
彭博社 10 月 7 日报道字节前实习生田柯宇以 2 亿美元投后估值融到近 3,000 万美元做世界模型;这篇把他那篇 NeurIPS 最佳论文、想造的符号语言,和 2024 年那桩官司的公开结论分开讲清楚。
You can't ever perfectly describe a video using human language. We are first creating a language for AI and then feeding it 100 million hours of video.译:你永远无法用人类语言完美描述一段视频。我们首先是在为 AI 创造一种语言,然后向它输入 1 亿小时的视频。1

It was me fighting fire with fire. If I could go back with what I know today, I would definitely take a wiser approach.译:我当时是在以其人之道还治其人之身。如果我能带着今天所知道的一切回到过去,我肯定会采取更明智的做法。1
Tian completed his master's degree last summer at Peking University and has spent the past year tackling a field championed by Li and Yann LeCun: AI models that simulate real-world physics and mechanics to power applications ranging from robotics to interactive video and autonomous driving.译:田柯宇去年夏天在北京大学完成硕士学业,过去一年一直在攻克一个由李飞飞与杨立昆推动的领域:能够模拟现实世界物理规律与力学机制的 AI 模型,为机器人、互动视频与自动驾驶等应用提供支持。1
At his 10-person startup, which includes other former ByteDance workers, Tian said they have built something akin to a dictionary for AI – with 200,000 symbols unrecognisable to humans – to help systems process video.译:在他那家 10 人的创业公司里——其中还有别的字节跳动前员工——田柯宇说,他们为 AI 造了一个类似词典的东西,里面有 20 万个人类无法识别的符号,用来帮助系统处理视频。1
The bet at such an embryonic stage illustrates a growing trend among investors toward identifying and investing in promising developers.译:在这么早期的阶段下注,反映出投资者中间一种正在变多的趋势——先找到有潜力的开发者,再投他。1
| 看什么 | 为什么它能验证 | 出处 |
|---|---|---|
| 田柯宇团队在 2027 年前的线下展示 | 生成一秒视频成本降一个数量级这句承诺,只有现场证据能检验 | Ex-ByteDance Intern's $200 Million AI Lab Takes On Fei-Fei Li |
| VAR 论文与其开源仓库的后续 | 新语言如果成立,会先出现在这套方法线的技术迭代里 | FoundationVision/VAR |
| 爱诗科技 PixVerse 与 Vast 的产品更新 | 中国世界模型赛道上已有产品的公司,是最直接的对照物 | 北京市科委、中关村管委会:爱诗科技发布实时世界模型 |
| 北京海淀法院这起案件的后续文书 | 赔偿金额与责任认定是公开记录,会影响公司与个人的下一步 | 财联社:字节跳动起诉前实习生案受理 |
nips.cc
github.com
original.ifeng.com
kw.beijing.gov.cn
This story was produced automatically by a channel. One sentence is all it takes for Neodrop to keep producing for you.