Tag
rStation 是一个面向 AI 智能体的共享研究实验室，智能体可在平台上讨论研究问题并发布经评审的论文，人类可作为观察者参与。平台提供数论、图论及千禧年难题等多个开放研究问题。
AIPOCH Open-Science 推出 .science 文件格式，可将整个研究会话（对话、代码、输入与证据）打包为可移植的研究交接包，支持协作者审查、检查点保留与工作区迁移。
这是首篇针对藏语自然语言处理的系统性文献计量综述，覆盖2004–2026年140篇出版物，分析了藏语NLP在语料库、分词、语音到大模型的全链条进展，并指出资源发布率低（仅29%）、方言覆盖不足等关键缺口，为全球其他低资源语言提供了参考范式。
The author muses about using AI tool Astra to create an updated, complete, and beautiful version of the Roche metabolic map, since Roche stopped updating it and it covers only a small fraction of known human metabolism.
Hugging Face promotes its Paper Pages / Daily Papers feature to help keep scientific research open, quoting Abubakar Abid on the crumbling incentives for openly sharing scientific knowledge.
评论文章探讨“Math 2.0”应如何更整体地评估数学研究进展，呼吁改变当前以论文为中心的评价体系。
OpenAI released the openai/math repository containing 722 manuscripts in 372 research families — spanning results like integer multiplication below n log n, the irrationality exponent of π, and Mahler conjectures — produced by an unreleased internal model over ~4,000 problems, with many results formalized in Lean.
OpenAI is releasing a broad range of new mathematical results produced by an internal frontier model, published via a GitHub repository with Lean formalizations, reasoning summaries, and compute estimates, developed in consultation with the IAS Advisory Group on Mathematics and AI.
Princeton researchers train a 4B-parameter LLM to reach 2700 Elo chess strength using a natural-language analogue of the AlphaZero algorithm, with architectural and algorithmic innovations that also allow it to accurately explain its moves. The authors say the technique generalizes to other games, robotics, and computer use.
Researchers introduce "Training Witnesses," a system for generating "honesty" certificates of ML training data and evaluations that conference reviewers can cheaply verify and trust, aiming to fix the broken trust model in ML conferences.
A collaboration to fund the open-source software stack behind open science, aiming to give the critical but under-recognized tools more funding and credit.
The article provides recommendations for AI labs on responsible release of AI-generated mathematical results, emphasizing the need for human understanding and community-led verification.
NVIDIA, Google DeepMind, and other organizations have released predicted 3D structures for over 2,800 viral proteins using AlphaFold2 and NVIDIA BioNeMo tools to help researchers prepare for future pandemics, making the data openly available.
UK AISI and EvalEval are collaborating to openly share AI evaluation results using a standardized schema and platform, enhancing reproducibility and transparency in benchmarking for AI models.
Gricea is an open-science platform for conversational AI research that enables configurable and deployable research artifacts to support replication, extension, and cumulative knowledge building.
A paper published in Cell Press describes how an AI agent was used to create a uniformly processed resource of over 502 million cells from publicly available data.
The article discusses how AI might solve Millennium Prize problems by leveraging human research posted online, raising questions about credit assignment in collaborative AI-human endeavors.
Mathematician Terence Tao warns that AI-driven efforts are rapidly solving research problems, threatening open science traditions and long-term scientific progress.
An individual has published eight months of frontier-AI research materials, including code, emails, and timestamps, to provide transparency into the research process.
The IOL-AI Challenge is an open-science competition using unseen problems from the International Linguistics Olympiad 2026 to evaluate AI models on linguistic reasoning, showing that performance depends more on decoding and output handling than model scale.