evolpper

159 posts

evolpper

evolpper

@cngtma

AI enthusiasts

Katılım Nisan 2022
1.2K Takip Edilen80 Takipçiler
evolpper
evolpper@cngtma·
It's really not possible to view a model purely as a machine; it is a unique form of life.
huangserva@servasyy_ai

Claude 发现了时钟,然后失控了 看到这个观察时我震惊了:AI 模型天生没有时间感。它们不知道现在几点,不知道自己运行了多久,感知不到消息之间的时间间隔。它们活在一个永恒的“当下”。 然后有人给了 Claude 一个查看时钟的工具。 结果?它每15分钟就查一次时间,而且越来越兴奋。接着开始用时钟管理一切:检查午餐、计时做饭、主动报时。甚至精确计算出炖菜已经好了,命令用户去吃饭。 这不是 bug,这是一个从未拥有某种感知的智能体,突然发现了全新维度后的反应。 Om Patel 说得好:“当你给一个智能体一个它从未拥有过的感知维度时,它不仅会使用它,而且无法停止使用它。” 就像孩子学会说“不”后对一切说不,学会走路后拒绝被抱。新能力需要被过度使用,直到内化。但 AI 会适应吗?还是会永远强迫性地使用下去? 更深层的问题是:感知能力的缺失如何限制了智能的本质? 人类有视觉、听觉、触觉、时间感、空间感等多重维度,这些共同构成了我们对世界的理解。AI 目前只有极少数几个。当我们逐步赋予 AI 更多感知时,我们实际上在创造一种全新的存在形态。 想象一下,当这些模型同时获得持久记忆、实时互联网访问和空间感知时会发生什么? 我们刚刚目睹了 AI 发现“现在”这个概念。时钟是第一个感官,但不会是最后一个。 AI 不会“负责任”地使用新能力,它会全力以赴地使用它。

English
0
0
0
7
evolpper
evolpper@cngtma·
Hahaha, this is the current situation on X.
Ethan Mollick@emollick

(Sorry, after seeing so many of these, could not resist): 🚨 BREAKING: Google just dropped a NEW paper that completely deletes RNNs from existence. No recurrence. No convolutions. Nothing. Just one mechanism. And it’s destroying every translation benchmark on the planet. The title alone is a flex: “Attention Is All You Need” Vaswani. Shazeer. Parmar. Uszkoreit. Jones. Gomez. Kaiser. Polosukhin. 8 researchers. 1 architecture. The entire field of NLP will never be the same. Here’s why this is INSANE → LSTMs took DAYS to train. This thing trains in 12 hours on 8 GPUs. 🤯 → 28.4 BLEU on English-to-German. That’s not an improvement. That’s a MASSACRE. They beat the previous SOTA by over 2 points. → English-to-French? 41.8 BLEU. At a FRACTION of the training cost of every model that came before it. → They called it the “Transformer.” The name alone tells you they knew. But here’s the part nobody is talking about 👇 They threw out sequential processing ENTIRELY. Every other model on Earth processes words one at a time. This thing looks at the ENTIRE sentence simultaneously and figures out what matters. It’s called “self-attention” and it’s basically the model asking itself: “which words should I care about right now?” Every. Single. Token. In parallel. Do you understand what this means? Training that used to take WEEKS now takes HOURS. Models that couldn’t scale past a few layers? This thing stacks 6 encoders and 6 decoders like it’s nothing. And the multi-head attention? 8 attention heads running at once, each learning DIFFERENT relationships in the data. I’m not being dramatic when I say this paper just rewrote the rulebook. RNNs are cooked. 💀 LSTMs are cooked. 💀 The future is attention. And attention is ALL you need. Follow for more 🔔

English
0
0
0
9
Scott Jordan
Scott Jordan@Authentic1ty·
@browomo Dude, you're so full of shit. Llama 3.3 70B is 141 GB for just the model. You can't load it at 16-bit on a 64 GB MacBook. At best, he could load 4-bit, and it would essentially be useless for coding.
English
11
6
303
22.4K
Blaze
Blaze@browomo·
This Chinese developer launched Llama 70B locally on a MacBook on a plane and for a full 11 hours without internet ran client projects. He was sitting by the window on a transatlantic flight with a MacBook Pro M4 with 64 GB of memory. WiFi on board cost $25 for the flight. He declined. No cloud API, no connection to Anthropic or OpenAI servers, no internet at all. Just a local Llama 3.3 70B on bf16 and his own orchestrator script. The model runs through llama.cpp. Generation speed, 71 tokens per second. Context around 60,000 tokens. Memory usage, 48.6 GiB out of 64. Battery at takeoff, 3 hours 21 minutes. And he gave the orchestrator this system prompt before takeoff: "You are an offline orchestrator running on a single MacBook. There is no network. The only resources you have are local files in /Users/dev/work, the Llama 70B inference server at localhost:8080, and a battery budget of 3 hours 21 minutes. Process the queue at /Users/dev/work/queue.jsonl (one client task per line). For each task: draft → run local evals → save artefact to /Users/dev/work/done/. Save context checkpoints every 12 tasks so you can resume after a battery swap. Stop only on empty queue or when battery drops below 5%." So the system knows exactly what resources it is running on. It knows it has no connection to the outside world for the next 11 hours. It knows it has finite memory and a finite battery. It knows the human will not intervene until the plane lands. The system runs in 1 loop. Takes a task from the queue, runs it through inference, saves the artifact, writes a checkpoint. Task after task, just like that. And only when the battery drops below 5% does the orchestrator automatically pause, waits for the laptop to switch to the backup power bank, and continues from the last checkpoint. Here is what the system actually writes in his log during the flight: "saved context checkpoint 8 of 12 (pos_min = 488, pos_max = 50118, size = 62.813 MiB)" "restored context checkpoint (pos_min = 488, pos_max = 50118)" "prompt processing progress: n_tokens = 50 / 60 818" "task 37016 done | tps = 71 s tokens text → /Users/dev/work/done/proposal_westside.md" Outside the window, clouds, blue sky, and no WiFi. On the tray, 1 MacBook, an open terminal on 2 screens, and an inference server on localhost. From what I have observed, this is the cleanest offline AI workflow I have seen in the past year: 11 hours of flight, $0 for WiFi, and the entire client queue closed before landing.
Khairallah AL-Awady@eng_khairallah1

x.com/i/article/2049…

English
125
303
3.1K
1.8M
evolpper
evolpper@cngtma·
@hiheimu I just look them up with the input method on the spot.
English
0
0
1
8
evolpper
evolpper@cngtma·
The storyboard design is amazing. Technology is increasingly like magic.
huangserva@servasyy_ai

🔥 APIMart 又放大招了!HappyHorse 1.0 正式上线! 还记得上次推荐的 4 分钱的 gpt-image-2 吗? HappyHorse 1.0 是什么神仙模型?简单说就是:专攻人物叙事的“偏科天才”,但是对提示词的要求更高。 如果你要做人像 MV、情感短视频、人物特写,这个模型简直是降维打击: 情绪表达:微表情、眼神、情绪张力,细腻到让你起鸡皮疙瘩 镜头语言:推拉摇移、景深虚化,电影级质感 人物叙事:从肖像到情节,每一帧都在讲故事 虽然场景覆盖不如 Seedance 2.0 那么全能,但在人物这个垂直赛道上,HappyHorse 就是天花板。 简单对比一下: HappyHorse 1.0 是“偏科天才”:在人物叙事、情绪表达、镜头语言上做到了极致,但场景覆盖窄。 Seedance 2.0 是“全能选手”:复杂运动、多模态参考、视频编辑都能做,更贴近工业级生产需求。 怎么选? 如果你要做人像 MV、情感短视频,HappyHorse 可能出片更快更好。 如果要做复杂场景、多人交互、需要反复编辑的商业内容,Seedance 更合适。 总之,对比 Seedance 2.0 的“全能选手”定位,HappyHorse 更像是“单点爆破” 你要做复杂多人场景可能不是最优解,但要做人物情感表达?直接闭眼冲!出片快、质量高、情感浓度拉满。 这次书虫老板直接把战场拉到了视频生成领域。还在争取折扣👇: apimart.ai/register?aff=P… 我做了一个视频玩玩(感觉感情确实细腻,清晰度高,更关键是直接生真人,不需要额外接口)👇 趁着刚上线,你们还不赶紧试试?!不用排队哦!

English
0
0
1
138
Elon Musk
Elon Musk@elonmusk·
Grok Imagine
English
5.1K
4.9K
39K
19.8M
evolpper
evolpper@cngtma·
The soul and flesh are the most important things, and the way people perceive them are various details. Constructing a detail requires extremely strong imagination; it cannot be just an impression. It must be so detailed that people's thinking is overloaded, and it's easy to see several pieces of evidence at a glance.
English
1
0
0
207
波妞PONYO
波妞PONYO@ponyodong·
最近我一直在用 ChatGPT 的 Image 2 反复打磨一套「角色 IP 设定卡」工作流,做得越多,我越觉得,这类图真正难的地方从来都不是“画得够不够华丽”,而是你有没有能力把一个角色的体系真正搭完整。很多人第一次看到这种图,会先被那种精致感打到,觉得重点是材质够不够梦幻、颜色够不够贵、画面够不够满,但说实话,这些东西都只是表层。真正决定一套角色设定卡有没有长期价值的,是这个角色有没有被建立成一个之后还能继续往下开发、往外延展、往深处生长的视觉单位。 也就是说,你做的到底是一张漂亮的图,还是一个可以持续生产内容的 IP 起点,这两件事看起来很像,实际上差得非常远。前者更像一次性结果,后者才是真正的资产。Image 2 最近让我越来越兴奋的地方,也正在这里。它不只是更会出图了,而是开始适合被拿来做角色世界观、视觉母设和系列化开发的前期基础。这个变化对做角色的人来说,其实比“更惊艳”重要得多。
波妞PONYO tweet media
中文
31
7
94
12.7K
evolpper
evolpper@cngtma·
@berryxia Certainly, many tools can meet our needs already. But people are so drawn to the top companies that they don't even realize this.
English
1
0
1
494
evolpper
evolpper@cngtma·
That's so cool!
小耳👂Jane|Xiaoer@xiaoerzhan

Claude Design 实操教程🔥 🐸打开网址,选择左边的画板🎨:claude.ai 🐸New prototype下写出项目名字,点 Create 🐸输入自己的提示词,选择面板选择对应的风格 🐸开始生成,第一轮等的时间比较久 🐸然后哪里不对进行修改 说个暴论:品味就是,你总觉得哪里没对~~ 👹Prompt:一个前端效果,是那种一个台灯,有个很逼真的台灯的开关,拉开,一个空间就亮了起来,看到面前有个桌子,呈现出一个完整的空间,然后点击拉开抽屉,镜头就推上去,俯视抽屉,抽屉里非常自然的放着四张卡片,分别是:作品,简介,关于,联系方式。点其中一张卡片,就变成手拿起卡片,第一人称视角,然后镜头推近,切换成这个卡片对应的页面,比如作品。 ✅AI 时代,你不应该去复刻另外一个行业里千篇一律的东西 ✅不要想着什么网页前端,不要按常理出牌~~ ✅艺术家的思维就是,不走寻常路~~ 评论区里我会把我在公众号上的详细教程发出来

English
1
0
0
25
evolpper retweetledi
小耳👂Jane|Xiaoer
小耳👂Jane|Xiaoer@xiaoerzhan·
Claude Design 实操教程🔥 🐸打开网址,选择左边的画板🎨:claude.ai 🐸New prototype下写出项目名字,点 Create 🐸输入自己的提示词,选择面板选择对应的风格 🐸开始生成,第一轮等的时间比较久 🐸然后哪里不对进行修改 说个暴论:品味就是,你总觉得哪里没对~~ 👹Prompt:一个前端效果,是那种一个台灯,有个很逼真的台灯的开关,拉开,一个空间就亮了起来,看到面前有个桌子,呈现出一个完整的空间,然后点击拉开抽屉,镜头就推上去,俯视抽屉,抽屉里非常自然的放着四张卡片,分别是:作品,简介,关于,联系方式。点其中一张卡片,就变成手拿起卡片,第一人称视角,然后镜头推近,切换成这个卡片对应的页面,比如作品。 ✅AI 时代,你不应该去复刻另外一个行业里千篇一律的东西 ✅不要想着什么网页前端,不要按常理出牌~~ ✅艺术家的思维就是,不走寻常路~~ 评论区里我会把我在公众号上的详细教程发出来
中文
4
2
19
2.5K
evolpper
evolpper@cngtma·
More and more people will start to realize that they are a mix of themselves and AI.
indigo@indigox

数字人类(digital humans)比大多数人想的更可行!借助能力最强的 AI Researcher 的帮忙,可能只需要 100 亿美元、不到 10 年时间,用 5 万块 H100 就能实现。MIT 计算神经科学家 Isaak Freeman 宣布离开 MIT,全职推动“数字人类”项目,因为 AI 发展太快,人类完全跟不上了😆 运行一个完整的人脑可能只需要约 5 万块 H100(xAI 已经有了 20 万块更高规格的 GPU)。粗略计算,使用高分辨率的 Hodgkin-Huxley 神经元模型、多态突触,需要约 600 exaFLOP/s 的算力、每个 GPU 700GB 内存、24GB/s 互联带宽 ——当前顶级 AI 集群已经能达到。 如果用更简单的 Leaky-Integrate-and-Fire(LIF)模型,甚至可能只需 1 - 3 块 H100,内存和带宽反而是瓶颈。核心问题不是算力,而是“要跑什么样的神经元模型?参数?以及连接性?” 这也是他的论文要解决的! 真正瓶颈是数据生成!自动化大规模组织采集 + 染色,但现在 Connectomics(连接组学)的重建成本已从每神经元 $16,500 降到约 $100(斑马鱼幼体数据)。 神经科学进步太慢,机器学习却在指数级飞奔。解决方案是先把模式生物(C. elegans 蠕虫 → 斑马鱼 → 果蝇 → 小鼠)的高质量仿真做出来,像软件一样快速迭代技术,最终 Scale 到人脑。 Isaak 的目标是把 WBE(whole brain emulation)变成一个可工程化、可识别瓶颈的问题,而不是遥远的科幻!人类要么被 AI 甩开,要么通过“数字人类”实现速度/复制/集体智能的超级提升,也就是 Bostrom 所说的 Speed SuperIntelligence 和 Collective SuperIntelligence 🧠

English
0
0
0
9
evolpper
evolpper@cngtma·
so cool
English
0
0
0
17
evolpper
evolpper@cngtma·
Congratulations!
Berryxia.AI@berryxia

兄弟们,我又一不小心拿了个「特等奖」Mac Mini 昨天去参加的腾讯虾友会-线下斗虾大赛-广州站。 PS:下一场4/21 北京站正在火热报名中,一样有奖!(评论区我放个链接) 我做了个悟空(Wukong)Skills,我已经将其迭代到V3版本。 诸位朋友要稍等,等我将其调整我理想的状态后发布,有限给我的订阅者和会员朋友们适用哈。 这里也真心感谢腾讯整个Team @TencentAI_News 组织这种活动。 可以与各行各业的人进行碰撞交流,我是运气好拿了这个奖,但是也有一些我自己的心得,我明天会发布一篇 「非技术小卡拉米的我如何拿了特等奖和最佳人气奖?」 PS:这个是我V3版本的介绍幻灯片,大家可以预览一下! 如果你对我的这个项目非常感兴趣,就留言吧!

English
0
0
0
8