Google Research
PulseAugur coverage of Google Research — every cluster mentioning Google Research across labs, papers, and developer communities, ranked by signal.
4 天有情绪数据
-
Claude AI primarily cites credible health institutions, study finds
A new study analyzed the citations provided by Anthropic's Claude AI when answering consumer health questions. Researchers applied the Authority Signals Framework to over 10,000 citations from a dataset of 3,075 questio…
-
TurboQuant使用PolarQuant将LLM KV缓存压缩4.2倍
一篇技术深度解析文章解释了TurboQuant的内部工作原理，这是一种用于压缩大型语言模型KV缓存的新颖方法。TurboQuant利用一种称为PolarQuant的技术，将KV嵌入转换为极坐标并量化所得角度。该方法旨在通过将KV缓存压缩4.2倍以上，显著减小其内存占用，而KV缓存是长上下文LLM的一个主要瓶颈。
-
Turbovec offers Rust vector index with Python bindings for efficient AI
Turbovec is a new open-source vector index library written in Rust with Python bindings, designed to reduce the memory footprint of vector embeddings for AI applications. It utilizes Google's TurboQuant algorithm, a dat…
-
Google Research's ERA AI aids scientists in writing and optimizing code
Google Research has developed Empirical Research Assistance (ERA), an AI tool designed to help scientists write and optimize code for computational experiments. Published in Nature, ERA utilizes Gemini to search scienti…
-
Google DeepMind's WeatherNext AI predicts historic Category 5 hurricane landfall
Google DeepMind's AI model, WeatherNext, has demonstrated its capability by accurately predicting the rapid intensification and historic landfall of Hurricane Melissa in Jamaica. The model provided a Category 5 intensit…
-
AI代理拥抱HTML，自动化研究，并集成到工作流程中，引发安全担忧
Google Research正在开发新的AI代理，以增强学术工作流程和数据生成。PaperVizAgent和ScholarPeer旨在自动化图形创建和同行评审，而Simula框架则为AI训练生成合成数据集。此外，Viktor等AI工具正被集成到Slack等平台中，通过自动化任务和生成报告来提高生产力。然而，工具和内存的集成到AI代理中也扩大了它们的攻击面，引入了新的攻击向量。
-
Google Research champions open science with global partnerships and shared resources
Google Research is emphasizing its commitment to open science by fostering global partnerships and providing open-source tools and datasets. The company highlights its contributions to fields like genomics and neuroscie…
-
Google Research 利用数据挖掘实现人工智能突破和效率提升
Google Research 正在利用先进的数据挖掘和建模技术，在人工智能领域取得重大突破。他们的工作侧重于改进地理空间推理、语言建模和整体人工智能效率。通过重新定义大型数据集的处理和利用方式，他们实现了数据分析时间的显著缩短，从而在各个研究领域取得了前所未有的进展。
-
Google 的 ERA 工具加速了公共卫生和宇宙学的科学发现
Google Research 的科学家们正在利用一款名为 Empirical Research Assistance (ERA) 的新 AI 工具来加速各个领域的科学发现。ERA 已被用于生成专家级的经验软件，有助于实际应用，例如预测流感、COVID-19 和 RSV 的住院率，以及分析宇宙弦等宇宙学现象。该工具旨在普及计算建模并从数据中揭示更深层次的见解，早期结果显示其在公共卫生预测方面的表现与现有方法相当或更优。
-
ICLR Workshop on Mechanism Design and Decision Intelligence Packed
A workshop at ICLR 2026 focused on mechanism design and decision intelligence, co-led by Alibaba Mama, drew significant attention from both industry and academia. The event explored how AI is reshaping fundamental econo…
-
Google's Perch 2.0 AI model excels at recognizing whale calls using bird training
Google DeepMind has developed Perch 2.0, an AI audio model initially trained on millions of bird recordings, which has shown surprising effectiveness in recognizing whale vocalizations. Researchers utilized transfer lea…
-
Google DeepMind details 2025 AI breakthroughs with Gemini 3 and new models
Google DeepMind and Google Research have detailed significant AI advancements throughout 2025, highlighted by the release of their Gemini 3 and Gemini 3 Flash models. These models demonstrate state-of-the-art performanc…
-
Google's Generative UI creates dynamic, interactive experiences on the fly
Google Research has introduced Generative UI, a new system that allows AI models to create entire interactive user experiences on the fly. This technology generates custom interfaces, tools, and simulations tailored to …
-
Google DeepMind launches WeatherNext 2, an AI model for faster, more accurate weather forecasts
Google DeepMind has unveiled WeatherNext 2, an advanced AI model for weather forecasting that generates predictions 8x faster and with hourly resolution. This new model, built on a Functional Generative Network (FGN) ap…
-
Google DeepMind releases VaultGemma, the most capable differentially private LLM
Google DeepMind has introduced VaultGemma, a 1-billion parameter language model trained from scratch with differential privacy. This release is accompanied by research detailing new scaling laws for differentially priva…
-
A collaborative approach to image generation
OpenAI has integrated its latest image generation model, powering GPT-4o, into its API, allowing developers to incorporate high-quality image creation into their applications. This model demonstrates enhanced capabiliti…
-
OpenAI 的 o1 模型展现出高级推理能力，而谷歌和苹果则在探索新的 LLM 训练方法。
OpenAI 发布了其新模型 OpenAI o1-preview 的早期版本，该模型在推理能力方面相比 GPT-4o 有显著提升。该模型在竞赛编程、高级数学考试和复杂的科学基准测试中表现出色，在某些领域超越了人类专家的表现。这种进步归功于一种大规模强化学习算法，该算法通过思维链教会模型进行生产性思考，并且性能随着训练和测试时间的计算量而扩展。
-
New benchmarks and frameworks emerge for evaluating LLMs in healthcare
Researchers have developed new benchmarks and frameworks to evaluate the performance of large language models (LLMs) in the medical domain, addressing limitations in existing datasets. Google Research introduced AfriMed…
-
研究人员揭示提高 LLM 推理速度和效率的新方法
Google Research 推出了“投机级联”（speculative cascades），一种通过将投机解码与标准级联相结合来提高大型语言模型（LLM）效率的新颖方法。这种混合方法旨在降低计算成本和推理延迟，同时不损害输出质量。通过策略性地使用较小的模型来预测 token，然后用较大的模型进行验证，投机级联与单独使用任一技术相比，提供了更好的成本-质量权衡，Gemma 和 T5 模型已证明了这一点。
-
Google Research evaluates LLM alignment and improves factuality
Google Research has developed a framework to evaluate the alignment of Large Language Models (LLMs) with human behavioral dispositions, using established psychological assessments adapted into situational judgment tests…