模型发布/更新
-
Nemotron 3.5 Content Safety: customizable multimodal safety for global enterprise AI
Nemotron 3.5 Content Safety is based on Gemma 3 4B IT and offers a 128K context window, supporting unified multimodal…
-
Nex-N2-Pro released: a 397B MoE reasoning model built on Qwen3.5, matching GPT-5.5 performance
neolab releases Nex-N2-Pro, a 397B-parameter MoE reasoning model built on Qwen3.5-397B-A17B, supporting 262K context…
-
NVIDIA Nemotron 3 Ultra 为长时间运行的智能体带来更快、更高效的推理能力
NVIDIA 发布 Nemotron 3 Ultra 模型,专为长时间运行的 AI 智能体设计。该模型能够在多轮对话中保持上下文、调用工具、调用子智能体,并高效处理复杂工作流。随着多智能体协作导致模型 token 数量快速增长,Nemot…
-
Google releases Magenta RealTime 2 (MRT2) real-time music model
Google AI for Developers announces Magenta RealTime 2 (MRT2), an open-weight real-time music model. The model can be…
产品发布/更新
-
Dreaming: ChatGPT launches a stronger memory system to better remember user preferences
ChatGPT introduces a new memory system called Dreaming that remembers user preferences more effectively and keeps…
-
Gemini for macOS attaches the active window with a double Command press
Get tailored help with what's on your screen using the Gemini app for macOS. 💻 Just press both Command ⌘ keys at once…
-
OpenAI API adds moderation scores
Moderation scores are now available in the Responses API and Completions API. Moderation signals are returned in the…
-
Meet OpenJarvis: a local-first on-device personal AI agent framework with tools, memory and learning
Researchers at Stanford have released OpenJarvis, an open-source framework that runs inference, agents, memory and…
-
NotebookLM source attribution goes live
Today we're launching another highly requested feature: source attribution! 🥳 No more guessing. You can now see the…
-
Codex launches iOS app building plugin
More iOS app loops, now integrated into Codex. The Build iOS Apps plugin lets Codex view and test your iOS app in an…
行业动态
-
台积电难以跟上AI需求:“我们只能支持这么多”
全球最大芯片制造商台积电表示,通过美国本土生产满足客户需求可能需要“非常长的时间”,凸显AI需求带来的产能压力。
-
UN report: AI data center water and power use to double by 2030
A report from the United Nations University Institute for Water, Environment and Health notes that driven by AI…
-
微软AI负责人:Anthropic模型太贵,正自研更便宜的替代模型
微软AI部门负责人表示,Anthropic推出的模型成本过高,公司目前正在内部研发更廉价的替代模型,以降低成本。
-
DeepSeek tops token share leaderboard for four consecutive weeks
DeepSeek has now ranked first on our platform's token share leaderboard for four consecutive weeks:…
-
Nemotron Parakeet ASR 印尼语准确率达 97.7%
当法律和监督依赖于转录内容时,70-80% 是不够的。 http://Rafiqspace.ai 通过微调 Nemotron Parakeet ASR 达到了 97.7% 的印尼语准确率(2.3% WER)—— 优于全球工具,同时每小时成…
-
GitHub Universe 回归:齐聚智能体时代
GitHub Universe 2026 将于 10 月 28–29 日重返旧金山 Fort Mason Center,主题聚焦智能体(agentic)时代。
论文研究
-
Google Research releases passive heart rate monitoring system PHRM
Google Research has developed a passive heart rate monitoring (PHRM) system that uses a smartphone's front camera to…
-
Task-seeded synthetic QA generation for Nemotron pretraining
In a 100B-token continued-training experiment on the Nemotron-3 Nano model, Task-Seeded SDG improved MMLU-Pro by 1.8…
-
NVIDIA PPISP:补偿光度变化提升3D重建
辐射场的质量取决于其背后的图像。 PPISP 可帮助补偿不同拍摄之间的光度变化,使 3D 重建在光照和相机设置不完全一致时更加鲁棒。 🔗 项目:https://nvda.ws/43JeJpk
技巧与观点
-
Boson AI and LMSYS release end-to-end Higgs Audio v3 TTS service on SGLang-Omni
Boson AI and LMSYS jointly launch an end-to-end Higgs Audio v3 TTS service built on the SGLang-Omni inference…
-
Nemotron 3.5 ASR: fine-tune for your language, domain or accent
Nemotron 3.5 ASR is a 600M-parameter multilingual streaming speech recognition model whose single checkpoint covers 40…
-
The End of Coexistence and Collaborative Intelligence
In the One Useful Thing blog, Ethan Mollick writes under the title "The End of Coexistence and Collaborative…
-
OpenRouter tests 11 LLMs to find the fastest decision model: Claude vs. Grok lead
OpenRouter spent $482 in total inference costs to pit 11 large language models against each other in a 30-round…
-
Alex Imas 和 Phil Trammell:AGI 后什么仍然稀缺?
经济学家 Alex Imas 和 Phil Trammell 指出,AGI 时代机器人数量可以快速复制增长,但人类独特技能(以芭蕾舞演员为例)的数量保持不变,揭示了即使技术大幅进步,某些稀缺资源仍不可替代。
-
OpenAI 发布《智能时代的生物防御》行动计划,以 AI 驱动生物韧性
OpenAI 提出一项名为“Biodefense in the Intelligence Age”的行动计划,旨在利用 AI 增强生物防御与韧性。该计划聚焦于构建 AI 驱动的生物安全能力,以应对未来可能出现的生物威胁。