and Akash Ramdas, raised $9 million in seed funding to use swarms of AI agents and physics simulations to find semiconductor materials that reduce heat in chips.3 min read
ResearchAgentRadio's asynchronous messaging nearly doubles multi-agent performance on enterprise code tasksResearchers from Coral AI Labs and partner universities introduced AgentRadio, an asynchronous message-passing layer that gives concurrently running coding agents passive awareness of each other.6 min read
ResearchStanford builds a 37,000‑agent 'Virtual Biotech' and validates a drug design later confirmed by MerckAt VB Transform 2026, James Zou of Stanford described a shift from single, powerful AI agents to systems of tens of thousands of specialized agents coordinated by an AI orchestration layer.5 min read
ResearchAI-driven single-cell analysis helps trace origins of metastatic cellsResearchers at the Szeged Biological Research Centre, in collaboration with Swedish and Swiss partners, have combined AI-based image analysis with single-cell proteomics and transcriptomics to map molecular differences between tumor cell groups.3 min read
ResearchStanford team uses genome language models to design functional bacteriophages, raising governance concernsResearchers at Stanford applied genome language models—AI tools analogous to chatbot LLMs—to design bacteriophages and produced the first functional viruses created by AI.2 min read
ResearchTime and Token Budgets Explain Divergent Qwen 3.8‑Max Benchmark ResultsTwo recent benchmark sets reached opposite conclusions about Alibaba’s Qwen 3.8‑Max because they used very different time and token budgets.6 min read
ResearchNo universal model: different AIs are strong for different tasksA recent comparison found there is no single AI model that outperforms on every task: Grok 4.5 is strong on routine tasks, GPT-5.6 Sol is effective at planning and understanding codebases, Opus 5 excels at execution-intensive work, and Fable 5 performs well in debugging and visual implementation.1 min read
ResearchAI's sudden dominance in pure mathematics: a month of breakthroughs and their consequencesIn mid‑2026, large AI models—from OpenAI’s GPT‑5.6 Sol and the internal Astra family to systems developed at Anthropic—produced a rapid sequence of solutions and counterexamples to long‑standing mathematical problems.6 min read
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
