VThe Verge@verge原帖11月16日 01:11
该公司搭载 Siri AI 的智能眼镜将于明年 WWDC 发布。The company’s Siri AI-powered eyewear is set to be announced at WWDC next year. https://t.co/spxCp4oX2z
转帖 7 · 引用 2 · 赞 108
TTechCrunch@TechCrunch原帖11月16日 01:09
欢迎回到 TechCrunch Mobility——关注交通出行的未来,以及 AI 在其中前所未有的重要作用。Welcome back to TechCrunch Mobility, your hub for the future of transportation and now, more than ever, the role AI is playing in it.
https://t.co/iM1N3VoX15
转帖 3 · 引用 2 · 赞 76
VThe Verge@verge原帖11月15日 21:01
NVIDIA 前 CEO 的处境变化得非常快。Things moved fast for the now-former CEO of Nvidia. https://t.co/F8g3QKfmzj
转帖 8 · 引用 0 · 赞 61
SSemiAnalysis@SemiAnalysis_原帖11月16日 03:05
逐层拆解麒麟 9050,从顶面到凸点面。LogicFolding 究竟长什么样?欢迎联系了解更多!Layer by layer of the Kirin 9050, from topside to bumpside. What does LogicFolding actually look like? Reach out to learn more!
sales@semianalysis.com | https://t.co/NVaBFeGVWv https://t.co/WAcQhBArEL
转帖 2 · 引用 0 · 赞 114
J机器之心 JIQIZHIXIN@jiqizhixin原帖11月16日 01:24
如果量子纠错自身的测量与反馈会带来额外开销,它还能起作用吗?中国科学技术大学团队在《自然·物理》报告了一项自主式离子阱实验,对量子信息的保护超越了最佳物理比特水平。How can quantum error correction help if its own measurements and feedback introduce additional overhead?
A University of Science and Technology of China team reports an autonomous trapped-ion experiment in Nature Physics, protecting quantum information beyond the best physical https://t.co/Ogvg0XWCRc
转帖 0 · 引用 0 · 赞 6
J机器之心 JIQIZHIXIN@jiqizhixin原帖11月15日 20:00
学习得到的机器人模拟器可以改进策略,并在物理模拟器中得到验证。PrimeBot Research 在 arXiv 预印本中提出动作条件世界模型 PrimeSim。三个要点:1. 训练流程从视频开始……A learned robot simulator can improve a policy that is subsequently evaluated in a physics simulator.
PrimeBot Research introduces PrimeSim, an action-conditioned world model described in an arXiv preprint. Three things to know:
1. The training recipe starts with a video https://t.co/CdgwBha008
转帖 1 · 引用 1 · 赞 12
J机器之心 JIQIZHIXIN@jiqizhixin原帖11月15日 14:36
在 OpenDS-World 的 200 个真实网络研究任务上,12 个前沿模型中最好的也只答对 39%。该基准由港科大(广州)、复旦、华盛顿大学、帝国理工等机构的团队提出。The best of 12 evaluated frontier models answers just 39% of OpenDS-World's 200 real-Web research tasks correctly.
A team spanning HKUST (Guangzhou), Fudan University, the University of Washington, Imperial College London and other institutions introduces the benchmark in an https://t.co/soAGWQsbPR
转帖 1 · 引用 0 · 赞 10
SSemiAnalysis@SemiAnalysis_原帖11月15日 12:00
很好的调查,控制接受长度/接受率(AL/AR)对我们也非常重要。设定合成 AL/AR 是为了防止专门针对基准提示词微调草稿模型,但对狭窄场景的模型用途来说,它仍是一种不错的优化。Great investigation, controlling AL/AR is very important to us too. Synthetic AL/AR was imposed to prevent finetuning draft models specifically on the benchmark prompts, but it remains a good optimization for narrow model use cases.
https://t.co/AXN6XvTXh8
转帖 6 · 引用 3 · 赞 169
J机器之心 JIQIZHIXIN@jiqizhixin原帖11月15日 09:12
如果一台机器人的失败能把整个机队引向另一台机器人的成功经验呢?清华大学、智元机器人和无问芯穹提出集体经验学习(CEL),在 arXiv 技术报告中用最多 96 台机器人的训练机队进行了演示。What if one robot's failure could point the whole fleet toward another robot's successful experience?
Tsinghua University, AgiBot and Infinigence AI introduce Collective Experience Learning (CEL), an arXiv technical report demonstrated with a training fleet of up to 96 robots. https://t.co/tKHrOye4DN
转帖 1 · 引用 1 · 赞 7
SSemiAnalysis@SemiAnalysis_原帖11月15日 06:00
NVIDIA 七大客户中有六家在自研芯片。「NVIDIA 最大的七家客户占据了绝大部分 GPU 部署:谷歌、微软、Meta、亚马逊、SpaceX、Anthropic 和 OpenAI。除 SpaceX 外,它们都在自研芯片。」Six of Nvidia’s seven biggest customers are building their own chips.
“NVIDIA’s seven biggest customers account for the vast majority of GPUs being deployed: Google, Microsoft, Meta, Amazon, SpaceX, Anthropic and OpenAI. All of them, except SpaceX, are making their own chips.” https://t.co/j5Rxb7Lzku
转帖 235 · 引用 61 · 赞 2276