Deconstructing the sophistication and universality of technology, covering cutting-edge AI for Science case studies.
This article compiles 22 AI for Science datasets across biomedical, agricultural, materials chemistry, and earth science domains, detailing each dataset's source, size, and application for model training and benchmarking.
Researchers from Shanghai Jiao Tong University, NUS, Meituan, and others built UrbanGround, a real-scale interactive Hong Kong environment using 3D geographic data, to evaluate MLLM agents on local perception, navigation persistence, and dynamic adaptation, revealing models excel at local tasks but fail at long-range navigation and dynamic recovery.
The newly open‑sourced Qwen 3.8‑27B model, built on the Qwen 3.5 architecture, expands its 27‑billion‑parameter foundation with stronger inference, programming, and autonomous‑agent abilities, achieving competitive benchmark scores and offering a step‑by‑step HyperAI tutorial for rapid deployment.
At the Meet AI Compiler #9 salon, Huawei AscendNPU IR architect Hai Lijuan detailed the open‑source compiler’s MLIR‑based architecture, its Tile‑level abstraction, support for Triton and other front‑end languages, and the latest hardware‑aware optimizations for Ascend 950, including Reg‑based SIMD, CV pipeline and MultiBuffer techniques.
The article analyzes why traditional model deployment fails for embodied intelligence, outlines three prevailing solution routes, and presents Lamp—a unified compiler framework that transforms complex robot pipelines from Python prototypes into edge‑ready, high‑performance products, achieving up to 22 Hz inference and offering five integrated capabilities.
The Argonne National Laboratory team introduces ChemGraph, an LLM‑driven agent for computational chemistry, and evaluates it across 13 benchmark tasks, showing that small models excel on simple tasks while larger models and multi‑agent designs dramatically improve performance on complex molecular simulations.
The article compiles 15 NVIDIA Nemotron series datasets—totaling over 10 trillion tokens and 40 million post‑training samples—covering general text pre‑training, supervised fine‑tuning, code generation, math reasoning, and multilingual persona dialogue, all hosted on HyperAI for LLM researchers.
The Virginia Tech team introduced RNAbpFlow, a SE(3)-equivariant flow‑matching model that predicts full‑atom RNA 3D structures using only sequence and base‑pair information, achieving higher TM‑score and lDDT than AlphaFold 3 on CASP16 targets and outperforming existing methods across multiple benchmarks.
This week’s AI paper roundup covers DeepMind’s D4RT framework that accelerates dynamic 4D reconstruction by up to 300×, a Columbia‑NYU proposal of Superhuman Adaptable Intelligence that questions AGI, MIT‑UW findings on chatbot delusional spiraling, security risks of autonomous agents, a new ARA protocol for executable research artifacts, a vision of AI‑driven software engineering, and a memory‑caching approach that expands RNN capacity while reducing complexity.
AgentSociety² is an integrated, human‑in‑the‑loop research environment that lets AI Social Scientists handle repetitive tasks such as literature mining, hypothesis generation, experiment configuration, simulation execution and report drafting, while human researchers retain control over problem definition, hypothesis revision, constraint setting, mechanism interpretation, and the judgment of social significance.
