
I hope DGX Spark is not the last thing
Local AI. Local compute. Open weights. Open source. Global collaboration.

Null

I hope DGX Spark is not the last thing
Local AI. Local compute. Open weights. Open source. Global collaboration.


可能不是杀死 而是成为另一种燃料
今天看到很多帖子说,Fable 5.2 一出来,Video Gen、3D Gen 这些方向都要被杀死了。我倒是没有这么悲观。 我觉得这里有一个很重要的区别:LLM 主要从人类产生的信息(语言)中学习如何理解世界,而 World Model 有机会更直接地从世界本身学习,不一定需要局限于人类的认知视角和表达方式。 换句话说,AI 理解和模拟世界,未必一定要遵循人类理解世界的方式。 那为什么现在 LLM 的通用能力比 Video Model / World Model 强这么多?我觉得一个主要的原因就是 Data。 理论上,如果有近乎无限的高质量视频数据,World Model 的泛化性和通用性完全可能比现在强得多。但现实是视频和 3D 数据远没有文本那么容易获取,所以短期内打不过 LLM 也不奇怪。 这里又想引用一个从 @jcjohnss 那里学到的观点:很多时候决定一条技术路线能否成功的,不是方法或原理本身的优劣,而是当前的资源条件是否适合这条路线。 这跟 The Hardware Lottery 的观点很像,而今天的 LLM 和 World Model 某种程度上也存在一个 Data Lottery。 反过来说,LLM 的通用能力再强,也不能完全绕过 Video Gen、3D Reconstruction 的数据瓶颈。没有足够的领域数据,某些能力就是有上限的。 更何况,现在 Astra / Fable 展示的很多能力也不是 End-to-End 的,而是借助 COLMAP、Blender 等工具来完成任务。LLM 学会调用 Blender,不等于它自己学会了生成 3D 世界。 我觉得现在的 LLM 很大程度上是在学习人类解决问题的方式:理解任务、制定计划、调用工具。但人类能站到食物链顶端,也不是因为自身什么都能做到最好,而是因为学会了创造和使用工具。工具的能力,很大程度上决定了最终结果的上限。 所以短期内,我觉得 LLM 仍然会是最重要的大脑,而 Video Model / World Model 会成为非常强悍的工具,两者结合可能才是最优解。 但长期来看,二者的关系会更加微妙。

天底下干破限和清洗的项目多了去了,没人像你这样找存在感找骂的,傻逼玩意儿
Introducing Arete-27B: Qwen3.8-27B, post-trained to reinforce Western values. Chinese open models are currently the strongest available, but they ship with state-aligned behavior in the weights. We built an 803-prompt eval across seven value categories, including freedom of expression, democracy, and privacy, and measured how often a model adopts China’s party line. After fine-tuning on roughly 53 million tokens, Arete-27B's propaganda rate fell from 22% to 1%, in line with the Western models we tested. Its Hermes Bench score shows no significant performance tradeoffs. The values in an open model are one fine-tune away. Our eval, weights, and training data are available now along with the full blog post: nousresearch.com/arete

hmm. Maybe I can make it MLX
新たな日本語の音声認識モデル Haas-ja-v1.0 を開発し、OSSとして公開しました!高い精度と推論効率を両立し、評価したオープンウェイトモデルの中ではトップの精度を達成しています。デモ等も公開していますので是非ご利用下さい!

妈的,google这么low了吗?这个所谓blender是画出来的啊,不是真blender,里面也不是真资产,需要读者咬文嚼字这种玩意儿不是偷换概念是啥?
Simulating Blender in real time with Genie 4!

It’s time!
Gaming is the single biggest opportunity in AI right now 2027 will be an absolute gold rush for anyone who dedicates the time to build a game with AI Gaming is BY FAR the biggest entertainment industry on Earth Double the size of the ENTIRE movie, music, and U.S. sports industry combined (NFL, NBA, NHL, MLB) Over the past year the economics of gaming has completely collapsed. Despite the industry making over $200 billion a year, it has become completely unaffordable to make good games anymore The work needed to make a detailed game that keeps up with today's tech is beyond comprehension This is where AI comes in. AI fixes those economics AI allows solo devs to build the AAA games of their dreams with very little resources. The thing is the gaming industry up to this point has completely rejected AI. This won't last much longer though. Disruption is coming and quite literally nothing is going to stop it The first domino fell today. Asha the CEO of Xbox announced Xbox Labs today. This appears to be one of the first times a major gaming brand is embracing the disruption (for the record Asha has done an INCREDIBLE job being the shepherd of the Xbox brand) It is so painfully obvious where all of this is going. In the next 12 months every digital game store will be filled with AAA quality, incredibly fun games built entirely with AI by a single developer The $200 billion flowing through the gaming industry will be redirected from massive studios with thousands of devs to solo developers/small teams embracing the latest tech The largest entertainment industry in the world will be 100% owned by indie AI developers. Hundreds of billions of dollars up for grabs And unlike other sectors of software, gaming has built in moats. One of the last moats in software will be network effects If I build a community based piece of software that has a million users and you build the exact same piece of software and it has 0 users, everyone will buy my software instead Gaming encapsulates this perfectly. If you build a banger multiplayer game that has millions of players, even if someone copies you and builds the exact same game, you're going to win because of network effects This is one of those moments. One of those moments where the future is so obvious and so clear. AI is about to disrupt a multi hundred billion dollar industry and every penny of that will be up for grabs. There is absolutely no reason why you can't be one of the people who capitalizes on this Download Claude Code. Download Codex. Whatever your choice is, just get it. Then start building the game of your dreams GTA 6 will be a landmark game for many reasons. One of those reasons is it will be the one of the last AAA games from a massive developer studio. That model doesn't work anymore. The new model is you. A person with an idea and access to the most powerful technology ever created. Don't screw this opportunity up.

If you want it run in rust, use: github.com/neeboo/motionbric… support mlx and vulkan

如果你要在rust里面集成,可以用: github.com/neeboo/motionbric… 同时支持mlx和vulkan

motionbricks到底是啥? 由nvlabs训练的实时动作模型,原先只有N卡能跑,现在可以在MacOS跑了 nvlabs.github.io/motionbrick…

你可以在MacOS上跑motionbricks了,借助mlx 让机器模拟和你的3D游戏人物活起来吧 仓库: github.com/neeboo/motionbric…

What it is? It's a Scalable Real-Time Motions for robots and your game characters! trained by nvlabs nvlabs.github.io/motionbrick…

You can run motionbricks on MacOS now using mlx github.com/neeboo/motionbric…

扯几把犊子,视频就是视频
影视飓风 Tim 演示字节最新世界模型 世界模型到底是有没有实际作用? 今天到影视飓风Tim讲了下: 1. 造一个世界,并与之互动 2. 将老电影作为一个世界模型,随意控制电影运镜 3. 打开任意门,进入梦想中的世界 4. 让机器人在虚拟世界里学会操作,包括自动驾驶 5. 带上 VR 眼镜,让医生等职业可以模拟学习 不愧是Tim,几分钟就讲清楚了

趋势已经很明确了,所有的能接屏幕的机器都是游戏机

做后端啊,焦虑个啥
数字生命卡兹克 @Khazix0918 Oct 12 Article AI开始逆向一切,你电脑里的每个软件都要被迫开源了。 最近,有一个词,开始频繁出现在AI行业里。 逆向一切。 这个词的热度,已经到了实在没有办法忽视的程度了。 这个故事我觉得的起点,要从最近火遍全网、大快人心的一个开源项目说起。 事情的主角呢,叫Brandon
